JP7590735B2 - Search label generation system and search system - Google Patents

Description

The present invention relates to a search label generation system that generates labels used when searching for disaster information, which is text related to disasters, and a search system that uses the labels.

It has been proposed to utilize information on disasters at construction sites to ensure the safety of future work. For example, Patent Document 1 describes a system that stores information on disasters that have occurred in the past and displays the information in response to input data for work to be carried out in the future.

Japanese Patent Application Publication No. 6-44211

Traditionally, at construction sites, cases of accidents, such as work-related accidents (accidents), have been compiled into a database. The data for each case is text (sentences), such as the circumstances under which the accident occurred, the details of the accident, and the cause of the accident. It is possible to use this database using a system such as that shown in Patent Document 1. For example, it is possible to input search keywords or text and use them to search for data stored in the database. However, when the amount of data stored in the database is enormous, it is conceivable that the amount of data obtained by a search will also be large. In such cases, the search results that are output are not necessarily convenient for the user, and as a result, there is a risk that they may not be appropriately used to ensure the safety of future work.

The present invention has been made in consideration of the above, and aims to provide a search label generation system and a search system that can appropriately output disaster information related to disasters obtained by search.

In order to achieve the above object, the search label generation system according to the present invention is a search label generation system that generates labels used when searching for disaster information, which is text related to a disaster, and includes a disaster information acquisition means for acquiring multiple pieces of disaster information to be searched, an extraction means for performing dependency analysis on each piece of disaster information acquired by the disaster information acquisition means, and extracting generation text to be used for generating labels from the disaster information using the result of the dependency analysis and pre-stored extraction expressions, and a label generation means for grouping multiple pieces of disaster information based on the generation text of each piece of disaster information extracted by the extraction means, and generating a label corresponding to each group.

In the search label generation system according to the present invention, the results of dependency analysis and pre-stored extraction expressions are used to extract generation text, which is then used to group disaster information. As a result, disaster information is appropriately grouped, and labels corresponding to each group are generated. By using the generated labels, disaster information obtained by search is appropriately output. In other words, the search label generation system according to the present invention can appropriately output disaster information related to disasters obtained by search.

The extraction means may extract the generation text using any one of the extraction expressions included in the initial phrase of the generation text to be extracted, the extraction expressions included in the final phrase of the generation text to be extracted, and the extraction expressions included in the phrases in the middle of the generation text to be extracted, which are stored in advance. With this configuration, the generation text can be extracted appropriately and reliably. As a result, the disaster information related to the disaster obtained by the search can be output appropriately and reliably.

The extraction means uses the divided expressions stored in advance to divide the text for each disaster information acquired by the disaster information acquisition means, performs dependency analysis on one of the divided disaster information, and extracts generation text composed of a phrase including an extraction expression and consecutive phrases having a dependency relationship with the phrase from the divided disaster information that has been subjected to dependency analysis. With this configuration, it is possible to appropriately and reliably extract generation text. As a result, it is possible to appropriately and reliably output disaster information related to the disaster obtained by the search.

The search system according to the present invention is a search system including the above-mentioned search label generation system, and further includes a storage means for storing disaster information of each group in association with the labels generated by the search label generation system, and a search means for inputting search information to be used in a search, searching for disaster information using the input search information, and outputting search results according to the labels associated with the disaster information found . With this configuration, for example, a search can be performed for disaster information related to work to be performed. Furthermore , disaster information related to the disaster obtained by the search can be appropriately output.

The search means inputs search text as search information, performs morphological analysis on the input search text to extract morphemes, determines morphemes to be used in the search based on the frequency of occurrence of the extracted morphemes in disaster information, and searches for disaster information using the determined morphemes . The search means may also change the number of morphemes to be used in the search according to the frequency of occurrence based on the result of the search for disaster information using the morphemes, and search for disaster information again. With these configurations, for example, it is possible to output a number of disaster information pieces that can be appropriately recognized by the user as search results.

According to the present invention, disaster information obtained through a search can be output appropriately.

1 is a diagram showing a configuration of a search label generation system and a disaster information search system which is a search system according to an embodiment of the present invention. 11 is a table showing examples of division expressions used to divide disaster information. 11 is a table showing examples of unnecessary expressions used when dividing disaster information. FIG. 13 is a diagram showing an example of a result of dependency analysis. FIG. 13 is a diagram showing examples of clue expressions used to extract a cause sentence. 11 is a table showing examples of cause sentences to be extracted and cause sentences to be used as cause labels of groups. 1 is a table showing the frequency of occurrence in disaster information counted for each morpheme (word). FIG. 13 is a diagram showing an example of a display used for searching for disaster information. 11 is a table showing an example of search information that has been subjected to morphological analysis. 1 is a flowchart showing a process executed before a search for disaster information in a disaster information search system, which is a search label generation system and a search system according to an embodiment of the present invention. 1 is a flowchart showing a process executed when searching for disaster information in a disaster information search system, which is a search label generation system and a search system according to an embodiment of the present invention.

