JP7583082B2 - Natural language processing system, natural language processing method, and natural language processing program - Google Patents

Description

The present invention relates to a natural language processing system, a natural language processing method, and a natural language processing program, and in particular to a technology for estimating information related to address information based on the address information.

Conventionally, technology has been developed that uses address information and a trained model to estimate information related to the address information. For example, Patent Literature 1 discloses a technology that uses address information and a trained model to determine whether the address information is false address information.

Special Publication No. 2020-502673

According to the technology disclosed in Patent Document 1, by determining whether the address information is false or not, the reliability of the address information is ensured, and the safety of the service execution of the logistics service can be improved. Meanwhile, in the case of logistics services, information on the building type (e.g., a building, a house, etc.) of the building located at the address indicated by the address information can also contribute to the safety of the service execution. In addition, not limited to logistics services, for example, when multiple users living in buildings of the same building type subscribe to or cancel any service at the same time, by understanding the building type, the relationship between the user behavior and the building type can be analyzed, which can be useful for further service development. However, conventionally, there was no technology for estimating the building type located at the address indicated by the address information using a trained model from the address information.

The present invention was made in consideration of the above problems, and aims to estimate the building type of a building located at an address using a learning model.

In order to solve the above problem, one aspect of the natural language processing system according to the present invention includes an acquisition unit that acquires address information, and an estimation unit that estimates the building type of a building located at an address indicated by the address information by inputting the address information into a learning model that has been trained by machine learning.

In order to solve the above problem, one aspect of the natural language processing method according to the present invention includes acquiring address information, and inputting the address information into a learning model that has been trained by machine learning, thereby estimating the building type of the building located at the address indicated by the address information.

In order to solve the above problem, one aspect of the natural language processing program according to the present invention is a natural language processing program for causing a computer to execute natural language processing, the program causing the computer to execute processes including an acquisition process for acquiring address information, and an estimation process for estimating the building type of a building located at an address indicated by the address information by inputting the address information into a learning model that has been trained by machine learning.

According to the present invention, it is possible to estimate the building type of a building located at an address using a learning model.
The above-mentioned objects, aspects, and advantages of the present invention, as well as objects, aspects, and advantages of the present invention not described above, will be understood by those skilled in the art from the following detailed description of the invention by referring to the accompanying drawings and the claims.

FIG. 1 shows an example of the configuration of a natural language processing system according to an embodiment. FIG. 2 shows an example of a hardware configuration of an information processing device according to an embodiment. FIG. 3A shows an example of an estimation procedure using a building type estimation model. FIG. 3B shows another example of an estimation procedure using a building type estimation model. FIG. 3C shows another example of an estimation procedure using a building type estimation model. FIG. 4 shows a specific example of estimating a building type from address information. FIG. 5 shows examples of building types identified by address information. FIG. 6 shows a flowchart of the overall process executed by the natural language processing system. FIG. 7A shows a flowchart of a first example of a building type determination process. FIG. 7B shows a flowchart of a second example of the building type determination process. FIG. 7C shows a flowchart of a third example of the building type determination process.

Below, an embodiment for carrying out the present invention will be described in detail with reference to the attached drawings. Among the components disclosed below, those having the same functions are given the same reference numerals, and their description will be omitted. Note that the embodiment disclosed below is one example of a means for realizing the present invention, and should be appropriately modified or changed depending on the configuration of the device to which the present invention is applied and various conditions, and the present invention is not limited to the following embodiment. Furthermore, not all of the combinations of features described in this embodiment are necessarily essential to the solution of the present invention.

[Configuration of natural language processing system]
1 shows an example of the configuration of a natural language processing (NLP) system 10 according to this embodiment. The natural language processing system 10 shown in FIG. 1 includes an address information acquisition unit 100, a first processing unit 110, a second processing unit 120, a learning processing unit 130, and a building type determination unit 140.

The natural language processing system 10 is configured using one or more information processing devices. The address information acquisition unit 100, first processing unit 110, second processing unit 120, learning processing unit 130, and building type determination unit 140 shown in FIG. 1 may be configured in one information processing device, or may be distributed across multiple information processing devices. The information processing device may be a device such as a desktop PC (Personal Computer), a notebook PC, a tablet, or a general-purpose machine. In addition, all or part of the information processing device may be realized, for example, in a virtual server constructed on a cloud server.

The address information acquisition unit 100 accepts input information including an address and acquires address information. For example, the address information acquisition unit 100 accepts input information from an operator (user) and acquires address information. Alternatively, the address information acquisition unit 100 may accept input information that has been set in advance in the natural language processing system 10 according to a predetermined program and acquire address information.

