Behavioral analysis assistance device, behavioral analysis assistance method, and behavioral analysis assistance program

Abstract

To provide an action analysis supporting device for converting a letter string into a real-valued vector suitable for an action analysis, storing the vector, and providing the vector in response to a request from the outside.SOLUTION: An action analysis supporting device includes: an acquisition unit for acquiring a letter string formed by arranging words about an action history of an action subject in order to passage of time; a conversion unit for converting at least a part of the letter string to a real-valued vector of a fixed length by a dispersion expression; a storage unit for storing the real-valued vector converted by the conversion unit in association with the action subject or words contained in the letter string; and a provision unit for providing the real-valued vector stored in the storage unit in response to a request from the action analysis program.SELECTED DRAWING: Figure 2

Description

[Technical field]

[0001] The present invention relates to a behavior analysis assistance device, a behavior analysis assistance method, and a behavior analysis assistance program. [Background technology]

[0002] Services that provide content suited to user needs via the Internet, etc. For example, an information processing device is known that extracts content that is highly similar to content previously viewed by a user and provides the content to the user (for example, see Patent Document 1). [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2020-42545 A Summary of the Invention [Problem to be solved by the invention]

[0004] The information processing device as described above expresses each content as a numerical vector by using, for example, a technique of distributed representation, and derives the similarity between the contents by calculating the cosine similarity. However, conventional numerical vectorization using distributed representation is generally performed using a general-purpose corpus or database, and is limited to expressing the linguistic characteristics of the target words and phrases. In particular, it does not reflect the behavioral characteristics of the subject of the operation as a consumer.

[0005] The present invention has been made to solve such problems, and provides a behavioral analysis assistance device that converts character strings into real-valued vectors suitable for behavioral analysis, stores the vectors, and provides them in response to external requests. [Means for solving the problem]

[0006] A behavior analysis support device in a first aspect of the present invention includes an acquisition unit that acquires a string in which words related to a behavior history of an agent are arranged over time, a conversion unit that converts at least a part of the string into a fixed-length real-valued vector using a distributed representation, a memory unit that stores the real-valued vector converted by the conversion unit in correspondence with the agent or a word contained in the string, and a provision unit that provides the real-valued vector stored in the memory unit in accordance with a request from a behavior analysis program.

[0007] A behavioral analysis assistance method in a second aspect of the present invention includes an acquisition step of acquiring a string in which words related to an agent's behavioral history are arranged over time; a conversion step of converting at least a part of the string into a fixed-length real-valued vector using distributed representation; a storage step of storing the real-valued vector converted in the conversion step in a memory unit in association with the agent or a word contained in the string; and a provision step of providing the real-valued vector stored in the memory unit in accordance with a request from a behavioral analysis program.

[0008] A behavior analysis assistance program in a third aspect of the present invention causes a computer to execute an acquisition step of acquiring a string in which words related to an agent's behavior history are arranged over time; a conversion step of converting at least a part of the string into a fixed-length real-valued vector using a distributed representation; a storage step of storing the real-valued vector converted in the conversion step in a memory unit in association with the agent or a word contained in the string; and a provision step of providing the real-valued vector stored in the memory unit in accordance with a request from the behavior analysis program. Effect of the Invention

[0009] According to the present invention, it is possible to provide a behavior analysis assistance device or the like that converts character strings into real-valued vectors suitable for behavior analysis, stores the converted vectors, and provides the converted vectors in response to an external request. [Brief description of the drawings]

[0010] [Figure 1]1 is a diagram illustrating an overall environment in which a behavior analysis assistance device according to an embodiment of the present invention is used. [Diagram 2] FIG. 2 is a diagram illustrating a hardware configuration of the behavior analysis assistance device. [Diagram 3] 11 is a diagram illustrating acquisition of a character string related to a behavior history of an action subject. FIG. [Figure 4] FIG. 13 is a diagram illustrating the concept of a process for converting an entire character string into a real-valued vector. [Diagram 5] FIG. 10 is a diagram illustrating the concept of a process for converting each word included in a character string into a real-valued vector. [Figure 6] FIG. 13 is a diagram illustrating the concept of a process for associating words included in a character string with their occurrence frequency and converting them into a real-valued vector. [Figure 7] FIG. 11 is a flowchart illustrating a process for accumulating a real-valued vector. [Figure 8] 11 is a flow diagram illustrating a processing procedure for providing a real-valued vector in response to a request from a behavior analysis program. FIG. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0011] The present invention will be described below through embodiments of the invention, but the invention according to the claims is not limited to the following embodiments. Furthermore, not all of the configurations described in the embodiments are necessarily essential as means for solving the problems.