Below, an embodiment of the search label generation system according to the present invention will be described in detail with reference to the drawings. Note that in the description of the drawings, the same elements are given the same reference numerals and duplicated descriptions will be omitted.

Figure 1 shows a disaster information search system 10, which is a search label generation system and search system according to this embodiment. The disaster information search system 10 is a system (device) that searches for disaster information (disaster data, disaster text), which is text related to disasters. Disaster information is, for example, text (character strings) describing past disasters. The disaster information search system 10 is used, for example, by a safety manager or instructor at a construction company. In this case, the disaster information is information related to past cases of work-related accidents (accidents) at construction sites. Specifically, the disaster information is a sentence (text) that explains, for each disaster case, the work being carried out when the disaster occurred, the cause of the disaster, the details of the disaster, etc. The disaster information is, for example, created by a safety manager at a construction site after the disaster occurs. Conventionally generated disaster information may be used as the disaster information in this embodiment.

The disaster information search system 10 searches for work that is expected to be done on a construction site, and by referencing disaster information related to the same or similar work, it predicts hazards for the work. Note that the disaster information to be searched does not necessarily have to be related to past work-related accidents at construction sites, and may be related to any type of disaster. In addition, the disaster information search system 10 may also be used outside of construction sites depending on the content of the disaster information, etc.

The disaster information search system 10 is also a search label generation system that generates labels to be used when searching for disaster information in advance, that is, by the time the search for disaster information is performed. When the amount of disaster information to be searched is enormous, the amount of disaster information obtained by the search is also likely to be large. In such cases, simply presenting the user with a large amount of disaster information obtained by the search may not necessarily be effective for predicting hazards in work, or may take a long time to refer to. The labels generated by the disaster information search system 10 are intended to appropriately output disaster information related to disasters obtained by the search. The disaster information search system 10 searches for disaster information that corresponds to the generated labels. Labels will be described in more detail below.

The disaster information search system 10 is specifically configured with a server device, which is a computer including hardware such as a CPU (Central Processing Unit) and memory. Each function of the disaster information search system 10, which will be described later, is realized by these components operating through programs or the like. The disaster information search system 10 may be realized with a single computer, or may be realized with a computer system configured by multiple computers connected to each other via a network.

The disaster information search system 10 can send and receive information between the terminal 20 and the disaster information database 30 via a network such as the Internet. The terminal 20 is used by a user such as a safety manager or instructor who performs searches using the disaster information search system 10. The terminal 20 makes a search request to the disaster information search system 10 and receives the search results from the disaster information search system 10. The terminal 20 is a computer such as a PC (personal computer), a tablet terminal, or a smartphone.

The disaster information database 30 is a database that stores disaster information to be searched by the disaster information search system 10. For example, the disaster information database 30 is managed by a construction company or a public institution. There may be multiple disaster information databases 30 for each managed entity (for example, the above-mentioned construction company or public institution). The disaster information database 30 is a computer such as a server device. The disaster information database 30 (functions thereof) may be included in the disaster information search system 10.

Next, the functions of the disaster information search system 10 according to this embodiment will be described. As shown in FIG. 1, the disaster information search system 10 includes a disaster information acquisition unit 11, an extraction unit 12, a label generation unit 13, and a search unit 14. The disaster information acquisition unit 11, the extraction unit 12, and the label generation unit 13 correspond to a search label generation system. The search unit 14 corresponds to a search system.

The disaster information acquisition unit 11 is a disaster information acquisition means that acquires multiple pieces of disaster information to be searched. For example, the disaster information acquisition unit 11 acquires disaster information from the disaster information database 30. If the disaster information database 30 publishes disaster information on the Web, the disaster information acquisition unit 11 may acquire the disaster information by performing Web scraping. Alternatively, disaster information acquired from the disaster information database 30 may be stored in advance in the disaster information search system 10, and the disaster information acquisition unit 11 may acquire the stored disaster information. Furthermore, the disaster information acquisition unit 11 may acquire disaster information by a method other than the above.