The address information may indicate a delivery address corresponding to product information that is the subject of transaction processing in an electronic commerce system (electronic commerce platform) that is composed of information processing devices that can communicate with the information processing devices that constitute the natural language processing system 10. In this case, the address information acquisition unit 100 may acquire address information from the electronic commerce system instead of accepting input information including an address from a user, and may accept input of address information by the electronic commerce system. In addition, the address information acquisition unit 100 may acquire address information corresponding to the product on the condition that a transaction is concluded in the electronic commerce system.

In this embodiment, the address information acquisition unit 100 is configured to acquire address information composed of text information (character strings). Therefore, when the accepted input information is composed of text information representing an address, the address information acquisition unit 100 acquires the accepted information itself as address information. When the accepted input information is composed of text information representing an address and text information other than the address or image information, the address information acquisition unit 100 may extract the text information (text portion) representing the address from the accepted input information to acquire the address information. When the accepted input information is voice information, the address information acquisition unit 100 converts the accepted voice information into text information (i.e., transcribes it) and acquires the converted information as address information. The conversion from voice information to text information may be performed using a known voice recognition function or the like.

The address information acquisition unit 100 outputs the acquired address information to the first processing unit 110 and the second processing unit 120. The first processing unit 110 is configured to estimate the building type of the building located at the address indicated by the address information input from the address information acquisition unit 100, based on the learning model. Meanwhile, the second processing unit 120 is configured to identify the building type of the building located at the address indicated by the address information, based on the address information input from the address information acquisition unit 100, on a rule basis.

The first processing unit 110 is a processing unit that estimates the building type of a building located at the address indicated by the address information input from the address information acquisition unit 100, based on the address information and the learning model. The first processing unit 110 is configured to include a building type estimation unit 111, a location information acquisition unit 112, an area information acquisition unit 113, and a learning model storage unit 114. The learning model storage unit 114 stores a building type estimation model 115. The building type estimation model 115 is a learning model for machine learning.

The building type estimation unit 111 uses the building type estimation model 115 to estimate the building type of the building located at the address indicated by the address information input by the address information acquisition unit 100. The building type estimation model 115 is a learning model configured to estimate (predict) and output the building type of the building located at the address indicated by the address information input by the address information acquisition unit 100, based on the address information. In this embodiment, the building types are assumed to be three types: "business building", "unit building", and "house". "Business building" is a building whose main purpose is as an office, and includes business offices and company buildings. "Unit building" is a building whose main purpose is as a residence for multiple households, and includes buildings, condominiums, and apartments (residences) with multiple residential sections (individually occupied spaces). "House" is a building whose main purpose is as a residence for a single household, and includes individual homes and detached houses.

The location information acquisition unit 112 acquires location information of the address indicated by the address information input by the address information acquisition unit 100. For example, the location information acquisition unit 112 may use an internet map service or location information service to acquire the latitude and longitude corresponding to the address as location information. Alternatively, the location information acquisition unit 112 may acquire the coordinates corresponding to the address in a preset two-dimensional or three-dimensional space as location information. The location information may be numerical information.

The area information acquisition unit 113 acquires, from the address information input by the address information acquisition unit 100, information on the area of the address indicated by the address information as area information. For example, the area information acquisition unit 113 acquires at least one of the zoning to which the address belongs, the nearest station to the address, and the distance from the nearest station to the address as area information. The zoning means an area that specifies the use of buildings that can be constructed, and indicates, for example, whether the area of the address is a residential zoning, a commercial zoning, or an industrial zoning. The area information acquisition unit 113 can acquire the zoning by searching a specified database using the address. The area information acquisition unit 113 can also acquire the nearest station to the address and the distance from the nearest station to the address using a map service or a location information service on the Internet. The nearest station is a station that satisfies the condition that it is within a specified distance from the address. If there are multiple stations that satisfy the condition, i.e., if there are multiple nearest stations, the area information acquisition unit 113 may acquire the multiple nearest stations and the distance from each of the nearest stations to the address as area information.

The building type estimation unit 111 may be configured to estimate the building type using at least one of the location information acquired by the location information acquisition unit 112 and the area information acquired by the area information acquisition unit 113 in addition to the address information. The building type estimation model 115 and the estimation procedure using the building type estimation model 115 will be described later. The building type estimation unit 111 outputs the estimated building type to the building type determination unit 140.

The second processing unit 120 is a processing unit that identifies, based on rules, the building type of the building located at the address indicated by the address information input by the address information acquisition unit 100. The second processing unit 120 includes a building type identification unit 121 and a rule memory unit 122. The rule memory unit 122 stores building type identification rules 123.

