JP7761185B2 - Analysis and Estimation Systems - Google Patents

Description

The present disclosure relates to analysis systems and estimation systems.

Patent Document 1 discloses a maintenance record input support device that is an analysis system. The analysis system supports maintenance personnel performing maintenance work on equipment when entering maintenance records. The analysis system extracts keywords from free-form fault condition information entered by the maintenance personnel. Of the options selected as maintenance records, the analysis system identifies options that correspond to phenomena that are likely to occur based on the extracted keywords. The analysis system can support maintenance personnel by suggesting the identified options to them.

Japanese Patent Publication No. 2023-019574

However, the analysis system described in Patent Document 1 identifies options based only on the probability of association with extracted keywords. Free text often contains words that are extracted as keywords but have little relevance to the option to be selected. As a result, there is a risk that the options suggested will be unlikely to be selected.

The present disclosure has been made to solve the above-mentioned problems. The purpose of the present disclosure is to provide an analysis system and an estimation system that can provide options that are likely to be selected as options to be added to the device's failure response record, or that can estimate those likely options.

The analysis system disclosed herein is a system that analyzes failure response records containing free text entered by people who responded to equipment failures, and includes: an extraction unit that extracts specific target terms and words indicating the details of the actions taken during the failure response from the free text included in the failure response records; an association calculation unit that calculates the association between each of multiple candidate codes that are candidates for judgment codes to be assigned to the failure response records and the target terms extracted by the extraction unit; an importance calculation unit that calculates the importance of the actions extracted by the extraction unit and associates them with one of the multiple candidate codes; and a list unit that creates a selection list in which the multiple candidate codes are ranked higher, based on the association calculated by the association calculation unit and the importance calculated by the importance calculation unit, so that the higher the likelihood of them being selected as a judgment code, the higher the ranking.

The estimation system disclosed herein is a system that analyzes failure response records containing free text entered by people who responded to equipment failures, and includes: an extraction unit that extracts specific target terms and words indicating the details of the actions taken during the failure response from the free text included in the failure response records; an association calculation unit that calculates the association between each of multiple candidate codes that are candidates for judgment codes to be assigned to the failure response records and the target terms extracted by the extraction unit; an importance calculation unit that calculates the importance of the actions extracted by the extraction unit and associates them with one of the multiple candidate codes; a list unit that creates a selection list in which the multiple candidate codes are ranked higher based on the association calculated by the association calculation unit and the importance calculated by the importance calculation unit, with the candidate codes being ranked higher the more likely they are to be selected as a judgment code; and an estimation unit that estimates the highest-ranked candidate code in the selection list created by the list unit to be the judgment code.

According to the present disclosure, a selection list is created in which multiple candidate codes are ranked higher the more likely they are to be selected as a judgment code, based on the relevance of target terms extracted from the free text and the importance of actions extracted from the free text. This makes it possible to provide options that are likely to be selected as options. Furthermore, the estimation system estimates that the candidate code with the highest ranking in the selection list is the judgment code. In this way, it is possible to estimate options with high likelihood.

1 is a diagram showing an overview of an elevator device in which an analysis system according to a first embodiment is used. 1 is a cross-sectional view of a refrigerator in which the analysis system of the first embodiment is used. FIG. 1 is a hardware configuration diagram of an analysis system according to a first embodiment. FIG. 1 is a functional block diagram of an analysis system according to a first embodiment. FIG. 2 is a diagram showing an example of a search user interface displayed by the analysis system in the first embodiment. FIG. 10 is a diagram showing an example of a setting request screen displayed by the analysis system in the first embodiment. 1 is a flowchart for explaining an outline of the operation of the analysis system in the first embodiment. FIG. 10 is a hardware configuration diagram of an analysis system according to a second embodiment. 10 is a flowchart for explaining an outline of the operation of the analysis system in the second embodiment.

The embodiments for implementing the present disclosure will be described with reference to the accompanying drawings. Note that in each drawing, identical or corresponding parts are designated by the same reference numerals. Duplicate descriptions of these parts will be appropriately simplified or omitted.

Embodiment 1.
Fig. 1 is a diagram showing an overview of an elevator device in which the analysis system according to the first embodiment is used. Fig. 2 is a cross-sectional view of a refrigerator in which the analysis system according to the first embodiment is used.

Figure 1 shows an elevator system 100, which is a first example of a system in which the analysis system 1 can be used. In the elevator system 100, a hoistway 101 passes through each floor of a building 102. A plurality of landings 103 are provided on each floor of the building 102. A plurality of landing doors 104 are provided at the entrances and exits of the plurality of landings 103. The landing doors 104 are sliding doors that move along door rails 105. A car 106 is provided inside the hoistway 101. The car 106 can move up and down inside the hoistway 101. A control panel 107 is provided in a machine room above the hoistway 101. The control panel 107 controls the overall operation of the elevator system 100. When a malfunction occurs in equipment of the elevator system 100, the control panel 107 can display an error code corresponding to the malfunction. The on-site terminal 108 is a terminal capable of inputting information and visually displaying information. For example, the on-site terminal 108 is a notebook computer. The on-site terminal 108 is capable of connecting to the network N via wireless communication or the like.

Maintenance worker H1 belongs to a company that performs maintenance work on elevator equipment 100. When maintenance worker H1 receives a report that elevator equipment 100 has broken down, he goes to building 102 and performs response work, such as replacing the faulty part of elevator equipment 100. Maintenance worker H1 inputs information indicating the results of the response work into the on-site terminal 108. At this time, maintenance worker H1 inputs information such as the error code displayed on the control panel 107 and the type of faulty equipment as the results of the response work. Furthermore, as part of the results of the response work, maintenance worker H1 inputs free text in natural language, including the situation upon arrival at building 102, the status of the equipment as heard from the owner of building 102, the cause of the failure, and the details of the corrective work performed to address the failure. Free text is a string of characters entered in natural language and is not pre-set as a choice. Based on the input information, the on-site terminal 108 creates information for a failure response record that includes the input free text.

The analysis system 1 includes a search terminal 2 and a server device 3. The analysis system 1 may further include an on-site terminal 108. In the example of FIG. 1, the search terminal 2 and server device 3 are provided at a company that performs maintenance work on the elevator device 100. Specifically, for example, the search terminal 2 and server device 3 are provided at the company's information center S. The server device 3 can communicate with the control panel 107 via the network N and a monitoring device (not shown) installed near the control panel 107. The server device 3 can receive fault response records from the on-site terminal 108 via the network. The search terminal 2 can display the fault response records received by the server device 3 and accept input of information related to the fault response records.

Manager H2 works at Information Center S. Manager H2 operates Search Terminal 2 to view the failure response records. Manager H2 inputs a judgment code, which is information for information management, into the failure response record. In this case, Server Device 3 creates information on the failure response history that includes the judgment code assigned to the failure response record and the failure response record itself, and registers this information in the failure response history database.

A judgment code is a tallying code used in a database related to failure responses. One or more types of judgment code are set for each failure response history. For example, types of judgment code include tallying device name, cause code, etc. A tallying device name is the name of the device that was responded to in a certain response work. For tallying device names, terms that indicate a fairly broad concept, such as "door," "interior," or "switch," are used. For example, if a part of a door is replaced in a response work, "door" is entered as the tallying device name. A cause code is a code that indicates the cause of a failure that was addressed in a certain response work, and may be a term that indicates a fairly broad concept, such as "broken wire" or "deterioration over time," or may be a string of numbers or other characters that indicate the cause. For each of multiple judgment codes, multiple candidate codes are set as input options.