[0012] 1 is a diagram for explaining the overall environment in which a database server 100 functioning as a behavior analysis support device according to this embodiment is used. The database server 100 is connected to the Internet 900, and directly or indirectly transmits and receives information to and from a media server 210, an application server 220, an information terminal 300, and an analysis server 400 via the Internet 900. The media server 210 is, for example, a server that provides a social network service (SNS). When a user operates his / her own information terminal 300, which is, for example, a tablet terminal, to transmit a comment such as a tweet, the comment is associated with a user ID, a transmission time, etc., and stored in a storage unit of the media server 210.

[0013] The application server 220 is, for example, a server that provides an online payment service. When a user operates his / her own information terminal 300, which is, for example, a smartphone, to make a payment at a store, information about the payment is stored in the storage unit of the application server 220 in association with a user ID, a payment time, a store name, and the like. In addition to the media server 210 and the application server 220, various servers that can store the user's behavior history may be connected to the Internet 900. For example, a server that records the user's behavior log using a GPS function installed in the information terminal 300 may be included. The server that can store the user's behavior history may also be a server connected to a credit card reading terminal 230, an electronic payment reading scanner 240, an automatic ticket gate device 250 that reads a transportation IC card, and the like installed in the store.

[0014] The database server 100 acquires words related to the user's behavior history stored in these servers as character strings arranged over time. Then, at least a part of the acquired character string is converted into a fixed-length real-valued vector using distributed representation, and stored in association with the subject of the action or the word contained in the character string. The real-valued vectors thus converted and stored eventually build a database of a certain size. The database server 100 provides the real-valued vectors from the database thus built in response to a request from the analysis server 400.

[0015] The analysis server 400 is a server that analyzes product demand forecasts, store visitor forecasts, optimal web advertisement placements, etc., based on the behavioral characteristics of the operating subject. Specifically, using real value vectors stored in the database server 100, for example, it analyzes the next product that a user who has purchased a certain product is likely to purchase.

[0016] 2 is a diagram showing a hardware configuration of a database server 100 which is a behavior analysis assistance device. The database server 100 is mainly composed of a calculation unit 110, a storage unit 120, and a communication unit 130. The calculation unit 110 is a processor (CPU: Central Processing Unit) which controls the database server 100 and executes programs. The processor may be configured to cooperate with a calculation processing chip such as an ASIC (Application Specific Integrated Circuit) or a GPU (Graphics Processing Unit). In particular, the calculation unit 110 executes various processes related to assistance in behavior analysis according to a behavior analysis assistance program stored in the storage unit 120 or transmitted from an external device.

[0017] The storage unit 120 is a non-volatile storage medium, and is configured, for example, by an HDD (Hard Disk Drive). The storage unit 120 may store various parameter values, functions, lookup tables, learned models, and the like used for control and calculation, in addition to programs for executing control and processing of the database server 100. The storage unit 120 particularly stores databases, namely, a subject-specific DB 121, a word-specific DB 122, and a frequency-specific DB 123. The subject-specific DB 121 is a database that stores and accumulates a real-valued vector generated by converting the entire character string in which words related to behavior history are arranged over time, in association with the subject of the action. The word-specific DB 122 is a database that stores and accumulates a real-valued vector generated by converting each word included in a character string in which words related to behavior history are arranged over time, in association with the word. The frequency-specific DB 123 is a database that stores and accumulates a real-valued vector generated by associating words included in a character string in which words related to behavior history are arranged over time with the frequency of appearance, in association with the subject of the action.

[0018] The storage unit 120 may be composed of multiple pieces of hardware, for example, a storage medium for storing the program and a storage medium for storing each database may be composed of separate hardware. The storage unit 120 does not have to be built into the database server 100, and may be, for example, a storage device connected via a network. In that case, the database server 100 and the storage device constitute a behavior analysis support device.

[0019] The communication unit 130 is responsible for connecting to the Internet 900 and exchanging data with external devices, and is configured by, for example, a LAN unit. The communication unit 130 exchanges information with the Internet 900 under the control of the processor 110.