The building type identification unit 121 analyzes the address information input by the address information acquisition unit 100, and identifies the building type of the building located at the address based on the building type identification rules 123 stored in the rule storage unit 122. The rule-based building type identification procedure will be described later. The building type identification unit 121 outputs the identified building type to the building type determination unit 140. Furthermore, the building type identification unit 121 may output the identified building type to the first processing unit 110.

The learning processing unit 130 is a processing unit that trains the building type estimation model 115. The learning processing unit 130 includes a learning data acquisition unit 131, a learning unit 132, and a learning data storage unit 133. The learning data storage unit 133 stores the learning data acquired by the learning data acquisition unit 131.

The learning data acquisition unit 131 acquires learning data 134 necessary for training the building type estimation model 115 and stores it in the learning data storage unit 133. In this embodiment, the learning data 134 includes data previously acquired by the natural language processing system 10 and a building type as correct answer data. For example, the learning data 134 includes address information previously acquired by the address information acquisition unit 100 and correct answer data of the building type of a building located at the address indicated by the address information. Furthermore, the learning data 134 may include location information acquired by the location information acquisition unit 112 based on the address information and area information acquired by the area information acquisition unit 113. Furthermore, the learning data 134 may include a building type identified by the building type identification unit 121 based on the address information. The learning data acquisition unit 131 may acquire the learning data 134 from the address information acquisition unit 100, the first processing unit 110, and the second processing unit 120, may acquire the learning data 134 by input from an operator (user), or may acquire the learning data 134 from an external device. The learning unit 132 uses the learning data 134 to train the building type estimation model 115.

The building type determination unit 140 determines the final building type based on the building type estimated by the building type estimation unit 111 of the first processing unit 110 and the building type identified by the building type identification unit 121 of the second processing unit 120. The procedure for determining the building type will be described later. The building type determination unit 140 may also be configured to output the determined building type.

[Hardware configuration of information processing device]
The natural language processing system 10 is configured using one or more information processing devices. Fig. 2 is a block diagram showing an example of a hardware configuration of an information processing device 20 according to this embodiment. The information processing device 20 can be implemented on a single or multiple computers, mobile devices, or any other processing platform.
2, the information processing device 20 is illustrated as being implemented in a single computer, but the information processing device 20 according to the present embodiment may be implemented in a computer system including multiple computers. The multiple computers may be connected to each other via a wired or wireless network so as to be able to communicate with each other.

2, the information processing device 20 may include a CPU (Central Processing Unit) 21, a ROM (Read Only Memory) 22, a RAM (Random Access Memory) 23, a HDD (Hard Disk Drive) 24, an input unit 25, a display unit 26, a communication I/F 27, and a system bus 28. The information processing device 20 may also include an external memory.
The CPU 21 generally controls the operation of the information processing device 20, and controls each of the components (22 to 27) via a system bus 28 which is a data transmission path.

The ROM 22 is a non-volatile memory that stores control programs and the like necessary for the CPU 21 to execute processes. Note that the programs may be stored in a non-volatile memory such as the HDD 24 or an SSD (Solid State Drive) or an external memory such as a removable storage medium (not shown).
The RAM 23 is a volatile memory and functions as a main memory, a work area, etc. of the CPU 21. That is, when executing a process, the CPU 21 loads necessary programs, etc. from the ROM 22 into the RAM 23 and executes the programs, etc. to realize various functional operations. The learning model storage unit 114, the rule storage unit 122, and the learning data storage unit 133 shown in FIG. 1 can be configured by the RAM 23.

The HDD 24 stores, for example, various data and various information required when the CPU 21 performs processing using a program. The HDD 24 also stores, for example, various data and various information obtained when the CPU 21 performs processing using a program.
The input unit 25 is composed of a keyboard and a pointing device such as a mouse.
The display unit 26 is configured with a monitor such as a liquid crystal display (LCD). The display unit 26 may be configured in combination with the input unit 25 to function as a GUI (Graphical User Interface).

The communication I/F 27 is an interface that controls communication between the information processing device 20 and external devices.
The communication I/F 27 provides an interface with a network and executes communication with an external device via the network. Various data, various parameters, etc. are transmitted and received between the external device and the communication I/F 27. In this embodiment, the communication I/F 27 may execute communication via a wired LAN (Local Area Network) or a dedicated line that complies with a communication standard such as Ethernet (registered trademark). However, the network that can be used in this embodiment is not limited to this, and may be configured as a wireless network. This wireless network includes wireless PANs (Personal Area Networks) such as Bluetooth (registered trademark), ZigBee (registered trademark), and UWB (Ultra Wide Band). It also includes wireless LANs (Local Area Networks) such as Wi-Fi (Wireless Fidelity) (registered trademark) and wireless MANs (Metropolitan Area Networks) such as WiMAX (registered trademark). It also includes wireless WANs (Wide Area Networks) such as 4G and 5G defined by 3GPP (Third Generation Partnership Project) (registered trademark). Note that the network only needs to be able to connect devices to each other and communicate with each other, and the communication standard, scale, and configuration are not limited to those described above.