The disaster information acquired by the disaster information acquisition unit 11 is made identifiable for each piece of disaster information. Furthermore, when disaster information is acquired from a plurality of disaster information databases 30, it may be possible to identify which disaster information database 30 the information was acquired from. Furthermore, metadata relating to the disaster information, such as the date of occurrence of the disaster, the type of disaster, the case title, and damage caused by the disaster (e.g., death or the period required for treatment), may be associated with each piece of disaster information. Furthermore, the disaster information and the associated metadata may be acquired, for example, as CSV (Comma-Separated Values) data, or may be data in another format. The disaster information acquisition unit 11 outputs the acquired disaster information to the extraction unit 12.

The extraction unit 12 is an extraction means that performs dependency analysis on each piece of disaster information acquired by the disaster information acquisition unit 11, and extracts generation text to be used for generating labels from the disaster information using the result of the dependency analysis and pre-stored extraction expressions. The extraction unit 12 may extract generation text using any of the pre-stored extraction expressions corresponding to the beginning of the generation text to be extracted, the extraction expressions corresponding to the end of the generation text to be extracted, and the extraction expressions corresponding to the middle of the generation text to be extracted. The extraction unit 12 may use the pre-stored division expressions to divide text for each piece of disaster information acquired by the disaster information acquisition unit 11, and perform dependency analysis on any of the divided disaster information.

The extraction unit 12 extracts, from the disaster information, a cause sentence indicating the cause of the disaster related to the disaster information as a generation text. As described above, disaster information usually includes a portion indicating the cause of the disaster. The extraction unit 12 extracts that portion as a cause sentence.

For example, the extraction unit 12 extracts a cause sentence, which is a text for generation, as follows. The extraction unit 12 inputs disaster information from the disaster information acquisition unit 11. Disaster information usually includes, in the first half, a portion indicating the work being performed when the disaster occurred, and in the second half, a portion indicating the disaster situation. For example, if the disaster information is a text of "While assembling scaffolding, the worker lost his footing and fell 3 meters below" (Disaster Information Example 1), the first half of the text, "While assembling scaffolding," indicates the work content, and the second half of the text, "The worker lost his footing and fell 3 meters below," indicates the disaster situation. If the disaster information is a text of "During tunnel construction, a rockfall occurred while excavating the tunnel" (Disaster Information Example 2), the first half of the text, "During tunnel construction, a rockfall occurred while excavating the tunnel," indicates the work content, and the second half of the text, "The rockfall occurred while the worker was excavating the tunnel," indicates the disaster situation. The portion indicating the cause of the disaster is included in the portion indicating the disaster situation.

First, the extraction unit 12 divides each piece of input disaster information into a part indicating the work content and a part indicating the disaster situation. The extraction unit 12 pre-stores, as a division expression, a division expression that appears at the boundary between the part indicating the work content and the part indicating the disaster situation. A division expression is an expression that appears at the end of the part indicating the work content, i.e., the first half to be divided. For example, the extraction unit 12 pre-stores, as a division dictionary, a number of division expressions as shown in FIG. 2.

The extraction unit 12 compares the input disaster information with the division expressions stored in advance to determine the division expressions contained in the disaster information. The extraction unit 12 divides the disaster information into two parts, with the first half being the part up to the division expression contained in the disaster information, and the second half being the part thereafter. Note that a priority may be set in advance for the division expressions, and if the disaster information contains multiple division expressions, the division may be performed using the division expressions according to the priority.

The extraction unit 12 regards the first half of the divided part as a part indicating the work content, and the second half of the divided part as a part indicating the disaster situation. The extraction unit 12 may also delete unnecessary expressions from the second half of the divided part as follows to determine the part indicating the disaster situation. Unnecessary expressions are expressions that are inappropriate for the beginning of a sentence, such as conjunctions and punctuation marks. The extraction unit 12 stores a plurality of unnecessary expressions as shown in FIG. 3 in advance as a dictionary. The extraction unit 12 compares the second half of the divided part with the unnecessary expressions stored in advance to determine whether the beginning of the second half part contains an unnecessary expression. If the beginning of the second half part contains an unnecessary expression, the extraction unit 12 deletes the unnecessary expression. The extraction unit 12 repeatedly deletes the unnecessary expressions until no unnecessary expressions are included at the beginning. The extraction unit 12 regards the above-mentioned second half part that does not contain an unnecessary expression at the beginning as the part indicating the disaster situation.

In the above disaster information example 1, the latter part of the segment becomes "However, he lost his footing and fell 3 meters down," and the part showing the disaster situation with the unnecessary expression "However," deleted becomes "He lost his footing and fell 3 meters down." In disaster information example 2, the latter part of the segment becomes "A rock fell into and he was caught in the mudslide," and the part showing the disaster situation with the unnecessary expression "to" deleted becomes "A rock fell into and he was caught in the mudslide."