The server device 3 analyzes the free-text content written in natural language contained in the failure response record and creates a selection list in which multiple candidate codes for a certain judgment code are rearranged. The selection list is a list in which multiple candidate codes are arranged in order of their likelihood of being selected as a judgment code.

Manager H2 refers to the selection list displayed on the search terminal 2 and selects a candidate code that corresponds to one or more judgment codes. In this case, failure response history information is created with the selected candidate code as the judgment code. Furthermore, manager H2 can refer to a sorted selection list by inputting sorting conditions via the search terminal 2 to further narrow down the multiple candidate codes included in the selection list. Using such a selection list, manager H2 can easily select an appropriate judgment code.

In addition, instead of entering the results of the response work into the field terminal 108, maintenance worker H1 may enter them into a separate format, such as a report. In other words, information equivalent to a failure response record is compiled into a report or the like. In this case, manager H2 may enter the contents of the report or the like into the search terminal 2. The search terminal 2 and server device 3 may regard the entered information as a failure response record.

Figure 2 shows a refrigerator 200, which is a second example of an apparatus in which the analysis system 1 can be used. The refrigerator 200 has a storage cabinet 201 inside the housing. For example, the storage cabinet 201 is divided into storage rooms 202, 203, 204, and 205. For example, the storage room 204 can be opened and closed by a sliding door 206. The sliding door 206 is realized by a case 208 sliding along a door rail 207.

Refrigerator 200 generates cold air through a refrigeration cycle including compressor 209 and cooler 210. The generated cold air is supplied to storage cabinet 201. Control board 211 is electrically connected to devices such as compressor 209 and cooler 210 via wiring (not shown). Control board 211 is capable of overall control of the operation of refrigerator 200. A switch 212 covered by a switch cover is provided on the wall of the interior of storage cabinet 201. For example, when switch 212 is operated, control board 211 may change the operating state of refrigerator 200, etc. If a malfunction occurs in refrigerator 200, control board 211 may display an error code indicating the malfunction on LCD panel 213, etc.

In the example of Figure 2, the maintenance technician belongs to a company that repairs refrigerator 200. When a report is received that refrigerator 200 has broken down, the maintenance technician goes to the building where refrigerator 200 is installed and performs response work such as replacing the broken part of refrigerator 200.

Although not shown in Figure 2, the maintenance worker inputs information indicating the results of the response work into the field terminal 108, as in the example of Figure 1. Based on the input information, the field terminal 108 creates information on the failure response record including the input free text. The manager belongs to the same company as the maintenance worker. The manager inputs information on the failure response record into the search terminal 2. At this time, a selection list is created by the server device 3.

After the owner of refrigerator 200 responds to a malfunction of refrigerator 200, information regarding the record of the response to the malfunction may be sent to server device 3. In this case, for example, the owner may input information regarding the record of the response to the malfunction via a device such as a smartphone. The information may include free text input by the owner. The smartphone or server device 3 may create a malfunction response record based on the input information. Even in this case, the manager may assign a judgment code to the malfunction response record.

Next, the analysis system 1 will be described with reference to Figures 3 and 4. In the following, the analysis system 1 will be described when processing the failure response record of the second example shown in Figure 2.
Fig. 3 is a hardware configuration diagram of the analysis system according to the first embodiment. Fig. 4 is a functional block diagram of the analysis system according to the first embodiment.

In the analysis system 1, one or more search terminals 2 are communicatively connected to a server device 3 via a network.

For example, the search terminal 2 is composed of a personal computer (PC). The search terminal 2 has a communication I/F 2a, a processing circuit, an input I/F 2b, and an output I/F 2c. The processing circuit includes a processor 2d and a memory. The memory is composed of a main memory device 2e and an auxiliary memory device 2f. Note that the processing circuit may refer to equipment such as multiple electronic circuits stored in separate housings.

The communication I/F 2a is an interface for connecting to a network such as a wired LAN or wireless LAN. The communication I/F 2a enables the search terminal 2 to communicate with other devices such as the server device 3. The input I/F 2b is an interface that accepts input operations by the administrator from the input device 2g. Input devices include a mouse, keyboard, touch pen, microphone, etc. The output I/F 2c is an interface that displays and outputs various information to the output device 2h. The output device 2h is a display, etc.

For example, processor 2d is a device that performs arithmetic processing, such as a CPU (Central Processing Unit), central processing unit, etc. Each function of search terminal 2 is realized by processor 2d reading and executing an information processing program stored in at least one of main storage device 2e and auxiliary storage device 2f, which are memories. In Figure 2, each function realized by processor 2d is represented as the configuration of processor 2d. Various types of information for realizing each function of search terminal 2 are stored in memory. In Figure 2, as an example, the information stored in memory is represented as the configuration of auxiliary storage device 2f.

For example, the server device 3 is configured by a PC. For example, the server device 3 includes a communication I/F 3a and a processing circuit. The communication I/F 3a and the processing circuit may have the same configuration and functions as the communication I/F 2a and processing circuit of the search terminal 2, respectively. The processing circuit includes a processor 3b and memories: a main storage device 3c and an auxiliary storage device 3d. The processor 3b, the main storage device 3c, and the auxiliary storage device 3d may have the same configuration and functions as the processor 2d, the main storage device 2e, and the auxiliary storage device 2f of the search terminal 2, respectively. In Figure 3, each function realized by the processor 3b is represented as the configuration of the processor 3b. Also, in Figure 3, as an example, the information stored in the memory is represented as the configuration of the auxiliary storage device 3d.

In addition, some of the functions of the search terminal 2 may be realized by dedicated hardware, and other functions may be realized by the processing circuitry of the search terminal 2. Also, some of the functions of the server device 3 may be realized by dedicated hardware, and other functions may be realized by the processing circuitry of the server device 3.

The server device 3 may be implemented on a cloud server. In this case, the processing circuit is composed of multiple partial circuits. The multiple partial processing circuits are provided in each of the multiple devices that make up the cloud server. The multiple devices that make up the cloud server may each be provided in a different building.

The search terminal 2 has the following functions: a display unit 20, an input unit 21, a reading unit 22, and an acquisition unit 23. The display unit 20 controls the content displayed on the output I/F 2c. For example, the display unit 20 causes the output I/F 2c to display a search user interface screen for creating a failure response history. The input unit 21 detects information entered via the input I/F 2b and reflects this in the operation of the equipment. The reading unit 22 causes the server device 3 to read the failure response record information selected by the administrator via the input unit 21. The acquisition unit 23 transmits information to be added to or appended to the failure response record, based on the information selected on the search user interface, acquired from a network, etc., to the server device 3.

The server device 3 has the following functions: a receiving unit 30, an extraction unit 31, an association calculation unit 32, a performance learning unit 33, an importance calculation unit 34, a priority calculation unit 35, a correspondence learning unit 36, a list unit 37, a display control unit 38, a history creation unit 39, and a setting request unit 40.