At least some of the functions of the natural language processing system 10 shown in FIG. 1 can be realized by the CPU 21 in the information processing device 20 executing a program. However, at least some of the functions of the natural language processing system 10 shown in FIG. 1 may be operated as dedicated hardware. In this case, the dedicated hardware may operate under the control of the CPU 21.

[Estimation procedure using building type estimation model]
Next, the building type estimation model 115 and an estimation procedure using the building type estimation model 115 will be described. FIG. 3A shows an example of an estimation procedure using the building type estimation model 115. The building type estimation model 115 shown in FIG. 3A is a trained learning model in which machine learning (including fine-tuning) is performed by the learning unit 132 using the address information and building type included in the training data 134. As shown in FIG. 3A, the building type estimation model 115 is composed of an encoder (encoder model) 116 and a classifier (classification model) 117. The encoder 116 generates a distributed representation (also called a vector representation, word embedding, etc.) of each token (small unit such as a word) of the input address information (i.e., text information). The classifier 117 estimates the building type 21 from the distributed representation.

Specifically, the encoder 116 divides (tokenizes) the address information into multiple tokens (small units such as words) and generates a distributed representation of each token. The distributed representation can be generated using a natural language processing (NLP) model such as BERT (Bidirectional Encoder Representations from Transformers), Word2Vec, GloVe (Global Vectors for Word Representation), FastText, etc. In other words, the building type estimation model 115 can be configured using a natural language processing model.
The classifier 117 estimates and outputs a building type from the distributed representation generated by the encoder 116. In this embodiment, the classifier 117 estimates (classifies) the probability (likelihood, probability) that the building corresponds to each of a "business building", a "unit building", and a "house" from the distributed representation. The classifier 117 then estimates and outputs the building type with the highest probability as the building type located at the address indicated by the input address information. The classifier 117 is configured, for example, with a FFNN (Feed-Forward Neural Network) model.

FIG. 4 shows a specific example of estimating a building type from address information. First, the building type estimation unit 111 inputs address information 40, such as "1-2-3, D Town, ABC Ward, Tokyo, inside abc Co., Ltd.", into the building type estimation model 115. The encoder 116 in the building type estimation model 115 divides the address information 40 into a plurality of tokens 41. For example, the address information 40 is divided into a plurality of tokens 41 including "Tokyo", "ABC Ward", "D Town", "1-2-3", "Inc.", "abc", and "inside". The encoder 116 further performs natural language processing such as BERT on the plurality of tokens 41 to generate a distributed representation 42 of each token. When BERT is used to generate a distributed representation, the address information (character string) is learned in both directions, so that it is possible to generate a distributed representation after analyzing the context of the address information. Furthermore, the classifier 117 estimates a building type 43 from the distributed representation 42. In this embodiment, the classifier 117 estimates the type of building located at the address indicated by the address information 40 from three building types, namely, "business building,""unitbuilding," and "house." FIG. 4 shows an example in which "business building" is estimated from the address information 40.

The building type estimation model 115 shown in FIG. 3A is configured to estimate the building type 31 from address information 30 as input data, but may also be configured to estimate the building type 31 using additional input data. Two examples are described here.

A first example of an estimation process using additional input data will be described with reference to FIG. 3B. FIG. 3B shows another example of an estimation procedure using the building type estimation model 115. As shown in FIG. 3B, the classifier 117 is configured to estimate the building type 31 using the building type 33 identified by the building type identification unit 121 based on the address information 30 as the additional input data 32. That is, the classifier 117 is configured to estimate the building type 31 using the identified building type 33 in addition to the input (distributed representation) from the encoder 116. The building type estimation model 115 shown in FIG. 3B is a trained learning model in which machine learning (including fine-tuning) has been performed by the learning unit 132. In the learning stage, machine learning is performed using the address information, building type, and identified building type included in the learning data 134.

By using the building type estimation model 115 shown in FIG. 3B, it becomes possible to estimate the building type taking into account the building type identified by the building type identification unit 121. In other words, the classifier 117 becomes able to estimate the building type according to the weight (gradient) of the building type identified by the building type identification unit 121.