The extraction unit 12 performs dependency analysis on the portion of the disaster information obtained as described above that indicates the disaster situation. The dependency analysis can be performed by a conventional method such as CaboCha or KNP (Kurohashi Nagao Parser). By performing dependency analysis, the extraction unit 12 obtains information indicating the phrases constituting the portion of the disaster information that indicates the disaster situation and the dependency relationship (modifier, modified) of the phrases, as shown in FIG. 4. The example shown in FIG. 4 is an example of a case where the disaster information is a text that reads, "While preparing the foundation for the exterior wall, I got off the stepladder without checking carefully what was below me, and so I sprained and broke my left foot on the step." (Disaster Information Example 3), and the latter half obtained is, "When I got off the stepladder, I sprained and broke my left foot on the step."

The extraction unit 12 stores clue expressions for extracting a cause sentence as extraction expressions in advance. Clue expressions are expressions that characteristically appear in a cause sentence. For example, the extraction unit 12 stores a plurality of clue expressions as shown in FIG. 5 as a dictionary in advance. A type indicating which part of a cause sentence the clue expression appears in is set in advance for the clue expression. The type of clue expression is one of three types: one that corresponds to (appears at) the beginning of the cause sentence to be extracted (type A: beginning of sentence type shown in FIG. 5), one that corresponds to (appears at) the end of the cause sentence to be extracted (type B: end of sentence type shown in FIG. 5), and one that corresponds to (appears at) the middle of the cause sentence to be extracted, i.e., a part that is not the beginning or end of the sentence (type C: middle of sentence type shown in FIG. 5).

The extraction unit 12 compares the portion indicating the disaster situation of each piece of obtained disaster information with pre-stored clue expressions to determine the clue expressions contained in the portion. Priorities may be set in advance for the clue expressions, and the extraction unit 12 may perform the above determination in order of priority until a clue expression contained in the portion is found. For example, the above determination may be performed in order from top to bottom on the left side of the clue expressions shown in FIG. 5, and then from top to bottom on the right side.

Based on the result of dependency analysis of the part indicating the disaster situation, the extraction unit 12 extracts a cause sentence using a clue expression included in the part as a clue. The extraction unit 12 prestores extraction rules for cause sentences for each type of clue expression. The extraction unit 12 extracts a cause sentence from the part indicating the disaster situation based on an extraction rule according to the clue expression included in the part.

First, the extraction unit 12 determines a phrase including a clue expression as a phrase that constitutes a cause sentence. If the clue expression included in a phrase is of the sentence-initial type (type A), the extraction unit 12 determines consecutive phrases that have a modification relationship (dependency relationship) following the phrase including the clue expression as a phrase that constitutes a cause sentence. For example, in the example shown in FIG. 4, if the phrase including the clue expression is phrase number 9, phrases number 9, 11, and 12 are determined as phrases that constitute a cause sentence. In this case, the extraction unit 12 determines a phrase that has a modification relationship (dependency relationship) going back from the last phrase among the consecutive phrases, has no modification relationship going back further (i.e., is an independent phrase), and appears before the phrase including the clue expression as a phrase that constitutes a cause sentence. For example, in the example shown in Figure 4, if the phrase containing the clue expression is phrase 9, and there is a phrase that can be found one step back from phrase 12, the last phrase above, then that phrase will be considered to be a phrase that constitutes the cause sentence. There is no such phrase in Figure 4, but if phrase 8 does not have a dependency relationship with phrase 7, then phrase 8 will be considered to be a phrase that constitutes the cause sentence.

If the clue expression included in a phrase is of the sentence-end type (type B), the extraction unit 12 regards consecutive phrases having a modification relationship (dependency relationship) going back from the phrase including the clue expression as phrases that constitute the cause sentence. In this case, if there are multiple phrases that can be traced back (i.e., when the modification relationship branches out), one phrase that is closest to the beginning of the part that indicates the disaster situation is regarded as a phrase that constitutes the cause sentence. Note that if one phrase can be traced back further, the further back phrases are also regarded as phrases that constitute the cause sentence. For example, in the example shown in FIG. 4, if the phrase including the clue expression is phrase 6, phrases 4 and 6 are regarded as phrases that constitute the cause sentence.

When the clue expression included in the phrase is of the mid-sentence type (Type C), the extraction unit 12 determines the phrases that make up the causal sentence based on both the extraction rules for the beginning-of-sentence type (Type A) and the end-of-sentence type (Type B). For example, in the example shown in FIG. 4, when the phrase that includes the clue expression is phrase 6, phrases 4, 6, 7, 8, and 12 are determined to be the phrases that make up the causal sentence.