The receiving unit 30 receives information about the failure response record from the field terminal 108 or the search terminal 2. The failure response record may include identification information for the equipment for which the failure was handled, the error code at the time of the failure, signal data transmitted from the equipment, the installation location of the equipment, and identification information for the maintenance personnel who handled the failure. Furthermore, this information included in the failure response record may be collected by the receiving unit 30 from another database or the equipment itself, and associated with the failure response record.

Failure response records also include multiple fields in which free text can be entered. The multiple fields include at least a first field for free text on the cause of the failure, indicating the cause of the failure, and a second field for free text on the details of the measures taken to address the failure during the response work. The multiple fields may also include a third field for free text on the arrival status, which is information collected from the time the maintenance technician arrives at the site until the cause is determined.

The following describes an example in which a failure response record contains free text for the cause of the failure in the first entry field and free text for the remedial action in the second entry field. The process when "equipment name for aggregation" is selected as the judgment code is also described. "Equipment name for aggregation" refers to an equipment name selected as a valid classification for aggregation purposes, rather than the actual product name, etc. Note that multiple entry fields may include fields other than those shown in this example, as long as they include first and second entry fields in which the free text written therein has different meanings.

The extraction unit 31 has a first extraction unit 31a and a second extraction unit 31b. The first extraction unit 31a extracts specified target terms from the free text included in the failure response record. The target terms are terms required when creating a selection list and are determined on a case-by-case basis. In this example, the first extraction unit 31a extracts device names as target terms. At this time, the first extraction unit 31a associates the extracted target terms with which of multiple entry fields the target terms are entered in. The second extraction unit 31b extracts words indicating measures to be taken in response to the failure from the free text included in the failure response record. At this time, the second extraction unit 31b extracts words indicating operations as measures. The measures basically describe measures to be taken on the equipment. The second extraction unit 31b associates the extracted measures with the device names, which are target terms. An example of the operation of the extraction unit 31 is explained using Tables 1 and 2 below.

Table 1 shows examples of free text included in three failure response records. The second column is free text on the cause of the failure, which is the first entry field. The third column is free text on the details of the treatment, which is the second entry field. The first extraction unit 31a extracts device names shown in bold in Table 1 as target terms. The second extraction unit 31b extracts treatment words shown in underlined parts in Table 1. Table 2 shows the description locations and treatments associated with the target terms extracted in the example of Table 1. Note that when the extraction unit 31 extracts words from the free text, known methods for extracting words may be used.

The relevance calculation unit 32 references the code DB 50 to retrieve a candidate list for the currently targeted judgment code. The code DB 50 stores candidate lists corresponding to each of the multiple judgment codes that may be targeted. The candidate list is a list in which multiple candidate codes corresponding to the judgment code are arranged in a specified initial order. Below, we will explain the multiple candidate codes corresponding to the target judgment code. Specifically, we will explain each candidate code included in the aggregation device name, which is the judgment code. In this case, the candidate list includes the candidate codes "inside the cabinet," "inside the cabinet thermometer," "refrigerator," "sliding door," "door rail," "switch," and "switch cover," in the order listed.

The relevance calculation unit 32 calculates the relevance between the target term extracted by the extraction unit 31 and each of the multiple candidate codes that have been called up. When the extraction unit 31 extracts multiple target terms from the failure response record, the relevance calculation unit 32 calculates the relevance between each of the multiple target terms and the candidate code. The relevance is an index value that indicates the textual or semantic relevance between the target term and the candidate code. For example, the relevance may be the textual similarity between the target term and the candidate code, or the attribute similarity between the target term and the candidate code. This similarity may be calculated based on information such as attributes attached to the target term and the candidate code using an ontology.

The edit distance between the target term and the candidate code may also be used as the textual similarity. Specifically, when the target term is an equipment name and the candidate code is a tabulation equipment name, the relevance calculation unit 32 calculates the edit distance between the equipment name and the tabulation equipment name as the relevance. For example, when the equipment name is "door" and the tabulation equipment name is "door rail," the minimum number of steps required to replace "door" with "door rail" is three, with one character insertion, deletion, or substitution being one step. In this case, the relevance calculation unit 32 calculates the relevance between the equipment name and the tabulation equipment name as the edit distance of 3.

The relevance calculation unit 32 associates the most highly relevant candidate code among multiple candidate codes with a given target term. The relevance calculation unit 32 may also associate the target term with one or more candidate codes whose relevance exceeds a specified threshold. In this case, the relevance calculation unit 32 may temporarily associate each of the multiple candidate codes in the candidate list with one or more target terms extracted from the currently processed failure response record.

The performance learning unit 33 calculates the performance importance of each action extracted from each of the multiple failure response histories based on multiple past failure response histories stored in the history DB 51. The performance importance is calculated based on the number of times the associated action associated with the judgment code selected in the failure response history was taken.

The importance calculation unit 34 associates the action extracted by the extraction unit 31 from the failure response record with one of multiple candidate codes. At this time, the importance calculation unit 34 associates the action with the candidate code associated with the target term corresponding to the action. Table 3 below shows examples of device names, which are target terms associated by the importance calculation unit 34, actions, and target device names, which are candidate codes.

As shown in Table 3, based on row #3-1, the action "inspection" corresponds to the equipment name "door" in the failure response record. In this case, the importance calculation unit 34 associates the action "inspection" with the aggregation equipment name "door rail."

The importance calculation unit 34 calculates the overall importance, which is the importance of the extracted actions. To do this, the importance calculation unit 34 first retrieves the basic importance for each action from the action DB 52. The basic importance is information stored in the action DB 52 and is the importance that has been set in advance for the action. The basic importance may also be set or updated manually in advance.

The processing of the performance learning unit 33 will now be described in more detail. The performance learning unit 33 retrieves multiple failure response histories stored in the history DB 51. The failure response histories include, for each of multiple judgment codes, candidate codes assigned to the judgment code. The candidate codes are associated with actions as shown in Table 3. The performance learning unit 33 identifies actions associated with judgment codes in each of the multiple failure response histories as related actions. Note that the related actions may include not only the action words associated with the candidate codes, but also synonyms of the actions associated with the candidate codes. For each action stored in the action DB 52, the performance learning unit 33 calculates the number of related actions, i.e., the number of times the action has become a related action. The performance learning unit 33 calculates the performance importance for each action so that the more times the action has become a related action, the higher the importance. For example, the performance importance value is calculated so that the more times the action has become a related action, the lower the value. Note that the performance importance may also be calculated based on the proportion of actions that have become related actions, rather than the number of times they have become related actions. The performance learning unit 33 associates the calculated performance importance with each of the multiple actions stored in the action DB 52. For example, the performance learning unit 33 may update the performance importance every time a new failure response record is stored in the history DB 51.

The importance calculation unit 34 calculates the overall importance based on at least one of the basic importance and the actual importance calculated by the performance learning unit 33. The basic importance, actual importance, and overall importance may be treated such that the smaller the value, the higher the importance. For example, it may be estimated that the smaller the importance value of a certain action, the more likely a candidate code corresponding to that action will be selected as a judgment code. As an example, as shown in the following formula (1), the importance calculation unit 34 calculates the overall importance of a certain action by calculating the weighted sum of the basic importance and actual importance.
Overall importance = A x basic importance + B x performance importance (1)
A and B are constants that are set in advance based on a person or machine learning. A or B may also be set to 0. When B is set to 0, the importance is determined only by the basic importance. When A is set to 0, the importance is determined only by the performance importance. Table 4 below shows an example of a method for calculating the overall importance.