A second example of the estimation process using additional input data will be described with reference to FIG. 3C. FIG. 3C shows another example of the estimation procedure using the building type estimation model 115. As shown in FIG. 3C, the classifier 117 is configured to estimate the building type 31 based on the address information 30 using the location information 34 acquired by the location information acquisition unit 112 and the area information 35 acquired by the area information acquisition unit 113 as the additional input data 32. That is, the classifier 117 is configured to estimate the building type 31 using the location information 34 and the area information 35 in addition to the input (distributed representation) from the encoder 116. The building type estimation model 115 shown in FIG. 3C is a trained learning model in which machine learning (including fine tuning) has been performed by the learning unit 132. In the learning stage, machine learning is performed using the address information, building type, location information, and area information included in the learning data 134.

By using the building type estimation model 115 shown in FIG. 3C, it becomes possible to estimate the building type taking into account the location information acquired by the location information acquisition unit 112 and the area acquired by the area information acquisition unit 113. That is, the classifier 117 is able to estimate the building type according to the location information acquired by the location information acquisition unit 112 and the weight (gradient) of the area acquired by the area information acquisition unit 113. Note that, although both the location information and the area information are input to the classifier 117 as additional input data 32 in FIG. 3C, it may be configured so that either information is input to the classifier 117.

In addition, without being limited to the above, the building type estimation model 115 may be configured so that any combination of the identified building type 33 shown in FIG. 3B and the location information 34 and area information 35 shown in FIG. 3C is input to the classifier 117. In this case, in the learning stage, machine learning is performed using information included in the learning data 134 that corresponds to the data input to the building type estimation model 115.

[Rule-based building type identification procedure]
Next, a description will be given of the procedure of the building type identification process by the building type identification unit 121. In this embodiment, the building type identification unit 121 identifies the building type based on the building type identification rules 123 stored in the rule storage unit 122. Specifically, the building type identification unit 121 identifies the building type by applying the address information (i.e., the character string of the address) to the building type identification rules 123.

The building type identification rule 123 is a pre-created rule for identifying a building type based on the characteristics of the address information. Note that the building type identification rule 123 described in this embodiment is an example, and although there may be cases where it does not match actual examples of combinations of addresses and building types, any rule capable of identifying a building type can be used. For example, a building type may be identified depending on whether or not the address information has a specified keyword. Furthermore, new rules may be added to the building type identification rule 123.

The building type identification rules 123 according to this embodiment include nine types of rules, (1) to (9). The rules (1) to (9) are configured to identify building types based on the characteristics of address information.
(1) If the address information includes "Inc.", the building type is identified as "Business Building."
For example, "1-2-3 D-cho, ABC-shi, Osaka Prefecture, inside ABC Co., Ltd." is identified as a "business building."

(2) If the address information includes "building" and includes three digits after the "building," the building type is identified as "unit building."
For example, "abc building 303, 1-2-3, D-cho, ABC-ku, Tokyo" is identified as a "unit building."

(3) If the address information includes "building" but does not include three digits after the "building," the building type is identified as "business building."
For example, "abc building, 5th floor, 1-2-3 D-cho, ABC-ku, Tokyo" is identified as a "business building."

(4) If the address information includes the final part "number", the building type is identified as "unit building".
For example, "Residence abc 303, 1-2-3, D-cho, ABC-shi, Osaka Prefecture" is identified as a "unit building."

(5) If the address information contains an "x1-x2-x3" format, where "x3" is not a three or four digit number, then the building type is identified as "home."
Example: "1-234-5 D-cho, ABC-shi, Saitama-ken" is identified as a "house."

(6) If the address information contains an "x1-x2" format, the building type is identified as "unit building."
For example, "DE123-4, ABC-cho, Kanagawa Prefecture" is identified as a "unit building."

(7) If the address information contains a single digit and includes that digit at the end, the building type is identified as "home."
For example, "1234 D-cho, ABC-shi, Chiba Prefecture" is identified as a "house."

(8) If the address information contains an “x1-x2-x3” or “x1-x2-x3-x4” format, where “x3” or “x4” is a three or four digit number, the building type is identified as a “unit building.”
Example: "FG1-2-3-303 DE ward, ABC city, Fukuoka prefecture" is identified as a "unit building."

(9) If the address information includes an “x1-x2-x3” or “x1-x2-x3-x4” format and includes a trailing number in addition to the format, the building type is identified as a “unit building.”
For example, "Residence abc 303, DE1-2-3, ABC City, Kanagawa Prefecture" is identified as a "unit building."