[0020] The calculation unit 110 also plays a role as a functional calculation unit that executes various calculations according to the processing instructed by the behavior analysis assistance program. The calculation unit 110 can function as an acquisition unit 111, a conversion unit 112, and a provision unit 113. The acquisition unit 111 acquires a character string in which words related to the behavior history of the subject of the action are arranged over time from the media server 210 and the application server 220 via the communication unit 130. The conversion unit 112 converts at least a part of the character string acquired by the acquisition unit 111 into a fixed-length real-valued vector using a distributed representation, and adds it to the corresponding database of the storage unit 120. The provision unit 113 provides the real-valued vector stored in each database of the storage unit 120 according to a request from the behavior analysis program. Specific processing will be described later.

[0021] 3 is a diagram for explaining acquisition of character strings related to the behavior history of an action subject. As described above, the acquisition unit 111 acquires the character strings from the media server 210 and the application server 220. In other words, if target category words can be collected from those servers for each action subject and arranged in a time sequence, the arranged character strings are imported. For example, if the target category is "stores," it is possible to arrange the stores visited and the order in which they were visited from the records of each person's posts to SNS and online payments.

[0022] Specifically, the acquiring unit 111 collects information on the subject of action, the action performed, and the time information on the time when the action was performed, which are stored and stored in the media server 210, the application server 220, and other servers that can store the user's behavior history, in association with each other. From the collected data group in this way, the acquiring unit 111 extracts target category words for each subject of action to be analyzed, and arranges them in a time sequence. If the category is a store, the acquiring unit 111 extracts a history of store visits in a time sequence by referring to the time information included in the data group, and generates a character string in which the store names are arranged as words, to determine the order in which each person visited the store. FIG. 3 shows N "words" that are "store names" collected in this way, arranged in the order of visit.

[0023] Specifically, for example, if it is possible to extract from person A's blog that A visited "Cafe BC," "R Ramen," "D Coffee," and so on to "ST Cafe" in that order, a character string arranged as "Cafe BC, R Ramen, D Coffee, ..., ST Cafe" is obtained. Similarly, if it is possible to confirm from person B's online payment records that B visited "B Camera," "Y Electric," "A Store," and so on to "A Store," a character string arranged as "B Camera, Y Electric, A Store, ..., A Store" is obtained. A character string arranged in this way over time contains the behavioral tendencies of the subject of the action, and it is possible to read a behavioral pattern, such as a tendency to stop by "Y Electric" after visiting "B Camera."

[0024] If the target category is "store", a string containing both cafes and department stores is generated. If it is assumed that behavior analysis will be performed within the framework of "store", such strings can be acquired and collected, but if it is assumed that behavior analysis will be performed within a more detailed framework, for example, "restaurant", then the target category can be set to "restaurant" and strings can be acquired and collected. Other targets include "shopping" and "fast food". In the case of "shopping", for example, a string arranged as "M department store, T department store, L mall, ..., T department store" is acquired from the behavior history of person C. In the case of "fast food", for example, a string arranged as "O burger, K burger, M donut, ..., O burger" is acquired from the behavior history of person D.

[0025] Regarding what kind of target word to extract, for example, a reference dictionary that defines the target to be extracted in advance may be prepared. For example, the reference dictionary for "fast food" lists fast food restaurant names such as "K Burger" and "M Donuts". Alternatively, a tag of an expected category may be added to each word included in the corpus to be referenced. For example, the tag of "fast food" or "restaurant" is added to "K Burger" included in the corpus. In this case, the acquisition unit 111 extracts words that match words having tags of the target category in the corpus from blogs, etc., and generates a character string.

[0026] Of course, the words related to the subject's behavior history are not limited to those related to "stores." In other words, it is not limited to the "stores" visited, but for example, the "sights" visited at tourist spots or the "colors" of clothes worn over a certain period of time can also be category targets. As described above, if the formed character string can contain the subject's behavioral tendency, it will be a selection target.

[0027] The database server 100 may generate such character strings arranged as words by collecting data groups from other servers as described above, but is not limited to this. If another server already holds a data set as such character strings, the database server 100 can simply obtain such a data set.

[0028] Next, a process of converting the acquired character string into a real-valued vector will be described. In this embodiment, the conversion unit 112 can perform a conversion process of converting the entire character string into a real-valued vector, a conversion process of converting each word included in the character string into a real-valued vector, and a conversion process of converting the words included in the character string into real-valued vectors by associating them with their occurrence frequencies.

[0029] 4 is a diagram for explaining the concept of the process of converting the entire character string into a real-valued vector. The real-valued vector obtained by converting the entire character string is stored and accumulated in the subject DB 121 in association with the subject of the action, and is therefore referred to as "subject-specific conversion" here.