In Table 4, the number of times the relevant procedure has been performed is shown in parentheses in the performance importance row. In this example, the relationship y<x<z holds. Therefore, for example, the performance importance value of the "maintenance" procedure is the smallest and therefore the most important. In Table 4, the constants A=B=1 are set.

The priority calculation unit 35 calculates an overall priority for each target term extracted by the extraction unit 31 based on at least one of the base priority and the occasional priority. Specifically, the priority calculation unit 35 first retrieves the base priority associated with the extracted target term from the target term DB 53, which stores anticipated target terms. The base priority is a preset priority stored in the target term DB 53. Note that the base priority may also be manually set or updated in advance.

The priority calculation unit 35 calculates an ad hoc priority for each target term using the target failure response record. The priority calculation unit 35 calculates the ad hoc priority so that its value differs depending on which of the multiple description columns included in the target failure response record the target term is included in. For example, the priority calculation unit 35 calculates the ad hoc priority when the target term is listed only in the first description column to be higher than the ad hoc priority when the target term is listed only in the second description column. The priority calculation unit 35 calculates the ad hoc priority when the target term is listed in both the first and second description columns to be higher than the ad hoc priority when the target term is listed only in the first description column. Note that the base priority value and ad hoc priority value may be calculated so that the higher the priority, the smaller the value. Tables 5 to 7 below show examples of base priority, ad hoc priority, and overall priority, respectively.

Table 5 shows an example of set base priorities. In this example, the target term "switch" has the highest base priority and the smallest value. Table 6 shows an example of setting values for ad hoc priority calculated based on the location of the target term. According to Table 6, the ad hoc priority is highest when the target term is written in both the first field for the cause of the failure and the second field for the remedial action. The priority is lowest when the target term is written only in the second field for the remedial action. Note that the base priority values and ad hoc priority values are not limited to the examples shown in Tables 5 and 6. For example, when the target term is written only in the second field, the ad hoc priority value may be set to "5" instead of "3".

As shown in Table 7, the priority calculation unit 35 calculates the overall priority for a certain target term by calculating the weighted sum of the base priority and the occasional priority. Specifically, the priority calculation unit 35 calculates the overall priority based on the following formula (2).
Overall priority = C x basic priority + D x occasional priority (2)
C and D are constants that are preset based on humans or machine learning. Alternatively, C or D may be set to 0. When D is set to 0, the priority is determined only by the base priority. When C is set to 0, the priority is determined only by the ad hoc priority. In the example of Table 7, the constants C = D = 1 are set.

The response learning unit 36 calculates a response index value assigned to each of the multiple candidate codes based on the multiple failure response histories stored in the history DB 51. The response index value may be the number of responses assigned to the failure response record, or a response ratio calculated based on the number of responses. Furthermore, if information contained in an error code included in the multiple failure response histories corresponds to a candidate code, the response learning unit 36 may add the number of pieces of information corresponding to the candidate code to the number of responses in the related index. For example, if the candidate code is a counting device name, the device included in the error code may be added to the number of responses for the corresponding counting device name. For example, if the candidate code is a cause code, the cause included in the error code may be added to the number of responses for the corresponding cause code. The response learning unit 36 stores information associating the response index value with the candidate code in the related DB 54. For example, the response learning unit 36 may update the response index value each time a new failure response record is stored in the history DB 51. Tables 8 and 9 below show examples of the number of responses and the response ratio, respectively.

Table 8 shows the number of times that each candidate code, which is a tabulation device name, was listed for multiple error codes, Code A to Code Z, included in the failure response history. In this example, it is assumed that no device corresponding to a candidate code was listed for Code B to Code Y. For example, the response learning unit 36 calculates that the number of times that the tabulation device name "inside" was listed for Code A to Code Z was five. For example, the response learning unit 36 calculates that the number of times that the tabulation device name was listed when the error code was Code Z was two for "door rail,""switch," and "switch cover," and zero for "inside,""insidethermometer/hygrometer,""refrigerator," and "sliding door." The number of times that the code was listed may be calculated as a weighted sum of the number of times the code was selected as a candidate code and the number of times the code was included in an error code, as shown in the following equation (3).
Number of times of correspondence = E × number of times selected as a candidate code
+ F × the number of times included in the error code (3)
E and F are constants that are set in advance based on a person or machine learning, and may be set to 0.

Table 9 shows the correspondence ratio calculated using the same number of correspondences as Table 8. The correspondence learning unit 36 may use the correspondence ratio instead of the number of correspondences. The correspondence ratio is the total value obtained by adding the individual correspondence ratios calculated for each error code for all error codes. By using the correspondence ratio as the correspondence index value, the correspondence index value of a candidate code that frequently appears only with a certain error code, such as "switch cover" in Table 9, may be larger than when the number of correspondences is used.

In addition to the error code, the number of responses or response rate may be calculated for each device installation location and added to the response index value of the candidate code corresponding to the device. For example, the device installation location may be divided into "indoors" and "outdoors," and the number of responses or response rate may be calculated.

The list unit 37 calls up the candidate list corresponding to the target judgment code, rearranges the order of the multiple candidate codes included in the candidate list, and creates a selection list. In this case, the list unit 37 creates the selection list by performing the rearrangement multiple times, as shown in Table 10 below. Table 10 shows an example where the free text for the cause of the failure in the first entry field reads "Liquid spilled inside the refrigerator," and the free text for the action to be taken in the second entry field reads "Inspect the door. Maintain the switch."

First, the list unit 37 identifies one or more inclusion candidate codes from among the multiple candidate codes that contain at least one of the one or more target terms extracted by the extraction unit 31 as a character string. The list unit 37 creates a first list in which the multiple candidate codes are ranked so that the inclusion candidate codes are ranked higher than those that are not inclusion candidate codes. Note that in Table 10, the ranking in the first list is indicated as ": 1st place." In the first list, multiple inclusion candidate codes all have the same ranking. In the first list, candidate codes that are not inclusion candidate codes all have the same ranking.

In the example of Table 10, the target terms "interior," "door," and "switch" are extracted from the target failure response record, so the six inclusion candidate codes are "interior," "interior thermo-hygrometer," "sliding door," "door rail," "switch," and "switch cover." In the first list, the target device names that are these six inclusion candidate codes are all set to the same order of 1st place. "Refrigerator," which is not an inclusion candidate code, is set to 7th place.

The list unit 37 then creates a second list by further ranking candidate codes with the same rank among the multiple candidate codes included in the first list. In Table 10, the ranking in the second list is indicated as "1st place." The list unit 37 rearranges candidate codes with the same rank based on at least one of the relevance calculated by the relevance calculation unit 32 and the importance calculated by the importance calculation unit 34. In this case, the list unit 37 rearranges the multiple candidate codes so that the higher the relevance of the target term associated with each candidate code and the higher the importance of the associated action, the higher the ranking. Furthermore, the list unit 37 may rearrange candidate codes with the same rank based on the priority calculated by the priority calculation unit 35 in addition to the relevance and importance. In this case, the list unit 37 rearranges the multiple candidate codes so that the higher the relevance of the target term associated with each candidate code, the higher the importance of the associated action, and the higher the priority of the associated target term, the higher the ranking.