In addition, the building type identification unit 121 can be configured to recognize "x1-x2-x3" in the above rules as the same string as variations such as "x1-chome x2-no. x3" or "x1-chome x2-x3."
In addition, the building type identification unit 121 can be configured to recognize, in the above rules, "x1-x2-x3-x4" as the same string as variations such as "x1-chome x no. 2 x no. 3 (building name) x 4" or "x1-chome x 2-x3 (building name) x 4."

Figure 5 shows examples of address information 50 listed in rules (1) to (9) above and building types 51 identified by the address information. As shown in Figure 5, the address information 50 is classified into one of the building types 51, "business building," "unit building," and "house," according to the characteristics of the address information.

For an address whose building type cannot be identified by the above rules (1) to (9), the building type identification unit 105 may identify the building type of the address as a predetermined building type such as "house." Alternatively, for an address whose building type cannot be identified by the above rules (1) to (9), the building type identification unit 105 may identify the building type of the address as "unknown type."

In addition, rules (1) to (9) may be applicable to some areas, such as Tokyo, but may not be applicable to other areas. Therefore, rules other than rules (1) to (9) may be created and applied to other areas. Furthermore, rules other than rules (1) to (9) may be created and applied even to areas within Tokyo.

[Overall processing flow]
6 shows a flowchart of the overall process executed in the natural language processing system 10 according to this embodiment. It is assumed that the learning model storage unit 114 stores a trained building type estimation model 115 that has been machine-learned by the learning unit 132. In this example, it is assumed that there are three building types: "business building", "unit building", and "house".

In S61, the address information acquisition unit 100 accepts input information including an address and acquires address information consisting of text information. When the accepted input information is composed of text information representing an address, the address information acquisition unit 100 acquires the accepted information itself as address information. When the accepted input information is composed of text information representing an address and text information other than the address or image information, the address information acquisition unit 100 may extract text information (text portion) representing the address from the accepted input information to acquire address information. When the accepted input information is audio information, the address information acquisition unit 100 converts the accepted audio information into text information (i.e., transcribes it) and acquires the converted information as address information. The address information acquisition unit 100 outputs the acquired address information to the first processing unit 110 and the second processing unit 120.

In S62, the building type estimation unit 111 in the first processing unit 110 inputs the address information acquired from the address information acquisition unit 100 to the building type estimation model 115 to estimate the building type of the building located at the address indicated by the address information. The procedure for estimating the building type is as described above. For example, as shown in FIG. 3A, the building type estimation unit 111 inputs the address information to the building type estimation model 115 to estimate the building type. Alternatively, as shown in FIG. 3B, the building type estimation unit 111 may input the address information and the building type identified in S63 described later to the building type estimation model 115 to estimate the building type. Alternatively, as shown in FIG. 3C, the building type estimation unit 111 may input the address information and at least one of the estimated building type, the location information acquired by the location information acquisition unit 112, and the area information acquired by the area information acquisition unit 113 to the building type estimation model 115 to estimate the building type. Furthermore, the building type estimation unit 111 may input a combination of the address information and at least one of the building type, the location information, and the area information to the building type estimation model 115 to estimate the building type.
The building type estimation unit 111 outputs the building type estimated in S62 to the building type determination unit 140 as an estimated building type.

In S63, the building type identification unit 121 in the second processing unit 120 uses the building type identification rule 123 to identify the type of building located at the address indicated by the address information acquired from the address information acquisition unit 100. The building type identification procedure is as described above. The building type identification unit 121 identifies one of a "business building", a "unit building", and a "house" from the address information.
The building type identification unit 121 outputs the building type identified in S63 to the building type determination unit 140 as an identified building type.

The building type estimation process in S62 and the building type identification process in S63 may be performed simultaneously, or one may be performed first. However, if the building type identified in S63 is used in the building type estimation process in S62, the process in S63 is performed first.

In S64, the building type determination unit 140 determines the final building type based on the estimated building type and the identified building type. The building type determination process in S64 will be described with reference to Figures 7A to 7C.

Figure 7A shows a flowchart of a first example of the building type determination process. In this first example, priorities are set in advance for the building types that the estimated building type and the identified building type can take, and when the estimated building type and the identified building type differ, the building type is determined according to the priorities. It is assumed that the building type determination unit 140 has acquired setting information regarding the priorities in advance. In this embodiment, it is assumed that the priorities for the building types are set so that "house" has the highest priority, "business building" has the second priority, and "unit building" has the third priority.

In FIG. 7A, the building type determination unit 140 determines whether the estimated building type and the identified building type are the same (S71). If the estimated building type and the identified building type are the same (Yes in S71), the building type determination unit 140 determines the estimated building type (or the identified building type) as the final building type (S72). On the other hand, if the estimated building type and the identified building type are different (No in S71), the building type determination unit 140 determines the building type with the higher priority between the estimated building type and the identified building type as the final building type according to a preset priority (S73). For example, if the estimated building type is a "unit building" and the identified building type is a "business building", then "business building" is determined as the final building type according to a preset priority.