The extraction unit 12 arranges the phrases that are determined to be the phrases that make up the cause sentence in the order in which they appear in the part indicating the disaster situation (in the example shown in FIG. 4, from smallest to largest numbers), and regards the arrangement as the cause sentence. For example, in the example of FIG. 4, the clue expression included in the part indicating the disaster situation is "do not" which is a sentence-final type (type B). Based on the above extraction rules, the cause sentence in this case is "without checking below" (phrases 4 and 6). The extraction unit 12 outputs the cause sentences extracted for each disaster information to the label generation unit 13.

The method of extracting a cause sentence, including the above-mentioned rules for extracting a cause sentence, was discovered by the inventor of the present invention through review and trial and error of existing disaster information, as a method capable of extracting a better cause sentence. However, the extraction of a cause sentence may be performed by a method other than the above.

The label generation unit 13 is a label generation means that groups multiple pieces of disaster information based on the generation text (cause sentence) of each piece of disaster information extracted by the extraction unit 12, and generates a label corresponding to each group. The grouping of disaster information by the label generation unit 13 is intended to make the output of search results for disaster information more convenient for users. The grouping of disaster information is performed based on the cause sentence, i.e., based on the cause of the disaster. Therefore, for example, by outputting the search results for disaster information by group, it is possible to output an output that allows the user to grasp the tendency of the cause of the disaster. Furthermore, as described below, the label may be a cause label that indicates the cause related to the group. For example, the cause label may be one that allows the user to easily grasp the cause of the disaster related to the disaster information based on the cause sentence.

For example, the label generation unit 13 groups the disaster information as follows, and generates a label corresponding to each group. The label generation unit 13 inputs the cause sentence for each disaster information from the extraction unit 12. The label generation unit 13 converts each cause sentence into a feature indicating the characteristics of the sentence, for example, a vector (cause sentence vector) with a preset number of dimensions. The conversion to a cause sentence vector can be performed by a conventional method, for example, using a general-purpose natural language processing model such as Pretrained BERT.

The label generation unit 13 clusters the cause sentences using the cause sentence vector. That is, the label generation unit 13 groups the disaster information. The clusters generated by the clustering correspond to groups of disaster information. Clustering using the cause sentence vector can be performed by a conventional method, for example, unsupervised clustering such as DBSCAN or k-means. A machine learning technique may also be used for the clustering. As shown in FIG. 6(a), the label generation unit 13 associates a group number, which is information indicating the group to which the cause sentence belongs, with each cause sentence (each disaster information). Note that the grouping of disaster information may be performed by a method other than the above.

The label generation unit 13 generates a cause label for each group (cluster) generated. For example, the label generation unit 13 sets the cause sentence corresponding to the cause sentence vector closest to the center of gravity of the cause sentence vectors included in the cluster as the cause label. Alternatively, the cause label may be generated by other methods. As shown in FIG. 6(b), the label generation unit 13 associates the cause sentence set as the cause label with a group number for each group. The label generation unit 13 associates the generated cause label with the disaster information included in the group related to the cause label, stores it in the disaster information search system 10, and makes it available to the search unit 14.

The search unit 14 is a search means that inputs search information to be used in the search, searches for disaster information using the input search information, and outputs search results according to the labels generated by the label generation unit 13. The search unit 14 may input search text as search information, perform morphological analysis on the input search text to extract morphemes, and search for disaster information using the extracted morphemes according to the frequency of occurrence of the morphemes in the disaster information. The search unit 14 may search for disaster information by changing the number of morphemes to be used in the search according to the frequency of occurrence based on the results of the search for disaster information using the morphemes.

The search unit 14 calculates the frequency of occurrence of morphemes in the disaster information in advance, that is, by the time the search for disaster information is performed. The search unit 14 acquires the disaster information to be searched. The acquisition of disaster information may be performed in the same manner as acquisition by the disaster information acquisition unit 11. Alternatively, the search unit 14 may input the disaster information acquired by the disaster information acquisition unit 11 from the disaster information acquisition unit 11.

The search unit 14 performs morpheme analysis on each of the acquired disaster information to acquire morphemes from the disaster information. The search unit 14 counts the frequency of occurrence in the disaster information for each morpheme. At this time, the frequency of occurrence is not counted for morphemes whose parts of speech obtained by the morpheme analysis are inappropriate for use in searching for disaster information (e.g., symbols, particles, numbers, and numerical particles). Alternatively, the frequency of occurrence may be counted only for specific morphemes (e.g., nouns) that are set in advance and are appropriate for use in searching for disaster information. The number to be counted may be the number of disaster information items that appear (1 is counted even if multiple items appear in one disaster information item), or the number of morphemes (the number is counted if multiple items appear in one disaster information item). The search unit 14 stores the number counted for each morpheme as shown in FIG. 7. The above is the function of the search unit 14 until the search is performed.