Specifically, the list unit 37 calculates an index value, which is a weighted sum of the relevance, importance, and priority, for each of the multiple candidate codes. The list unit 37 compares the index values assigned to candidate codes with the same ranking and sorts them so that the smaller the index value, the higher the ranking, to create the second list. The index value may be calculated as a weighted sum of the number of times the code has been selected as a candidate code and the number of times it has been included in an error code, as shown in the following formula (4).
Index value = G × relevance + H × importance + I × priority (4)
G, H, and I are constants that are set in advance based on a person or machine learning. I may also be set to 0. When I is set to 0, the index value is calculated from the relevance and importance. In the example of Table 10, H = G = I = 1. The list unit 37 arranges six counting device names that are ranked the same as the first in the first list so that the smaller the index value, the higher the ranking. Multiple candidate codes that are ranked the same in the first list and have the same index value in the second list are assigned the same rank in the second list. In the example of Table 10, the list unit 37 ranks "inside storage" and "sliding door" in fourth place.

Then, the list unit 37 creates a third list by further ranking and rearranging candidate codes that have the same ranking among the multiple candidate codes included in the second list. Note that in Table 10, the ranking in the third list is shown as "1st place." The list unit 37 rearranges multiple candidate codes that have the same ranking in the second list so that the higher the correspondence index value calculated by the correspondence learning unit 36, the higher the ranking, and creates the third list.

In the example of Table 10, the number of correspondences is used as the correspondence index value. The number of correspondences between "interior" and "sliding door", which were both ranked fourth in the second list, is 5 and 8, respectively. The list unit 37 ranks "sliding door", which has a high correspondence index value, i.e., a high number of correspondences, as fourth, and "interior", which has a low number of correspondences, as fifth. The list unit 37 rearranges multiple candidate codes based on the updated ranking, and creates a third list. Table 11 below shows the third list created from the example of Table 10.

Then, the list unit 37 creates a fourth list by deleting candidate codes that are not applicable from the multiple candidate codes present in the third list based on at least one of the model information and error code information contained in the failure response record. If there are no candidate codes that are not applicable, the list unit 37 creates a fourth list that is the same as the third list. For example, when the fourth list is created based only on model information, the list unit 37 obtains model correspondence information from the model DB 55 that indicates the correspondence between the model and multiple candidate codes.

For example, there may be codes that are listed as candidate codes in the candidate list but are not included in the actual device model and therefore will never be selected as candidate codes. The model compatibility information indicates whether each of multiple candidate codes is "eligible" or "ineligible" for selection as a candidate code for each model. The list unit 37 compares the model compatibility information with the models included in the failure response record and deletes from the third list any candidate codes that are "ineligible" for that model. Tables 12, 13, and 14 below are examples of when the fourth list is created based solely on model information.

In the example of Table 12, the target failure response record includes information that the model of the device for which the failure was handled is model A. In this case, as shown by the underlined parts in Table 13, according to the model support information, "interior," "interior thermo-hygrometer," "refrigerator," and "sliding door" are "target." Furthermore, "door rail," "switch," and "switch cover" are "not target." As shown in Table 14, the list unit 37 creates a fourth list by deleting the "not target" aggregation equipment names from the third list.

If the fourth list is further created based on an error code, the list unit 37 similarly removes candidate codes from the list shown in Table 14 based on the "error response information," and creates the fourth list. Specifically, the list unit 37 obtains the error response information from the error DB 56. The error response information is information that indicates the correspondence between an error code and a plurality of candidate codes. In other words, the error response information is information that indicates, for each of a plurality of candidate codes, whether the code is "eligible" or "not eligible" for selection as a candidate code. The list unit 37 removes candidate codes that are "not eligible" based on the error code indicated in the failure response record currently in question from the list shown in Table 14, and creates the fourth list.

In addition, when the fourth list is created based only on the error code, the list unit 37 similarly creates the fourth list by removing candidate codes that are "not applicable" from the third list based on the error response information and the error code contained in the failure response record.

Then, the list unit 37 outputs the fourth list as a selection list.

The list unit 37 may output any one of the first, second, and third lists as the selection list instead of the fourth list. In any of the first, second, third, fourth, and selection lists, the multiple candidate codes are arranged so that the higher the probability of being selected as a judgment code, the higher the ranking. The more recently created the list, the more accurate the probability of being selected as a judgment code.

The display control unit 38 displays the selection list output by the list unit 37 on the search terminal 2. For example, the display control unit 38 displays a search user interface screen on the display unit 20 of the search terminal 2. For example, the display control unit 38 first displays only the highest-ranked candidate code among the multiple candidate codes included in the selection list. A button for accepting selection operations is also displayed near the highest-ranked candidate code. When the button for accepting selection operations is operated by the administrator, the display control unit 38 displays the multiple candidate codes included in the selection list so that the higher the ranking, the higher the position. The display control unit 38 accepts a selection of a candidate code corresponding to a certain judgment code from the administrator via the search terminal 2. Note that the display control unit 38 may accept a selection of multiple candidate codes for one judgment code.

The display control unit 38 can further accept input of sorting conditions or rearrangement conditions via the search user interface.

The sorting conditions are conditions for deleting non-applicable candidate codes from the selection list while maintaining the ranking of the multiple candidate codes included in the selection list. The display control unit 38 deletes candidate codes from the selection list that do not satisfy the sorting conditions, and displays the multiple candidate codes included in the selection list after this deletion process so that the higher the ranking, the higher the position.

For example, the sorting condition may be a condition that accepts input into at least one of the first and second fields and displays candidate codes corresponding to the target terms entered in the field for which input was accepted. In this case, the display control unit 38 may delete candidate codes corresponding to target terms that are not entered in the field for which input was accepted from the selection list, and display the multiple candidate codes included in the selection list after the deletion process has been completed.

For example, the sorting condition may be a condition that accepts input of a boundary value for performance importance and displays candidate codes corresponding to procedures with performance importance equal to or greater than the accepted boundary value. In this case, the display control unit 38 may delete candidate codes corresponding to procedures with performance importance less than the boundary value from the selection list and display multiple candidate codes included in the selection list after the deletion process has been completed.

The sorting conditions are conditions for rearranging the order of multiple candidate codes included in the selection list. When the display control unit 38 receives input of the sorting conditions, the list unit 37 rearranges the multiple candidate codes included in the candidate list based on the sorting conditions to create a selection list. The display control unit 38 then displays the rearranged selection list on the search terminal 2.

For example, the sorting condition is a condition for changing a constant that is the weighting ratio when calculating each value, such as importance value, priority value, index value, etc. Specifically, the display control unit 38 accepts input of a condition for changing the weighting ratio between the base priority and the ad hoc priority as the sorting condition. When creating a selection list from the candidate list, particularly when calculating the priority, the list unit 37 calculates the overall priority of the target term using constants C and D that correspond to the input ratio. After the selection list is created again in this way, the display control unit 38 displays the selection list reflecting the priority sorting condition.