Figure 7B shows a flowchart of a second example of the building type determination process. In this second example, it is set in advance which of the estimated building type and the identified building type will be prioritized when the estimated building type and the identified building type differ, and the building type is determined according to the setting regarding the priority. It is assumed that the building type determination unit 140 has acquired setting information regarding the priority in advance. In this embodiment, it is assumed that the setting is such that the identified building type is prioritized. In other words, when the estimated building type and the identified building type differ, the identified building type is prioritized over the estimated building type.

In Fig. 7B, the processes of S71 and S72 are the same as those in Fig. 7A, and therefore the description thereof will be omitted. If the estimated building type and the identified building type are different (No in S71), the building type determination unit 140 determines the identified building type as the final building type in accordance with the setting regarding the priority (S74).

Figure 7C shows a flowchart of a third example of the building type determination process. In this third example, if the estimated building type and the identified building type differ, the operator is notified that the estimated building type and the identified building type differ, and the operator determines the final building type.

In FIG. 7C, the processes of S71 and S72 are the same as those in FIG. 7A, and therefore will not be described. If the estimated building type and the identified building type are different (No in S71), the building type determination unit 140 outputs a message prompting the operator to perform a manual operation (S75). The message may include address information. For example, the operator can visually determine the building type located at the address indicated by the address information in the message output in S75 using a map service or the like. If an instruction is given by the operator after the message is output (for example, the operator inputs a building type), the building type determination unit 140 determines the specified building type as the final building type (S76).

Returning to the explanation of FIG. 6, once the final building type is determined in S64, the process proceeds to S65. In S65, the building type determination unit 140 outputs the final building type. For example, the building type determination unit 140 may output the final building type to an external device (not shown).

In this way, according to this embodiment, the building type of the building located at the address indicated by the address information, which is composed of text, can be estimated from the address information using a trained learning model constructed using a natural language processing model. Furthermore, to complement the estimation process, the building type of the building located at the address indicated by the address information is identified on a rule-based basis, and the final building type is determined based on the estimation result and the identification result. This increases the accuracy of the building type determined based on the address information, which is expected to improve the reliability of services that use the building type.

Although an embodiment of the identification has been described above, this embodiment is merely an example and is not intended to limit the scope of the present invention. The apparatus and method described in this specification can be embodied in forms other than those described above. Furthermore, the above-described embodiments can be omitted, substituted, and modified as appropriate without departing from the scope of the present invention. Forms in which such omissions, substitutions, and modifications have been made are included within the scope of what is described in the claims and their equivalents, and belong to the technical scope of the present invention.

The disclosure of this embodiment includes the following configuration.
[1] A natural language processing system having an acquisition unit that acquires address information, and an estimation unit that estimates a building type of a building located at an address indicated by the address information by inputting the address information into a learning model that has been trained through machine learning.

[2] The natural language processing system described in [1], further comprising an identification unit that identifies the building type from the address information using predetermined rules.

[3] The natural language processing system described in [2], further comprising a determination unit that determines the building type based on a first building type indicating the building type estimated by the estimation unit and a second building type indicating the building type identified by the identification unit.

[4] The natural language processing system described in [3], in which, when the first building type and the second building type are the same, the determination unit determines the first building type or the second building type as the building type.

[5] A natural language processing system according to [3] or [4], in which a priority is set for each of the multiple building types that the first building type and the second building type can take, and when the first building type and the second building type are different, the determination unit determines, from the first building type and the second building type, the building type with the higher priority as the building type.

[6] The natural language processing system described in [3] or [4], in which, when the first building type and the second building type are different, the determination unit determines the second building type as the building type.

[7] The natural language processing system described in [3] or [4], in which, when the first building type and the second building type are different, the determination unit determines the building type specified by the operator as the building type.

[8] A natural language processing system according to any one of [3] to [7], wherein the estimation unit estimates the building type by inputting the address information and the second building type into the learning model.

[9] A natural language processing system according to any one of [3] to [8], further comprising a location information acquisition unit that acquires location information of an address indicated by the address information, and the estimation unit estimates the building type by inputting the address information and the location information into the learning model.

[10] A natural language processing system according to any one of [3] to [9], further comprising an area information acquisition unit that acquires, as area information, at least one of the following: a land use district to which the address indicated by the address information belongs, the nearest train station to the address, and the distance from the nearest train station to the address; and the estimation unit estimates the building type by inputting the address information and the area information into the learning model.