[0030] As described above, a character string arranged as "Cafe BC, R Ramen, D Coffee, ..., ST Cafe" can be obtained from person A in FIG. 3. The character string is regarded as one sentence, and the conversion unit 112 converts the entire character string into a real-valued vector of fixed length by distributed representation. Specifically, the Doc2vec method is used. The conversion into a real-valued vector by Doc2vec can be performed by applying a known technique (for example, see the paper "Distributed Representations of Sentences and Documents", https: / / cs.stanford.edu / ~quocle / paragraph_vector.pdf). By converting in this way, a real-valued vector of fixed length, i.e., a set number of dimensions, such as [0.12, 0.02, 0.05, ...], is obtained as shown in the figure. Similarly, the character string "B Camera, Y Electrical Appliance, A Store, ..., A Store" obtained from the behavior history of person B is also converted into a real-valued vector of the same number of dimensions, such as [0.20, 0.03, 0.08, ...], by Doc2vec. The conversion process is performed similarly for other acquired character strings. The conversion unit 112 stores and accumulates each of the real-valued vectors thus converted in the subject-specific DB 121 in association with the subject of the action, "Person A" and "Person B." Here, storing and accumulating the real-valued vector in association with the subject of the action means storing and accumulating the converted real-valued vector in association with information or data that directly or indirectly indicates the subject of the action.

[0031] In this way, a real-valued vector obtained by converting the entire character string in which the behavior history of each subject is arranged over time can be evaluated as including the behavior tendency of the subject. Therefore, such a real-valued vector can be used when extracting people who show similar behavioral tendencies. For example, when the behavior analysis program determines to which of multiple groups classified by purchasing patterns person X, who is the evaluation target, belongs, it first converts person X's past behavior history into a real-valued vector in the same way, and extracts a real-valued vector similar to this from the subject DB 121. Then, it determines the group to which person X belongs to a large number of people associated with the extracted real-valued vector (for example, "belongs to a group that has a strong tendency to purchase home appliances at mass retailers").

[0032] Such real-valued vectors can also be used to predict future behavior of a person who shows a similar behavioral tendency. For example, when the behavior analysis program predicts future behavior based on the past behavior history of person X, first, the past behavior history of person X is converted into a real-valued vector in the same way, and a real-valued vector similar to this is extracted from the subject DB 121. Then, future behavior is predicted by extracting behavior that person X has not yet performed from the information of the person associated with the extracted real-valued vector (for example, "high possibility of going to 'Store A'"). When the providing unit 113 receives a request command requesting a real-valued vector from the subject DB 121 from the behavior analysis program executed in the analysis server 400, the providing unit 113 selects the target real-valued vector from the subject DB 121 and provides it.

[0033] 5 is a diagram explaining the concept of the process of converting each word included in a character string into a real-valued vector. The real-valued vector obtained by converting each word included in a character string is stored and accumulated in the word-specific DB 122 in association with the word, and is therefore referred to as "word-specific conversion" here.

[0034] As described above, for example, from person C in FIG. 3, a character string arranged as "M Department Store, T Department Store, L Mall, ..., T Department Store" can be obtained. If this is regarded as one sentence and attention is paid to "T Department Store," the words "M Department Store" and "L Mall" that come before and after it can be regarded as peripheral words. Therefore, the conversion unit 112 collects character strings generated from other actors including "T Department Store" and converts "T Department Store" into a fixed-length real-valued vector using distributed representation by learning the occurrence probability of each peripheral word for "T Department Store." Specifically, the Word2vec method is used. The conversion into a real-valued vector using Word2vec can be performed using a known technique (for example, see the paper "Efficient Estimation of Word Representations in Vector Space," https: / / arxiv.org / pdf / 1301.3781.pdf). By converting in this way, a fixed-length real-valued vector, i.e., a real-valued vector with a set number of dimensions, such as [0.17, 0.42, 0.15, ...], is obtained as shown in the figure. The conversion unit 112 stores and accumulates each real-valued vector converted for each word in the word-specific DB 122 in association with the original word (for example, "T department store").

[0035] In this way, the real-value vector obtained by converting the words included in the character string in which the behavior history is arranged over time can be evaluated as including general behavioral tendencies of various actors. Therefore, such a real-value vector can be used to estimate behavior that is likely to be performed before or after the behavior of the word in a general behavior pattern. For example, when the behavior analysis program estimates what kind of store the evaluation target "M Donuts" is likely to visit after visiting, it extracts a real-value vector similar to the real-value vector of "M Donuts" from the word-specific DB 122. Then, it determines the store represented by the word associated with the extracted real-value vector as the target store. For example, when it is determined to be "L Mall", it can be estimated that as a general behavior pattern, it is highly likely that "M Donuts" will be visited after visiting "L Mall". When the providing unit 113 receives a request command requesting a real-value vector of the word-specific DB 122 from the behavior analysis program executed by the analysis server 400, the providing unit 113 selects the target real-value vector from the word-specific DB 122 and provides it.

[0036] 6 is a diagram explaining the concept of the process of converting words included in a character string into a real-valued vector by associating the words included in the character string with their frequency of occurrence. The real-valued vector converted by associating the words included in the character string with their frequency of occurrence is stored and accumulated in the frequency-based DB 123 in association with the subject of the action, and is therefore referred to as "frequency-based conversion" here.

[0037] As described above, for example, from person B in FIG. 3, a character string arranged as "B Camera, Y Electric, A Store, ..., A Store" can be obtained. When converting this into a fixed-length real-valued vector by distributed representation, the conversion unit 112 considers the frequency of occurrence of overlapping words. In this example, person B has visited "A Store" twice, and it is estimated that he is highly interested in "A Store". In order to consider the frequency, the TF-IDF method is used here. Then, by using IDF as appearance frequency information, the importance of each word is listed together with the word. By converting in this way, frequency information of each word is obtained, such as ["A Store": 1.2, "Y Electric": 0.5, ...], and by extracting and listing the numerical values ​​from the information, a real-valued vector is obtained. The conversion unit 112 stores and accumulates each of the converted real-valued vectors in the frequency-based DB 123 in association with the subject of the action, "person A", "person B", ....

[0038] In this way, a real-value vector converted based on the frequency of occurrence of words included in a character string in which the behavior history of each subject is arranged over time can be evaluated as including the behavioral bias of the subject. Therefore, such a real-value vector can be used when extracting people who show similar behavioral bias. When the providing unit 113 receives a request command requesting a real-value vector from the frequency-based DB 123 from the behavior analysis program executed by the analysis server 400, the providing unit 113 selects a target real-value vector from the frequency-based DB 123 and provides it.

[0039] Next, a description will be given of the process relating to the accumulation of real-valued vectors by the database server 100. Fig. 7 is a flow diagram illustrating the procedure of the process relating to the accumulation of real-valued vectors. The flow starts when an instruction to start accumulation is given.

[0040] In step S101, the acquisition unit 111 collects words related to the action history of the action subject from the media server 210, the application server 220, etc. via the communication unit 130, and acquires them as character strings arranged chronologically. In step S102, the conversion unit 112 executes the subject-specific conversion described using Fig. 4, and in step S103, stores and accumulates the converted real-valued vector in the subject-specific DB 121 in association with the action subject.

[0041] Next, in step S104, conversion unit 112 executes the word-specific conversion described using Fig. 5, and in step S105, stores and accumulates each converted real-valued vector in association with the original word in word-specific DB 122. Furthermore, in step S106, conversion unit 112 executes the frequency-specific conversion described using Fig. 6, and in step S107, stores and accumulates the converted real-valued vector in frequency-specific DB 123 in association with the subject of the action.

[0042] In step S108, the calculation unit 110 checks whether or not an instruction to complete accumulation has been received. If an instruction to complete accumulation has not been received, the process returns to step S101 to continue the accumulation process. If an instruction to complete accumulation has been received, the process ends.

[0043] When each database accumulates a certain number of real-valued vectors, it can accept a request command from the behavior analysis program executed by the analysis server 400. Fig. 8 is a flow diagram explaining the procedure of a process for providing a real-valued vector in response to a request from a behavior analysis program. The flow starts when the providing unit 113 receives a request command from the behavior analysis program in a state in which the request command can be accepted.

[0044] In step S201, providing unit 113 checks whether the request command requests a real-valued vector for the acting subject. That is, it checks whether the request command requests a real-valued vector stored in subject-specific DB 121. If the request command requests such a real-valued vector, it proceeds to step S202, selects and provides the target real-valued vector from subject-specific DB 121, and proceeds to step S203. If not, it skips step S202 and proceeds to step S203.

[0045] In step S203, providing unit 113 checks whether the request command requests a real-valued vector for a word. That is, it checks whether the request command requests a real-valued vector stored in word-specific DB 122. If the request command requests such a real-valued vector, the process proceeds to step S204, where the target real-valued vector is selected and provided from word-specific DB 122, and the process proceeds to step S205. If not, step S204 is skipped and the process proceeds to step S205.

[0046] In step S205, providing unit 113 checks whether the request command requests a real-valued vector for the frequency. That is, it checks whether the request command requests a real-valued vector stored in frequency-based DB 123. If the request command requests such a real-valued vector, the process proceeds to step S206, where the target real-valued vector is selected and provided from frequency-based DB 123, and the process proceeds to step S207. If not, step S206 is skipped and the process proceeds to step S207.

[0047] In step S207, providing unit 113 checks whether all the request commands have been processed. If there are unprocessed request commands remaining, the process returns to step S201 and continues the process of providing a real-valued vector. If not, the process ends. Note that the process of accumulating the real-valued vector in FIG. 7 may be continued even after the process of providing the real-valued vector in FIG. 8 becomes available for execution.

[0048] In the above-described embodiment, the conversion unit 112 converts the acquired character string into three types of real-valued vectors and stores them in the corresponding databases, but the mode of the conversion unit 112 is not limited to this. The conversion unit 112 may be capable of converting one type, or may be capable of converting two of the three. The conversion unit 112 may also be capable of realizing other types of conversion.

[0049] In addition, in the embodiment described above, the subject of the operation has been described as an individual person, but the subject of the operation is not limited to an individual person, but may be a group that is a collection of people, and also may not be limited to a person, but may be mobility such as a passenger car used by multiple people. [Explanation of symbols]

[0050] 100...database server, 110...calculation unit, 111...acquisition unit, 112...conversion unit, 113...providing unit, 120...storage unit, 121...subject-specific DB, 122...word-specific DB, 123...frequency-specific DB, 130...communication unit, 210...media server, 220...application server, 230...credit card reading terminal, 240...electronic payment reading scanner, 250...automatic ticket gate device, 300...information terminal, 400...analysis server, 900...Internet

Claims

1. an acquisition unit that collects words related to the behavior history of an action subject from a server connected to the Internet and arranges the words into a character string over time; a conversion unit that converts the entire string, each of the words included in the string, or the words included in the string based on an occurrence frequency of the string into a fixed-length real-valued vector using a distributed representation; a storage unit that stores the real-valued vector converted by the conversion unit in association with the subject of the action or a word included in the character string; a providing unit that provides the real-valued vector stored in the storage unit in response to a request from a behavior analysis program; A behavioral analysis assistance device comprising:

2. The conversion unit converts the entire character string into a fixed-length real-valued vector using a distributed representation; The storage unit stores the real-valued vector converted by the conversion unit in association with the subject of the action; The behavior analysis assistance device according to claim 1 , wherein the providing unit provides the real-valued vector for each of the subject of action stored in the storage unit in response to a request from a behavior analysis program.

3. The conversion unit converts each of the words included in the character string into a fixed-length real-valued vector using a distributed representation; the storage unit stores the real-valued vector converted by the conversion unit in association with each of the words; 3. The behavior analysis assistance device according to claim 1, wherein the providing unit provides the real-value vector for each word stored in the storage unit in response to a request from a behavior analysis program.

4. The conversion unit converts the character string into a real-valued vector using a distributed representation based on an appearance frequency of the word included in the character string; The storage unit stores the real-valued vector converted by the conversion unit in association with the subject of the action; 4. The behavior analysis assistance device according to claim 1, wherein the providing unit provides the real-value vector for the occurrence frequency stored in the storage unit in response to a request from a behavior analysis program.

5. An acquisition step in which an acquisition unit collects words related to an action history of an action subject from a server connected to the Internet and arranges the words into a character string over time; A conversion step in which a conversion unit converts the entire character string, each of the words included in the character string, or the words included in the character string based on an appearance frequency into a fixed-length real-valued vector using a distributed representation; a storage step in which the calculation unit stores the real-valued vector converted in the conversion step in a storage unit in association with the subject of the action or a word included in the character string; a providing step in which a providing unit provides the real-valued vector stored in the storage unit in accordance with a request from a behavior analysis program; A behavioral analysis assistance method comprising:

6. An acquisition step of collecting words related to the action history of the action subject from a server connected to the Internet and arranging the words into a character string over time; A conversion step of converting the entire string, each of the words contained in the string, or the words contained in the string based on the frequency of occurrence of the string into a fixed-length real-valued vector using a distributed representation; a storage step of storing the real-valued vector converted in the conversion step in a storage unit in association with the subject of the action or a word included in the character string; a providing step of providing the real-valued vector stored in the storage unit in response to a request from a behavior analysis program; A behavioral analysis assistant program that causes a computer to execute the above.

Images