The history creation unit 39 assigns the candidate code selected by the display control unit 38 as the corresponding judgment code to the target failure response record. Note that if multiple candidate codes are selected, the history creation unit 39 may assign multiple candidate codes as judgment codes. When all judgment codes related to the target failure response record have been assigned, the history creation unit 39 creates failure response history information including the failure response record and the assigned judgment codes. The history creation unit 39 stores the created failure response history in the history DB 51. In other words, the history DB 51 stores multiple failure response histories created in the past by the history creation unit 39.

The setting request unit 40 compares the failure response history contained in the history DB 51 with the action DB 52 to extract actions for which a basic importance level has not been set. The setting request unit 40 creates an importance request list listing the extracted actions. The setting request unit 40 compares the failure response history contained in the history DB 51 with the target term DB 53 to extract target terms for which a basic priority level has not been set. The setting request unit 40 creates a priority request list listing the extracted target terms. The setting request unit 40 compares the failure response history contained in the history DB 51 with the model DB 55 to extract combinations of aggregation equipment names and models for which target status has not been set. The setting request unit 40 creates a combination request list listing the extracted combinations.

The setting request unit 40 may cause the search terminal 2 to display a setting request screen displaying at least one of the importance request list, priority request list, and combination request list at any timing, such as when operated on the search terminal 2. When settings are input on the setting request screen, the information in each DB is updated.

Next, an example of a search user interface will be described with reference to FIG.
FIG. 5 is a diagram showing an example of a search user interface displayed by the analysis system according to the first embodiment.

Figure 5 shows a search user interface (hereinafter also referred to as the "selection UI") in which the display control unit 38 displays multiple candidate codes Ex1, sorting conditions Ex2, and rearrangement conditions Ex3 for a certain judgment code. The selection UI displays a button Ex4 that accepts selection operations. When the "Change" button Ex4 is operated, multiple candidate codes Ex1 are displayed in addition to the highest-ranked candidate code, "Switch."

In this example, the sorting condition Ex2 is selected and entered as "Location of equipment name" to display only candidate codes corresponding to the target terms listed in "Procedure content."

In this example, the sorting condition Ex3 is entered as a condition for sorting after calculating an overall priority in which the weight constant D of the ad hoc priority is greater than the weight constant C of the base priority. Specifically, the priority calculation unit 35 displays multiple candidate codes Ex1 sorted using the overall priority calculated with the weight constant C of the base priority = 0.5 and the weight constant D of the ad hoc priority = 1.0. Note that when "Sort by prioritizing important device names" in the lower row is selected and entered, sorting is performed after calculating an overall priority in which the weight constant C of the base priority is greater than the weight constant D of the ad hoc priority.

Next, an example of a setting request screen will be described with reference to FIG.
FIG. 6 is a diagram showing an example of a setting request screen displayed by the analysis system according to the first embodiment.

As shown in FIG. 6, the setting request screen displays an importance request list Ex5, a priority request list Ex6, and a combination request list Ex7. The setting request unit 40 accepts updates to the setting values in each list. When a setting is input, the setting request unit 40 updates the DB information corresponding to the input setting value to the input.

Next, the operation of the analysis system 1 will be described with reference to FIG.
FIG. 7 is a flowchart for explaining an outline of the operation of the analysis system according to the first embodiment.

For example, the flowchart begins when a maintenance technician completes troubleshooting work and inputs information into the field terminal 108. In step S1, the field terminal 108 creates a fault troubleshooting record and transmits it to the server device 3.

Then, in step S2, the server device 3 stores the failure response record. The administrator calls up the failure response record from the search terminal 2.

Then, in step S3, the server device 3 creates a selection list for one or more judgment codes corresponding to the target failure response record.

Then, in step S4, the server device 3 displays a search UI on the search terminal 2. The selection list created in step S3 is displayed in the search UI. In the search UI, the display control unit 38 accepts input of candidate codes from the administrator. Note that if a sorting condition is selected in the search UI, the server device 3 creates a selection list based on the sorting condition and displays it again in the search UI.

Then, in step S5, the display control unit 38 of the server device 3 determines whether a registration operation has been performed for the target failure response record. At this time, the administrator performs the registration operation after entering all management codes related to the target failure response record. If a registration operation has not been performed in step S5, the server device 3 repeats the operations from step S4 onwards.

If a registration operation is performed in step S5, the operation of step S6 is performed. In step S6, the server device 3 creates a failure response history for the target failure response record. The server device 3 stores the failure response history in the history DB 51.

Then, in step S7, the performance learning unit 33 of the server device 3 updates the performance importance in the treatment DB 52 based on the failure response record registered in step S6. The response learning unit 36 of the server device 3 updates the response index value in the related DB 54 based on the failure response history registered in step S6.

Then the operation of the flowchart ends.

According to the first embodiment described above, the analysis system 1 comprises an extraction unit 31, a relevance calculation unit 32, an importance calculation unit 34, and a list unit 37. The list unit 37 creates a selection list in which the order of multiple candidate codes is rearranged based on the relevance and importance. The selection list lists multiple candidate codes so that the higher the probability of them being selected as a judgment code, the higher the ranking. Therefore, the analysis system 1 can provide options that are likely to be selected as options.

The analysis system 1 also includes a display control unit 38. The display control unit 38 displays multiple candidate codes included in the selection list so that the higher the ranking, the higher the position. As a result, the analysis system 1 can visually provide, for example, a manager or the like, multiple options that are likely to be selected, including their ranking. Furthermore, for example, when searching for a candidate code that is considered appropriate from the candidate list, the manager can refer to the candidate codes in order, starting with the most likely. As a result, the manager's work time can be reduced in management tasks such as creating a failure response history. Furthermore, because the more likely candidate codes are displayed, the occurrence of an incident in which the manager selects a code with low management value, such as "Other," can be reduced.

The display control unit 38 also first displays only the highest-ranked candidate code from among the multiple candidate codes included in the selection list. This reduces the amount of time required by administrators to perform administrative tasks such as creating a failure response history.

Furthermore, when creating the selection list, the list unit 37 arranges multiple candidate codes so that the higher the relevance and the higher the demand, the higher the ranking. Therefore, multiple candidate codes can be arranged in the selection list so that the more likely they are to be selected as a judgment code, the higher their ranking. For example, the target term is an equipment name, the candidate code is a tabulation equipment name, and the relevance is the edit distance between the equipment name and the tabulation equipment name. In this case, due to the characteristics of free text, the selection list can be created based on an edit distance that is assumed to indicate a high relevance. As a result, multiple candidate codes can be arranged so that the more likely they are to be selected as a judgment code, the higher their ranking.

The analysis system 1 also includes a history creation unit 39 and a performance learning unit 33. The performance learning unit 33 calculates the number of times an action has become a related action based on the failure response history created in the past. The importance calculation unit 34 calculates the importance of the action using performance importance, which increases the more times an action has become a related action. In other words, the importance calculated by the analysis system 1 reflects the performance of the number of times an action has become a related action in the past. As a result, the analysis system 1 can increase the likelihood that the provided options will be selected.