[11] The natural language processing system according to any one of [1] to [10], wherein the learning model is constructed using a natural language processing model.

[12] A natural language processing method including acquiring address information, and inputting the address information into a learning model trained by machine learning, thereby estimating a building type of a building located at an address indicated by the address information.

[13] A natural language processing program for causing a computer to execute natural language processing, the program causing the computer to execute processes including an acquisition process for acquiring address information, and an estimation process for estimating the building type of a building located at an address indicated by the address information by inputting the address information into a learning model that has been trained by machine learning.

10: Natural language processing system, 100: Address information acquisition unit, 110: First processing unit, 111: Building type estimation unit, 112: Location information acquisition unit, 113: Area information acquisition unit, 114: Learning model storage unit, 115: Building type estimation model, 120: Second processing unit, 121: Building type identification unit, 122: Rule storage unit, 123: Building type identification rule, 130: Learning processing unit, 131: Learning data acquisition unit, 132: Learning unit, 133: Learning data storage unit, 134: Learning data, 140: Building type determination unit

Claims (12)

An acquisition unit for acquiring address information;
an estimation unit that estimates a building type of a building located at an address indicated by the address information by inputting the address information into a learning model that has been trained by machine learning;
having
The learning model is composed of an encoder and a classifier,
The estimation unit is
inputting the address information into the encoder, splitting the address information into a plurality of tokens, and generating a distributed representation of each token;
The distributed representation is input to the classifier to estimate a probability that the distributed representation corresponds to each of a plurality of predefined building types;
estimating the building type based on the probability;
Natural language processing system. an identification unit that identifies a building type from the address information by using a predetermined rule that is created in advance for identifying a building type of the reference address information based on a feature of the reference address information ;
a determination unit that determines the building type based on a first building type indicating the building type estimated by the estimation unit and a second building type indicating the building type identified by the identification unit,
The natural language processing system according to claim 1 . When the first building type and the second building type are the same, the determination unit determines the first building type or the second building type as the building type.
The natural language processing system according to claim 2 . a priority is set for each of a plurality of building types that can be the first building type and the second building type;
When the first building type and the second building type are different, the determination unit determines, as the building type, a building type having a higher priority among the first building type and the second building type.
The natural language processing system according to claim 2 . When the first building type and the second building type are different, the determination unit determines the second building type as the building type.
The natural language processing system according to claim 2 . When the first building type and the second building type are different, the determination unit determines the building type designated by an operator as the building type.
The natural language processing system according to claim 2 . The estimation unit is
inputting the distributed representation and the second building type into the classifier to estimate a second probability that the building type corresponds to each of a plurality of preset building types;
estimating the building type based on the second probability ;
The natural language processing system according to claim 2 . A location information acquisition unit that acquires location information of an address indicated by the address information,
The estimation unit is
inputting the distributed representation and the location information into the classifier to estimate a third probability that corresponds to each of a plurality of predefined building types;
estimating the building type based on the third probability ;
The natural language processing system according to claim 2 . The method further includes an area information acquisition unit that acquires, as area information, at least one of a land use district to which the address indicated by the address information belongs, a nearest station to the address, and a distance from the nearest station to the address;
The estimation unit is
inputting the distributed representation and the area information into the classifier to estimate a fourth probability that the area corresponds to each of a plurality of predefined building types;
estimating the building type based on the fourth probability ;
The natural language processing system according to claim 2 . The learning model is configured using a natural language processing model.
The natural language processing system according to claim 1 . A natural language processing method executed by an information processing device, comprising:
Obtaining address information;
inputting the address information into a learning model trained by machine learning to estimate a building type of a building located at an address indicated by the address information;
Including,
The learning model is composed of an encoder and a classifier,
The estimating step comprises:
inputting the address information into the encoder, splitting the address information into a plurality of tokens, and generating a distributed representation of each token;
The distributed representation is input to the classifier to estimate a probability that the distributed representation corresponds to each of a plurality of predefined building types;
estimating the building type based on the probability.
Natural language processing methods. A natural language processing program for causing a computer to execute natural language processing, the program comprising:
An acquisition process for acquiring address information;
and an estimation process of estimating a building type of a building located at an address indicated by the address information by inputting the address information into a learning model that has been trained by machine learning ,
The learning model is composed of an encoder and a classifier,
The estimation process includes:
inputting the address information into the encoder, splitting the address information into a plurality of tokens, and generating a distributed representation of each token;
The distributed representation is input to the classifier to estimate a probability that the distributed representation corresponds to each of a plurality of predefined building types;
estimating the building type based on the probability.
Natural language processing program.

Images