For example, the search unit 14 searches for disaster information as follows. The search by the search unit 14 is performed, for example, via a display on the terminal 20, which is a user interface shown in FIG. 8. The display is performed, for example, by the terminal 20 accessing the disaster information search system 10. When search information is entered in the work content input field on the display and a search button is pressed, the search unit 14 of the disaster information search system 10 uses the entered search information to search for disaster information. The search results are displayed in the section following the analysis result. The search information is, for example, text (character string, keyword). By entering text related to work to be performed at the construction site as search information, disaster information related to the work is searched for. Note that a predetermined API (Application Programming Interface) may be provided for the above search function by the search unit 14.

In addition, if there are multiple types of disaster information databases 30, it may be possible to search for each disaster information database 30. For example, in the display shown in FIG. 8, it may be possible to select the disaster information database 30 to be searched in the target data selection field.

The search results are displayed using the cause labels generated by the label generation unit 13 and associated with each piece of disaster information. For example, a display like the pie chart portion in the display shown in FIG. 8 may be displayed. For example, among the disaster information hit by the search, those associated with the same cause label are displayed as one pie chart. The example shown in FIG. 8 shows that the search hit 16 pieces of disaster information associated with the cause label "scaffolding board came off." The pie chart shows, for example, the number of disaster information items by damage caused by the disaster (for example, death, if treatment was required, and how many months of treatment were required). Note that the graph may be of a type other than a pie chart, as shown in FIG. 8.

In addition, disaster information found in the search is displayed in the case list column. For example, in the display shown in FIG. 8, DB type, disaster information, and cause statement are displayed in a table format. DB type is information indicating the disaster information database 30 that stores disaster information. Disaster information is disaster information found in the search. Cause statement is a cause statement extracted from the disaster information by the extraction unit 12. Other metadata associated with the disaster information may also be displayed. When a pie chart is selected by the user, disaster information corresponding to the pie chart may be displayed. That is, disaster information associated with the same cause label may be displayed together.

The search unit 14 receives and inputs search information to be used in a search as a search request from the terminal 20. Input of search information in the terminal 20 and transmission to the disaster information search system 10 are performed by the user operating the terminal 20 using the display shown in FIG. 8 and the like. The input search information is search text, which is a sentence. The search text indicates the work that will be performed at the construction site, such as "assembling formwork on the scaffolding."

The search unit 14 performs morphological analysis on the input search text to extract morphemes. FIG. 9(a) shows an example of search text from which morpheme analysis has been performed and morphemes (words) have been extracted. Note that EOS (End Of Sentence) in FIG. 9 indicates the end of the search text. Next, the search unit 14 excludes morphemes whose parts of speech obtained by the morphological analysis are inappropriate for use in searches for disaster information (e.g., symbols, particles, numbers, and numerical particles). Alternatively, only certain morphemes (e.g., nouns) that are preset as appropriate for use in searches for disaster information may be used in searches for disaster information.

An example of morphemes (words) excluded is shown in FIG. 9(b). The morphemes with a horizontal line are excluded morphemes. Furthermore, the search unit 14 may exclude, from the obtained morphemes, those whose frequency of occurrence in disaster information meets a certain criterion. For example, the search unit 14 may refer to the frequency of occurrence (number of occurrences) of each morpheme that has been calculated and stored in advance, and exclude the morphemes that appear in the top 5% of frequency in disaster information. This is because morphemes that appear frequently in disaster information and are considered to be common are not necessarily suitable for searching disaster information.

The search unit 14 searches for disaster information using the morphemes obtained as described above. The search unit 14 may store disaster information obtained from the disaster information database 30 in advance and perform a search on the stored disaster information, or may perform a search on the disaster information database 30. The search unit 14 searches for disaster information that contains all of the obtained morphemes in the text (partial match search). That is, the search unit 14 performs an AND search using the obtained morphemes as search keywords.

The search unit 14 obtains disaster information that contains all of the morphemes used in the search in the text, i.e., disaster information that is hit by the search. Based on the search result, the search unit 14 determines whether to output the search result or to perform another search. The search unit 14 compares the number of hit disaster information items with a preset threshold. If it is determined that the number of hit disaster information items is equal to or greater than the threshold, the search unit 14 determines to output the search result. If it is determined that the number of hit disaster information items is less than the threshold, the search unit 14 determines to perform another search. This is because if the number of hit disaster information items is small, the user cannot refer to a sufficient amount of disaster information, which is not necessarily appropriate from the standpoint of work hazard prediction, etc.

When it is determined that a search should be performed again, the search unit 14 refers to the frequency of occurrence (number of occurrences) of each morpheme that has been calculated and stored in advance, excludes the morpheme that appears most frequently from among the morphemes used in the search, and performs the search again. For example, if a search is performed again after a search is performed using the morphemes (words) shown in FIG. 9(b), the morpheme "work" that appears most frequently (number of occurrences) as shown in FIG. 9(c) is excluded from the morphemes to be used in the next search. The search unit 14 repeats the exclusion of morphemes and the search until the number of hit disaster information items reaches or exceeds a threshold value.

When it is determined that the search results should be output, the search unit 14 outputs the disaster information hit. In doing so, the search unit 14 references the cause label of each piece of disaster information generated and stored by the label generation unit 13, and generates and outputs information indicating the search results according to the disaster label as shown in FIG. 8. Specifically, the search unit 14 transmits the generated information indicating the search results to the terminal 20. The terminal 20 receives and displays the information, or outputs it in a format that can be recognized by the user.

The search unit 14 may perform a search in a manner other than the above. In the above example, the number of morphemes used in the search is reduced to increase the number of hits for disaster information, but the number of morphemes used in the search may be increased to decrease the number of hits for disaster information. In addition, the search unit 14 may perform a search using, for example, the input keywords as is, instead of a search using morphological analysis of the search text as described above.

The input and output of information by the search unit 14 does not necessarily have to be performed with the terminal 20 as described above, but may be performed in any manner with any device. The above are the functions of the disaster information search system 10 according to this embodiment.

Next, the process executed by the disaster information search system 10 according to this embodiment will be described with reference to the flowcharts in FIG. 10 and FIG. 11. First, the process executed before searching for disaster information will be described with reference to FIG. 10.

In this process, first, the disaster information acquisition unit 11 acquires multiple pieces of disaster information to be searched (S01). Next, the extraction unit 12 uses a division expression to divide the disaster information into a first half that indicates the work content and a second half that indicates the disaster situation (S02). Next, the extraction unit 12 performs dependency analysis on the part that indicates the disaster situation (S03). Next, the extraction unit 12 uses the result of the dependency analysis and clue expressions to extract a cause sentence from the disaster information (S04).

Then, the label generation unit 13 performs clustering based on the cause sentences to group the disaster information (S05). The label generation unit 13 then generates a cause label for each group (S06). The generated cause label is associated with the disaster information of each group, stored in the disaster information search system 10, and used by the search unit 14.

The search unit 14 also acquires multiple pieces of disaster information to be searched, performs morphological analysis, and counts the frequency of occurrence of each morpheme in the disaster information (S07). The frequency of occurrence of each morpheme is stored in the disaster information search system 10 and used by the search unit 14. Note that the processes in S01 to S06 and S07 are performed independently of each other, and therefore do not necessarily have to be performed in the above order. The above is the process that is executed before searching for disaster information.

Next, the process executed when searching for disaster information will be described with reference to FIG. 11. In this process, first, search text is input by the search unit 14 (S11). For example, search text transmitted from the terminal 20 is received and input. Next, the search unit 14 performs morphological analysis on the search text to extract morphemes (S12). At this time, morphemes that are not used in the search may be excluded from the extracted morphemes. Next, the search unit 14 searches for disaster information using the morphemes (S13). Next, the search unit 14 determines whether the number of disaster information hits by the search is equal to or greater than a threshold (S14).

If it is determined that the number of hit disaster information items is less than the threshold value (NO in S14), the search unit 14 then removes morphemes used in the search according to the frequency of occurrence of the morphemes in the disaster information (S15). After that, the search unit 14 again searches for disaster information using the morphemes (S13), and thereafter, the same processing as above (processing from S14 onwards) is performed.

If it is determined in S14 that the number of hit disaster information items is equal to or greater than the threshold value (YES in S14), then the search unit 14 generates and outputs search results according to the cause labels associated with the hit disaster information items (S16). For example, information indicating the search results as shown in FIG. 8 is generated and transmitted to the terminal 20. The information is displayed on the terminal 20, allowing the user to refer to the disaster information. The above is the process executed when searching for disaster information.

In this embodiment, the result of dependency analysis and clue expressions, which are pre-stored expressions for extraction, are used to extract cause sentences, which are generation texts, and these are used to group disaster information. As a result, disaster information is appropriately grouped, and cause labels corresponding to each group are generated. By using the generated cause labels, disaster information obtained by search is appropriately output. For example, as described above, search results are output in a state where they are grouped according to the cause of the disaster. By performing such output, the user can appropriately grasp disaster information, which can be effectively used for predicting hazards in work, etc. In this way, according to this embodiment, disaster information related to disasters obtained by search can be appropriately output.

Also, as in the above-described embodiment, the cause sentence may be extracted using three types of clue expressions according to the position in the sentence in which they appear. With this configuration, the cause sentence can be extracted appropriately and reliably. As a result, disaster information related to the disaster obtained by the search can be output appropriately and reliably. However, the clue expressions do not have to be of the three types described above, and any type may be used as long as it extracts the cause sentence using the results of dependency analysis.

Also, as in the above-described embodiment, dependency analysis may be performed by dividing disaster information using a divided expression. As described above, the cause sentence is usually included in the latter part of the disaster information that indicates the disaster situation. Therefore, with this configuration, the cause sentence can be extracted appropriately and reliably. As a result, disaster information related to the disaster obtained by search can be output appropriately and reliably. Note that in this embodiment, dependency analysis is performed on the divided latter part, but dependency analysis may be performed on the first part depending on the disaster information to be searched and the generation text to be extracted. Also, it is not necessarily necessary to divide the disaster information, and dependency analysis may be performed on the disaster information itself to extract the generation text.

As in the embodiment described above, the disaster information search system 10 may be configured to include a search unit 14, i.e., may have the functionality of a search system. With this configuration, as described above, for example, it is possible to search for disaster information related to work to be performed. Furthermore, it is possible to appropriately output disaster information related to the disaster obtained by the search.

Also, as in this embodiment, morphological analysis may be performed on the search text to extract morphemes, and the extracted morphemes may be used to search for disaster information depending on the frequency of occurrence of the morphemes in the disaster information. Furthermore, the number of morphemes used in the search may be changed depending on the frequency of occurrence, and a search for disaster information may be performed again. With these configurations, for example, a number of pieces of disaster information that the user can properly recognize may be output as search results. However, the frequency of occurrence of morphemes does not necessarily need to be used in the search, and morphological analysis does not necessarily need to be performed.

In this embodiment, the disaster information search system 10 has the functions of both a search label generation system (disaster information acquisition unit 11, extraction unit 12, and label generation unit 13) and a search system (search unit 14). However, the search label generation system and the search system do not necessarily need to be included in one system, and may be separate systems. In this case, the search system acquires labels for each group of disaster information, such as cause labels, generated by a search label generation system, which is a separate system, and uses them to output search results.

In addition, in this embodiment, the generation text extracted by the extraction unit 12 is a cause sentence indicating the cause of the disaster, but it does not necessarily have to be a cause sentence, as long as it can appropriately group disaster information. Similarly, the labels generated by the label generation unit do not necessarily have to be cause labels, as long as they are labels that can be used to output appropriate search results.

10... disaster information search system, 11... disaster information acquisition unit, 12... extraction unit, 13... label generation unit, 14... search unit, 20... terminal, 30... disaster information database.

Claims (4)

A search label generation system that generates labels used when searching for disaster information, which is text related to a disaster, comprising:
A disaster information acquisition means for acquiring multiple pieces of disaster information to be searched;
an extraction means for performing dependency analysis on each of the disaster information acquired by the disaster information acquisition means, and extracting generation text to be used for generating a label from the disaster information using a result of the dependency analysis and an extraction expression stored in advance;
a label generating means for grouping a plurality of disaster information items based on the generation text of each disaster information item extracted by the extraction means, and generating a label corresponding to each group;
Equipped with
The extraction means uses pre-stored divided expressions to divide text for each piece of disaster information acquired by the disaster information acquisition means, performs the dependency analysis on any one of the divided disaster information, and extracts the generation text from the divided disaster information after the dependency analysis, the generation text being composed of a phrase including the expression for extraction and consecutive phrases having a dependency relationship with the phrase . 2. The search label generation system according to claim 1, wherein the extraction means extracts the generation text using any one of extraction expressions contained in an initial phrase of the generation text to be extracted, extraction expressions contained in an initial phrase of the generation text to be extracted, and extraction expressions contained in an intermediate phrase of the generation text to be extracted, which are stored in advance. A search system comprising the search label generation system according to claim 1 or 2 ,
a storage means for storing disaster information for each group and a label generated by the search label generation system in association with each other;
A search unit is further provided for inputting search information to be used in a search, searching for disaster information using the input search information, and outputting a search result according to a label associated with the disaster information found ,
The search means is a search system that inputs search text as the search information, performs morphological analysis on the input search text to extract morphemes, determines morphemes to be used in the search based on the frequency of appearance of the extracted morphemes in disaster information, and searches for disaster information using the determined morphemes . 4. The search system according to claim 3 , wherein the search means changes the number of morphemes used in the search in accordance with the occurrence frequency based on a result of the search for disaster information using the morphemes, and searches for disaster information again.

Images