The importance calculation unit 34 also calculates the importance of the action as the weighted sum of the actual importance and the set basic importance. Therefore, the importance is not only based on actual performance, but also on the settings of what actions a person or other person considers important. For example, even if a sufficient amount of failure response history has not been accumulated to ensure accuracy, actions that should be emphasized can be set based on the basic importance. As a result, the possibility that the provided options will be selected can be increased.

The analysis system 1 also includes a priority calculation unit 35. The priority calculation unit 35 calculates the priority of the target term. The list unit 37 arranges multiple candidate codes so that the higher the priority of the associated target term, the higher the ranking. This allows the analysis system 1 to provide options that are likely to be selected as options.

The priority calculation unit 35 also calculates an ad hoc priority, the value of which varies depending on whether the target term is included in the first entry field or the second entry field. The priority calculation unit 35 calculates the priority of the target term as the weighted sum of the ad hoc priority and the base priority. For example, when a first target term is entered in the first entry field, "Cause of Failure," and a second target term is entered in the second entry field, "Details of Treatment," there may be a case where the candidate code associated with the first target term is more likely to be selected than the candidate code associated with the second target term. By calculating the priority based on the entry field of the target term in this way, the analysis system 1 can increase the likelihood that the provided option will be selected.

The list unit 37 also calculates an index value that is a weighted sum of relevance, importance, and priority. Therefore, the list unit 37 can create a list with more specific indices in which multiple candidate codes are arranged so that the higher the relevance, importance, and priority, the better the ranking.

The display control unit 38 also accepts input of sorting conditions after displaying multiple candidate codes. The list unit 37 rearranges the order of the selection list based on the sorting conditions. Therefore, an administrator or the like can input sorting conditions after checking the display of the selection list. Administrators or the like can also refer to multiple candidate codes arranged in accordance with the sorting conditions. More specifically, the sorting conditions may be conditions that change the weighting ratio between the base priority and the priority at any time. In this case, candidate codes that meet the sorting conditions deemed necessary by the administrator or the like and are likely to be selected may be displayed. As a result, the analysis system 1 can reduce the administrator's work time.

Furthermore, the display control unit 38 displays the multiple candidate codes included in the selection list after deleting candidate codes that do not meet the received sorting conditions, with the higher the ranking, the higher the candidate code. This allows candidate codes that meet the sorting conditions deemed necessary by the administrator or other personnel and are likely to be selected to be displayed. As a result, the analysis system 1 can reduce the administrator's work time.

The analysis system 1 also includes a response learning unit 36. The response learning unit 36 calculates a response index value based on the failure response history. When there are two or more candidate codes with the same ranking determined by information including relevance and importance, the list unit 37 arranges the two or more candidate codes in a third list so that the higher the response index value, the higher the ranking. This allows the analysis system 1 to provide options that are likely to be selected as options.

The list unit 37 also identifies multiple candidate inclusion codes that contain multiple target terms as character strings, and creates a first list. The list unit 37 creates a second list from the first list. This allows the analysis system 1 to provide options that are likely to be selected as options.

The processes disclosed herein may also be applied when a code other than "aggregation equipment name" is selected as the judgment code. In this case, the words extracted or used in each process may be selected in accordance with the judgment code. For example, when "cause code" is selected as the judgment code, the second extraction unit 31b may particularly extract words indicating causes rather than words indicating treatment from the free text. The relevance calculation unit 32 may calculate the relevance between the cause code and the target term, cause, as the relevance. The performance learning unit 33 may calculate the number of times a word indicating a cause, rather than a treatment, is associated with the target code assigned to the failure response history and becomes a related cause, thereby calculating the performance importance. The importance calculation unit 34 may calculate the importance using a word indicating a cause, rather than a treatment, as the target term.

In addition, when creating the fourth list, the list unit 37 may determine whether a device is "targeted" or "not targeted" based on a database related to information contained in the failure response record, such as other codes issued by the device, information entered by the maintenance staff, building names, etc., rather than error codes.

In addition, the setting information contained in each DB may be updated by importing corresponding database information from outside.

Embodiment 2.
Fig. 8 is a hardware configuration diagram of the analysis system according to the second embodiment. Fig. 9 is a flowchart for explaining an outline of the operation of the analysis system according to the second embodiment. Note that the same reference numerals are used to designate parts that are the same as or equivalent to parts in the first embodiment. Explanation of these parts will be omitted.

In embodiment 2, analysis system 1 functions as an estimation system. The estimation system estimates the candidate code that is most suitable for the judgment code based on the failure response record. The estimation system creates a failure response history in which the estimated candidate code is assigned as the judgment code to the failure response record.

Specifically, as shown in FIG. 8, the estimation system, which is the analysis system 1, further includes an estimation unit 41 as a function. For example, in embodiment 2, the display control unit 38 does not need to be provided.

When a selection list based on failure response records is created by the list unit 37, the estimation unit 41 estimates that the candidate code with the highest ranking in the selection list is the judgment code to be assigned. In this case, the history creation unit 39 assigns the candidate code estimated by the estimation unit 41 to be a judgment code as the corresponding judgment code to the target failure response record, thereby creating a failure response history.

Figure 9 shows an overview of the operation of the analysis system 1, i.e., the estimation system, in embodiment 2. Step S1 in this flowchart is the same as step S1 in the flowchart of Figure 7 in embodiment 1.

After step S1, in step S11, the server device 3 stores the transmitted failure response record. The server device 3 then calls up the failure response record.

Then, in step S3, the server device 3 creates a selection list, similar to the flowchart of Figure 7 in embodiment 1.

Then, in step S12, the estimation unit 41 of the server device 3 estimates that for all judgment codes related to the target failure response record, the highest-ranked candidate code from the corresponding selection list is the judgment code to be assigned.

Then, in step S13, the history creation unit 39 of the server device 3 creates a failure response history in which the candidate code estimated by the estimation unit 41 in step S12 is used as the judgment code.

Then, in step S7, the server device 3 updates each DB, similar to the flowchart of Figure 7 in embodiment 1.

Then the operation of the flowchart ends.

According to the second embodiment described above, the estimation system comprises an extraction unit 31, a relevance calculation unit 32, an importance calculation unit 34, a list unit 37, and an estimation unit 41. The list unit 37 creates a selection list in which the order of multiple candidate codes is rearranged based on the relevance and importance. The selection list is arranged so that the higher the probability of selection as a judgment code, the higher the ranking. The estimation unit 41 estimates that the candidate code with the highest ranking in the selection list is the judgment code. Therefore, the estimation system can estimate options that are likely to be selected as options.

As described above, the analysis system disclosed herein can be used in a system for managing equipment maintenance records.

REFERENCE SIGNS LIST 1 Analysis system, 2 Search terminal, Communication I/F 2a, Input I/F 2b, Output I/F 2c, 2d Processor, 2e Main storage device, 2f Auxiliary storage device, 2g Input device, 2h Output device, 3 Server device, Communication I/F 3a, 3b Processor, 3c Main storage device, 3d Auxiliary storage device, 20 Display unit, 21 Input unit, 22 Reading unit, 23 Acquisition unit, 30 Receiving unit, 31 Extraction unit, 31a First extraction unit, 31b Second extraction unit, 32 Relevance calculation unit, 33 Performance learning unit, 34 Importance calculation unit, 35 Priority calculation unit, 36 Correspondence learning unit, 37 List unit, 38 Display control unit, 39 History creation unit, 40 Setting request unit, 41 Estimation unit, 50 Code DB, 51 History DB, 52 Treatment DB, 53 Target term DB, 54 Related DB, 55 Model DB, 56 Error DB, 100 Elevator device, 101 Hoistway, 102 Building, 103 Landing, 104 Landing door, 105 Door rail, 106 Cage, 107 Control panel, 108 Field terminal, 200 Refrigerator, 201 Storage, 202, 203, 204, 205 Storage room, 206 Sliding door, 207 Door rail, 208 Case, 209 Compressor, 210 Cooler, 211 Control board, 212 Switch, 213 Liquid crystal panel, Ex1 Candidate code, Ex2 Sorting condition, Ex3 Sorting conditions, Ex4 button, Ex5 importance request list, Ex6 priority request list, Ex7 request list, H1 maintenance staff, H2 management staff, N network, S information center

Claims (16)

A system for analyzing a failure response record including free text entered by a person who responded to a device failure,
an extraction unit that extracts specific target terms and words indicating the content of the measures taken during the failure response from free text included in the failure response record;
a correlation calculation unit that calculates a correlation between each of a plurality of candidate codes that are candidates for a judgment code to be assigned to the failure response record and the target term extracted by the extraction unit;
an importance calculation unit that calculates the importance of the action extracted by the extraction unit and associates the action with one of the plurality of candidate codes;
a list unit that creates a selection list in which the plurality of candidate codes are arranged in a ranking order based on the relevance calculated by the relevance calculation unit and the importance calculated by the importance calculation unit, so that the higher the probability of the candidate code being selected as the judgment code, the higher the ranking; and
An analysis system equipped with a display control unit that displays the plurality of candidate codes included in the selection list so that higher ranking candidate codes are positioned higher;
The analysis system according to claim 1 , further comprising: The analysis system described in claim 2, wherein the display control unit displays only the highest-ranked candidate code among the plurality of candidate codes included in the selection list, and when a selection operation is subsequently received, the plurality of candidate codes included in the selection list are displayed such that the higher the ranking, the higher the position of the candidate code. the relevance calculation unit associates the target term extracted by the extraction unit and the relevance of the target term with a candidate code having the highest relevance with the target term among the plurality of candidate codes;
3. The analysis system according to claim 2, wherein the list unit, when creating the selection list, arranges the plurality of candidate codes so that the higher the relevance of the associated target term and the higher the importance of the associated action, the higher the ranking. a history creation unit that creates a failure response history in which the candidate code selected from the selection list as the judgment code is added to the failure response record;
a performance learning unit that identifies an associated action that is an extracted action and corresponds to the assigned judgment code in each of a plurality of failure response histories created in the past by the history creation unit, and calculates the number of times that the associated action was taken in the plurality of failure response histories for each action;
Further provided with
The analysis system according to claim 2 , wherein the importance calculation unit calculates the importance of the action using a performance importance that increases as the number of times the action becomes the related action increases. The analysis system described in claim 5, wherein the importance calculation unit calculates the importance of a procedure as a weighted sum of the basic importance set for each procedure and the actual importance calculated for each procedure. a priority calculation unit that calculates the priority of the target terms extracted by the extraction unit;
Further provided with
3. The analysis system according to claim 2, wherein, when creating the selection list, the list unit arranges the plurality of candidate codes so that the higher the relevance of the associated target term, the higher the importance of the associated action, and the higher the associated priority, the higher the ranking. The analysis system of claim 7, wherein when creating the selection list, the list unit calculates an index value for each of the plurality of candidate codes that includes a weighted sum of the relevance of the associated target term, the importance of the associated treatment, and the priority of the associated target term, and arranges the plurality of candidate codes so that the higher the index value, the higher the ranking. The failure response record includes a first entry field and a second entry field in which free text can be entered,
The analysis system of claim 7, wherein the priority calculation unit calculates an on-demand priority whose value differs depending on whether the target term extracted by the extraction unit is included in the first description column or the second description column , and calculates the priority of the target term extracted by the extraction unit as a weighted sum of the on-demand priority and a basic priority set for each target term. The analysis system described in claim 9, wherein the display control unit accepts input of sorting conditions after displaying the plurality of candidate codes, and displays the plurality of candidate codes in the selection list after being sorted by the list unit based on the sorting conditions, with higher rankings positioned higher. the display control unit accepts, as the sorting condition, an input of a condition for changing a weighting ratio between the basic priority and the occasional priority;
The analysis system according to claim 10, wherein the list unit calculates a priority of the target term based on a weighting ratio indicated in the sorting conditions, and sorts the plurality of candidate codes based on the calculated priority so that the higher the relevance of the target term to which the plurality of candidate codes are associated, the higher the importance of the associated action, and the higher the associated priority, the higher the ranking. The analysis system described in any one of claims 2 to 11, wherein the display control unit accepts input of sorting conditions, and displays the plurality of candidate codes included in the selection list after removing candidate codes that do not satisfy the sorting conditions from the plurality of candidate codes included in the selection list, with higher rankings positioned higher. a history creation unit that creates a failure response history in which the candidate code selected from the selection list as the judgment code is added to the failure response record;
a correspondence learning unit that calculates a correspondence index value for each of the plurality of candidate codes based on at least one of the number of times the candidate code has been assigned to a plurality of failure response histories previously created by the history creation unit and the number of times the candidate code corresponds to an error code included in the plurality of failure response histories; and
Further provided with
12. The analysis system according to claim 2, wherein, when there are two or more candidate codes in the selection list that have the same ranking determined by information including the relevance and the importance among the plurality of candidate codes, the list unit arranges the two or more candidate codes that have the same ranking so that the higher the correspondence index value, the higher the ranking . The list section
Among the plurality of candidate codes, a plurality of inclusive candidate codes that include the associated target term as a character string are sorted so as to be ranked higher than candidate codes that are not the plurality of inclusive candidate codes, and then
The analysis system according to claim 13 , wherein the plurality of inclusion candidate codes are arranged so that the higher the relevance of the associated target term and the higher the importance of the associated action, the higher the ranking. the target terms extracted by the extraction unit are names of devices that constitute the device;
each of the plurality of candidate codes is a name of a counting device selected for counting;
The analysis system according to claim 2 , wherein the degree of association is an edit distance between the device name and the tabulation device name. A system for analyzing a failure response record including free text entered by a person who responded to a device failure,
an extraction unit that extracts specific target terms and words indicating the content of the measures taken during the failure response from free text included in the failure response record;
a correlation calculation unit that calculates a correlation between each of a plurality of candidate codes that are candidates for a judgment code to be assigned to the failure response record and the target term extracted by the extraction unit;
an importance calculation unit that calculates the importance of the action extracted by the extraction unit and associates the action with one of the plurality of candidate codes;
a list unit that creates a selection list in which the plurality of candidate codes are arranged in a ranking order based on the relevance calculated by the relevance calculation unit and the importance calculated by the importance calculation unit, so that the higher the probability of the candidate code being selected as the judgment code, the higher the ranking; and
an estimation unit that estimates that the candidate code with the highest ranking in the selection list created by the list unit is the judgment code;
An estimation system comprising: