Information processing device, information processing method, and program

Abstract

To provide an information processing device, an information processing method, and an information processing program which easily and accurately extract information regarding purchasing behavior of a user.SOLUTION: An attribute information extraction device 1 includes: a transaction information acquisition unit 11 which acquires transaction information containing entry which describes transaction in a store; a text separation unit 12 which separates transaction texts from the transaction information acquired by the transaction information acquisition unit 11; an attribute extracting unit 13 which inputs the transaction texts separated by the text separation unit 12 into a plurality of natural language processing models for extracting each of a plurality of attributes respectively corresponding to the transaction texts; and an output unit 14 which outputs the plurality of attributes extracted by the attribute extracting unit 13.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to an information processing device, an information processing method, and a program, and in particular to a technique for extracting attribute information related to a user's purchasing behavior. [Background technology]

[0002] In recent years, there are a wide variety of websites and applications that allow users to purchase products and services online. Users access desired online stores via these numerous websites and applications, and purchase desired products and services through online transactions using payment methods such as credit cards, debit cards, and virtual currencies.

[0003] In such online transactions, information on the purchasing behavior of users, such as which store they made what kind of online transactions, can be analyzed, each user can be profiled, and personalized advertisements can be provided to each user based on the results of this profiling. This is expected to improve the effectiveness of advertisements, as shown by indices such as conversion rate (CVR).

[0004] However, as described above, since there are a wide variety of websites and applications through which users can make online transactions, information on which stores a user has made online transactions at is distributed across many domains, and the formats of the information vary from domain to domain. For this reason, it is not easy to obtain this information from each domain, consolidate it centrally, and provide it for analysis.

[0005] Patent Document 1 discloses a usage details / store information link device that generates usage details information indicating the content of a payment actually made by a user in association with store information indicating the store where the payment was made. Specifically, the usage details / store information link device disclosed in Patent Document 1 transmits usage details information including information identifying the store where the user actually made the payment to a customer terminal, and upon receiving information identifying the store contained in the usage details information from the customer terminal, reads out the corresponding store information and transmits it to the customer terminal. When a user posts about a store that they have actually visited, posting function information included in the transmitted store information is sent from the customer terminal, and the usage details / store information link device disclosed in Patent Document 1 further stores and updates the posted information based on the posting function information. [Prior art documents] [Patent documents]

[0006] [Patent Document 1] Patent Publication No. 2018-72978 Summary of the Invention [Problem to be solved by the invention]

[0007] According to the technology of Patent Document 1, information for identifying the store where the user actually made the payment can be acquired from the transaction details information in a unified manner.

[0008] However, since the usage details information only requires that the user be able to identify the store to which the post was made, it only contains text information about the store name and image data of the store's exterior. It is not possible to directly obtain attribute information that characterizes the purchasing behavior of the user at the store where the payment was actually made from the usage details information. In addition, by referencing the store information to which the usage details information is linked, it is possible to obtain attribute information about a store. However, in order to do this, the usage details / store information link device, which is a server device, must constantly keep up with updates and changes to the store attribute information that occur from time to time in multiple domains, which is cumbersome and lacks flexibility in accommodating changes.

[0009] The present invention has been made to solve the above-mentioned problems, and its object is to provide an information processing device, an information processing method, and a program capable of easily and accurately extracting information related to a user's purchasing behavior. [Means for solving the problem]

[0010] In order to solve the above problem, one aspect of an information processing device according to the present invention includes a transaction information acquisition unit that acquires transaction information including an entry describing a transaction at a store, a text separation unit that separates transaction text from the transaction information acquired by the transaction information acquisition unit, an attribute extraction unit that inputs the transaction text separated by the text separation unit to a plurality of natural language processing models and extracts a plurality of attributes corresponding to the transaction text, respectively, and an output unit that outputs the plurality of attributes extracted by the attribute extraction unit.

[0011] One aspect of an information processing method according to the present invention is an information processing method executed by an information processing device, comprising the steps of acquiring transaction information including an entry describing a transaction at a store, separating transaction text from the acquired transaction information, inputting the separated transaction text into a plurality of natural language processing models to extract a plurality of attributes corresponding to the transaction text, and outputting the extracted plurality of attributes.

[0012] One aspect of the information processing program of the present invention is an information processing program for causing a computer to execute information processing, the program causing the computer to execute processes including a transaction information acquisition process for acquiring transaction information including an entry describing a transaction at a store, a text separation process for separating transaction text from the transaction information acquired by the transaction information acquisition process, an attribute extraction process for inputting the transaction text separated by the text separation process into a plurality of natural language processing models and extracting a plurality of attributes corresponding to the transaction text, and an output process for outputting the plurality of attributes extracted by the attribute extraction process. Effect of the Invention

[0013] According to the present invention, information related to a user's purchasing behavior can be extracted easily and with high accuracy. 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. [Brief description of the drawings]

[0014] [Figure 1] FIG. 1 is a block diagram showing an example of a functional configuration of an attribute information extraction device according to each embodiment of the present invention. [Diagram 2] FIG. 2 is a conceptual diagram illustrating an example of various pieces of attribute information extracted from a card statement by the attribute information extraction device according to this embodiment. [Diagram 3] FIG. 3 is a conceptual diagram showing an example of transaction text written in multiple languages ​​on a card statement input to the attribute information extraction device according to the present embodiment. [Figure 4] FIG. 4 is a flowchart showing an example of an outline of the process steps of the attribute extraction process executed by the attribute information extraction device according to the present embodiment. [Diagram 5]FIG. 5 is a flowchart showing an example of a detailed procedure of the natural language processing in step S3 of FIG. 4, which is executed by the attribute extraction unit of the attribute information extraction device. [Figure 6] FIG. 6 is a conceptual diagram illustrating an example of a process executed by the attribute extraction unit of the attribute information extraction device to predict a category to which a store belongs using a text classifier based on the store name described in a transaction text. [Figure 7] FIG. 7 is a conceptual diagram illustrating an example of direct attributes and derived attributes respectively extracted from a Japanese transaction text. [Figure 8] FIG. 8 is a conceptual diagram illustrating an example of direct attributes and derived attributes respectively extracted from an English transaction text. [Figure 9] FIG. 9 is a diagram showing an example of a display output of a processing result of a user profiling process based on attribute information, which is executed by a post-processing unit of the attribute information extraction device. [Figure 10] FIG. 10 is a diagram showing an example of a display output of a processing result of business entity profiling processing based on attribute information, which is executed by a post-processing unit of the attribute information extraction device. [Figure 11] FIG. 11 is a block diagram showing an example of a hardware configuration of the attribute information extraction device according to this embodiment. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0015] Hereinafter, with reference to the attached drawings, an embodiment for carrying out the present invention will be described in detail. 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 an 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 the present embodiment are necessarily essential to the solution of the present invention.

[0016] The attribute information extraction device according to this embodiment acquires transaction information including entries describing transactions at a store based on credit card details or the like, separates transaction text from the acquired transaction information, and inputs the separated transaction text into multiple natural language processing models to extract multiple attributes corresponding to each of the transaction text.

[0017] Below, an example is described in which this embodiment uses a credit card statement as a single input, identifies the store where the credit card transaction was performed as a key segment from the transaction text, and extracts various attributes of the identified store, but this embodiment is not limited to this. This embodiment is applicable not only to credit cards, but also to other transaction means for which transaction details can be recorded, such as debit cards, virtual currencies, etc. Furthermore, what is identified as a key segment from the transaction text of the transaction details is not limited to the store where the transaction was performed, but may be any information that can identify the purchasing behavior of a user, such as the store location, the purchased product or service, etc.

[0018] <Functional configuration of the attribute information extraction device> FIG. 1 is a block diagram showing an example of a functional configuration of an attribute information extraction device 1 according to this embodiment. The attribute information extraction device 1 shown in FIG. 1 includes a transaction information acquisition unit 11, a transaction text separation unit 12, an attribute extraction unit 13, an attribute output unit 14, and a post-processing unit 15.

[0019] The attribute information extraction device 1 reads out the card statement 2 stored in the storage device, acquires transaction information from the data of the read out card statement 2, and separates transaction text from the acquired card transaction information. The attribute information extraction device 1 further performs natural language processing on the separated transaction text using a plurality of natural language processing models 3 to extract a plurality of corresponding attributes from the transaction text, and outputs the extracted plurality of attributes to an attribute information database 4. The output attributes are subjected to various post-processing by the attribute information extraction device 1, such as profiling of users and stores.

[0020] The attribute information extraction device 1 may be communicably connected to a client device (not shown) configured by a PC (Personal Computer) or the like via a network. In this case, the attribute information extraction device 1 may be implemented in a server, and the client device may provide a user interface when the attribute information extraction device 1 executes input / output of information with the outside, and may also include some or all of the components 11 to 15 of the attribute information extraction device 1.

[0021] The transaction information acquisition unit 11 reads out the card statement 2 stored in the storage device, acquires transaction information from the data of the read out card statement 2, and supplies the acquired transaction information to the transaction text separation unit 12.

[0022] The card statement 2 is a history of card usage such as a credit card over a specified period of time that is issued to each user. This card statement 2 includes entries that describe transactions at a store for the card for which the statement is issued, for a specified period of time, for example every month. Each entry corresponds to one card transaction that occurred at a store. The card statement 2 may include entries for both web-based transactions, which are purchases of goods and services at online stores, and real transactions, which are purchases of goods and services at real stores. Furthermore, as long as the use of the card for which the statement is issued is not limited to domestic use, the card statement 2 may include entries for both domestic and overseas card transactions. In other words, the card statement 2 may be written in multiple languages ​​(multilingual).

[0023] In this embodiment, the card statement 2 is used as the single source of information for attribute extraction. For example, in order to obtain a higher advertising effect for a certain user, it is important to grasp various attributes in the user's purchasing behavior. In order to obtain more attributes, information is usually collected from many different domains, such as travel websites, electronic commerce (EC) websites, financial transaction applications, health-related applications, food-related applications, restaurant reservation websites, etc. However, collecting attributes across domains through different interfaces is cumbersome and lacks flexibility.

[0024] In contrast, in this embodiment, the card statement 2 is basically used as a single information source for various attributes of a user's purchasing behavior. The card statement 2 is a collection of all card transactions that the user, who is the card owner, performed at multiple stores where the card can be used over a specified period of time. Therefore, it is possible to easily and centrally extract various attributes of a user's purchasing behavior. However, the present embodiment is not limited thereto, and the transaction information may additionally be obtained from other information, for example information obtained from other domains, such as other websites or applications.

[0025] The storage device that stores the card statement 2 may be configured as a non-volatile storage device such as a hard disk drive (HDD) or a solid state drive (SSD). The transaction information acquisition unit 11 may acquire transaction information by reading out the card statement 2 pre-stored in such a storage device, or may receive the card statement 2 via a communication I / F from the same or a different opposing device that stores the card statement 2.

[0026] The transaction information acquiring unit 11 also accepts input of various parameters necessary for executing the attribute extraction process in the attribute information extracting device 1. The transaction information acquiring unit 11 may accept input of various parameters via a user interface of a client device communicatively connected to the attribute information extracting device 1.

[0027] Here, transaction information is information extracted from card statement 2 that identifies a card transaction at a store, and may include fields such as the cardholder's user ID, the date and time the transaction occurred, the transaction amount, the name of the store where the transaction occurred, and a description of the purchased product or service (described below with reference to FIG. 2). Of these, the store name and transaction amount may be written as transaction text, which is a text string of a predetermined length. Like card statement 2, each entry of the transaction information corresponds to one card transaction that occurred at a certain store, and may be written in multiple languages ​​(multilingual). Here, the transaction text is a text string (character string, text information) with a limited number of characters, a text string with a limited number of language types (one or more of a plurality of language types, such as Japanese and English), and a text string with a limited number of character types (one or more of a plurality of character types, such as Katakana and the alphabet). The number of characters in the text string may be, for example, 5-9, 10-15, 16-20, 21-25, 26-30, etc. If the number of characters in the transaction text exceeds the upper limit of the number of characters for a text string, the portion exceeding the upper limit of the number of characters is deleted before being written and treated as transaction information.

[0028] The transaction text separator 12 separates the transaction text from the transaction information supplied from the transaction information acquirer 11 and supplies it to the attribute extractor 13 . The transaction text separation unit 12 may further supply information other than the transaction text from the transaction information supplied from the card statement 2, such as the cardholder's user ID, the date and time the transaction occurred, the transaction amount, etc., to the attribute extraction unit 13 together with the transaction text.

[0029] The attribute extraction unit 13 extracts attributes from the transaction text supplied from the transaction text separation unit 12, and supplies the extracted attributes to an attribute output unit 14. Specifically, the attribute extraction unit 13 uses a plurality of trained natural language processing models 3 to apply natural language processing to extract various attributes from the transaction text.

[0030] In this embodiment, an attribute refers to any information that is related to the purchasing behavior of a user who is a card owner and can characterize the purchasing behavior of the user. The attributes extracted by the attribute extraction unit 13 include direct attributes that are directly extracted from the transaction text and derived attributes that are derived from the direct attributes.

[0031] The attribute output unit 14 outputs the attributes supplied from the attribute extraction unit 13 to the attribute information database 4. The attribute information database 4 may be configured, for example, with a non-volatile storage device such as an HDD or SSD, or may be placed in a Network Attached Storage (NAS) connected to the attribute information extraction device 1 via a network. The attribute output unit 14 may also output and display the attributes stored in the attribute information database 4 to the outside on an arbitrary Graphic User Interface (GUI) via a display device or the like.

[0032] The post-processing unit 15 executes post-processing to arbitrarily process the attributes stored in the attribute information database 4. The post-processing executed by the post-processing unit 15 with reference to the attribute information database 4 may include, for example, user profiling and store profiling, the details of which will be described later with reference to Figs. 9 and 10.

[0033] FIG. 2 is a conceptual diagram illustrating an example of various attributes extracted from transaction information by the attribute extraction unit 13 of the attribute information extraction device 1. As shown in FIG. 2, the transaction information acquired by transaction information acquisition unit 11 has entries similar to those of card statement 2 input to transaction information acquisition unit 11, with each entry including fields for user ID 21 identifying the card owner, transaction text 22, transaction amount 23 indicating the amount of the card transaction, and transaction date and time 24 indicating the date and time of the card transaction. Transaction text 22 may include, for example, the name of the store where the card transaction occurred, and a description of the product or service purchased in the card transaction.

[0034] In this embodiment, the attribute extraction unit 13 of the attribute information extraction device 1 extracts the direct attributes 5 and the derived attributes 6 from the card statement 2. The attribute extraction unit 13 may further extract the card transaction identification information 7 from the card statement 2.

[0035] The direct attributes 5 are attributes that are extracted directly from the transaction text 22 by the attribute extraction unit 13 applying natural language processing using the natural language processing model 3 to the transaction text 22 . Referring to FIG. 2 , the direct attributes 5 include, for example, a business / brand 51 indicating the name of the store where the card transaction occurred, a category 52 of the business / brand 51, a subcategory 53 into which the business / brand 51 category 52 is subdivided, a transaction location 54 indicating the location or place where the card transaction occurred, a state / country 55 to which the transaction location 54 belongs, and a latitude / longitude 56 of the transaction location 54.

[0036] The derived attributes 6 are attributes derived from the direct attributes 5 by the attribute extraction unit 13 extracting information from external information sources such as websites on the Internet and public databases based on the direct attributes 5 . Referring to FIG. 2, the derived attributes 6 include, for example, information on branches 61 of the business / brand 51, such as the location of the branch, the size of the branch, etc., and information on competitors 62 competing with the business / brand 51, such as the location, size, competing products or services, and the degree of competition of the competitor.

[0037] The card transaction identification information 7 is information for identifying one card transaction, and is obtained from each field other than the transaction text 22 of the card statement 2, the user ID 21, the transaction amount 23, and the transaction date and time 24. The card transaction identification information 7 includes, for example, a user ID 71, a transaction amount 72, and a transaction date and time 73. The direct attributes 5, derived attributes 6, and card transaction identification information 7 extracted by the attribute extraction unit 13 are stored in the attribute information database 4 in association with each card transaction.

[0038] In this embodiment, a large number of types of attributes such as those shown in Figure 2 can be centrally and easily extracted from a small text string described in the card statement 2, particularly the transaction text 22, without accessing multiple information sources. The attribute extraction process executed by the attribute extraction unit 13 will be described in detail later with reference to FIG.

[0039] FIG. 3 is a conceptual diagram showing an example of transaction text written in multiple languages ​​on a card statement 2 input to the attribute information extraction device 1. As shown in FIG. 3, Japanese transaction 31 includes entries 311-313 corresponding to three card transactions, and English transaction 32 includes entries 321-323 corresponding to three card transactions.

[0040] Entry 311 for Japanese transaction 31 includes transaction text 22 "Toy Store & Baby Store Yokohama," transaction amount 23 "￥2,643.00," and transaction date and time 24 "15th Aug." Similarly, entry 312 for Japanese transaction 31 includes transaction text 22 "Toy Store & Baby Store Matsudo," transaction amount 23 "￥3,100.00," and transaction date and time 24 "22nd Aug," and entry 313 for Japanese transaction 31 includes transaction text 22 "Hotel Shima," transaction amount 23 "￥20,400.00," and transaction date and time 24 "25th Aug." Meanwhile, entry 321 of English transaction 32 includes transaction text 22 "Toys Store, Baby Store Tokorozawa", transaction amount 23 "￥2,100.00", and transaction date and time 24 "2nd Aug". Similarly, entry 322 of English transaction 32 includes transaction text 22 "ABC KOREA SHOE STORE", transaction amount 23 "￥6,589.00", and transaction date and time 24 "5th Aug", and entry 323 of English transaction 32 includes transaction text 22 "Burger Waikiki", transaction amount 23 "￥950.00", and transaction date and time 24 "20th Aug".

[0041] 3, the text string of a card transaction written on the card statement 2 must be written within a limited number of characters, so the end of the string may be missing, and the content may contain abbreviations and a mixture of full-width, half-width, and kana characters. For this reason, if the transaction text 22 obtained from the text string of a card transaction written on the card statement 2 is processed as written, the accuracy of extracting the direct attribute 5 may decrease, and the query for deriving the derived attribute 6 may also be incomplete. In this embodiment, multiple natural language processing models 3 are used to perform multiple natural language processing operations including fuzzy text matching on the transaction text 22, thereby appropriately complementing the incompleteness of such transaction text 22 and improving the accuracy of attribute extraction.

[0042] <Attribute Extraction Process Executed by Attribute Information Extraction Device 1> FIG. 4 is a flowchart showing an example of an outline of the procedure of the attribute extraction process executed by the attribute information extraction device 1 according to the present embodiment. Each step in FIG. 4 is realized by a CPU reading and executing a program stored in a storage device such as an HDD of the attribute information extraction device 1. At least a part of the flowchart shown in FIG. 4 may be realized by other hardware such as a GPU. When realized by hardware, for example, a specific compiler may be used to automatically generate a dedicated circuit on an FPGA (Field Programmable Gate Array) from a program for realizing each step. Also, a gate array circuit may be formed in the same manner as an FPGA to realize the hardware. Also, it may be realized by an ASIC (Application Specific Integrated Circuit).

[0043] In S1, the transaction information acquisition unit 11 of the attribute information extraction device 1 acquires transaction information from the card statement 2 stored in the storage device. The transaction information obtained here is information that identifies a card transaction at a store, extracted from the card statement 2, and includes fields for the cardholder's user ID 21, transaction text 22, transaction amount 23, and transaction date and time 24, as shown in Figure 2.

[0044] In S2, the transaction text separation unit 12 of the attribute information extraction device 1 separates the transaction text 22 from the transaction information acquired by the transaction information acquisition unit 11 in S1. The transaction text 22 is a text string including key segments for attribute extraction, and referring to Figures 2 and 3, is a text string that typically includes the name of the store where the card transaction occurred, a description of the purchased product or service, etc. Below, an example is described in which a store name (business / brand 51) in a text string is identified as a key segment, but this embodiment is not limited to this, and instead of a store name, a product or service name or other text segment may be identified as a key segment.

[0045] In S3, the attribute extraction unit 13 of the attribute information extraction device 1 applies natural language processing to the transaction text 22 separated in S2 using multiple trained natural language processing models 3 to extract various attributes including direct attributes 5 and derived attributes 6 from the transaction text 22. In this embodiment, the attribute extraction unit 13 uses multiple natural language processing models 3 to combine multiple types of natural language processing, thereby being able to extract a large number of types of attributes sufficient for profiling users and stores from the small text strings described in the card statement 2. The attribute extraction process executed by the attribute extraction unit 13 will be described in detail later with reference to FIG.

[0046] In S4, the attribute output unit 14 of the attribute information extraction device 1 outputs to the attribute information database 4 the attributes extracted from the transaction text 22 by the attribute extraction unit 13 in S3. In S5, the post-processing unit 15 of the attribute information extraction device 1 processes the attributes stored in the attribute information database 4 and performs any post-processing, such as profiling for each user or profiling for each store (business / brand).

[0047] <Details of the attribute extraction process executed by the attribute extraction unit 13> FIG. 5 is a flowchart showing an example of a detailed procedure of the natural language processing in step S3 of FIG. 4, which is executed by the attribute extraction unit 13 of the attribute information extraction device 1. It should be noted that the processes from S31 to S36 shown in FIG. 5 do not need to be executed sequentially, but may be executed simultaneously, and the order of execution may also be arbitrary.

[0048] In S31, the attribute extraction unit 13 identifies a key segment in the transaction text 22 and recognizes the identified key segment. Specifically, the attribute extraction unit 13 may identify, for example, a store name described in the transaction text 22 as a key segment. The store name usually includes the name of the business operator / brand 51 as well as the store location information as shown in Fig. 3. Therefore, in S31, the attribute extraction unit 13 can extract the attributes of the business operator / brand 51 and the transaction location 54 as direct attributes 5. In this embodiment, the segments of the transaction text 22 that can be identified as key segments are not limited to the store name. For example, other segments of the transaction text 22, such as the name of a product or service, may be identified as key segments. However, the following describes an example in which the store name is identified as a key segment.

[0049] Once the transaction text 22 has been separated in S2, in S31 the attribute extraction unit 13 may use, for example, the Natural Language ToolKit (NLTK) to identify the store name, i.e., the business / brand 51, and possibly the transaction location 54, as named entities in the transaction text 22 and recognize the identified named entities as key segments.

[0050] NLTK is a natural language processing library for executing natural language processing in the Python language, and has functions such as text classification, tokenization, stemming (i.e., text matching with the stem of a word whose form changes), part-of-speech tagging, text syntax analysis, morphological analysis, and semantic inference for transaction text 22. NLTK is language-dependent, and can be multilingual by having interfaces to multiple corpora corresponding to multiple natural languages, each of which is a structured and large-scale accumulation of sentences in each natural language, language-dependent add-on modules, etc. NLTK is implemented in a natural language processing model, and may be trained in advance using training data for text classification, analysis, etc.

[0051] In S32, the attribute extraction unit 13 searches for names that are similar to the key segment recognized in S31 by fuzzy text matching. Specifically, the attribute extraction unit 13 may search for text strings (names) that are semantically similar to the key segments recognized in S31, for example, using a natural language processing model that implements a Locally Sensitive Hashing (LSH) algorithm. LSH is a method for compressing high-dimensional data by probabilistic processing, and can be applied to measuring text similarity. The basic idea of ​​hashing is to organize data so that similar data is likely to fall into the same bucket. In many cases, the number of buckets is much smaller than the number of input data samples. A parameter set for performing LSH is called an LSH family, and the LSH family is defined by a distance space, a threshold, and a proximity factor.

[0052] The natural language processing model used in S32 may have learned a parameter set of the LSH family in advance. For example, this natural language processing model may implement MinHash, an LSH family that uses the Jaccard distance (coefficient) to measure the similarity between sets. The Jaccard distance is calculated by applying a random hash function to each vector in the set and finding the probability that the minimum hash value matches.

[0053] Alternatively, the natural language processing model used in S32 may implement Jaro Similarity, which measures partial matches between two strings using Jaro distance, which measures the similarity between two strings by quantifying the number of matching characters and transpositions between the two strings.

[0054] The attribute extraction unit 13 may further search for names that are phonetically similar to the key segments recognized in S31 using a natural language processing model that implements a phonetic algorithm in S32. As a phonetic analysis algorithm, for example, Soundex, which is one of the algorithms that derives an approximation of the pronunciation from the spelling of a character string, can be used. In S32, the attribute extraction unit 13 can extract the attributes of the business operator / brand 51 and the transaction location 54 as direct attributes 5, including the approximate names obtained by the above-mentioned fuzzy text matching and phonetic analysis.

[0055] At S33, the attribute extraction unit 13 predicts the category of the key segment recognized at S31 using a text classifier. Specifically, the attribute extraction unit 13 inputs, to a text classifier, particularly the business operator / brand 51 in the store name, which is the key segment recognized in S31, and causes the text classifier to predict the category of the business operator / brand 51. Alternatively, the attribute extraction unit 13 may input, to the text classifier, the product name or service name described in the transaction text 22, and predict the category of the business operator / brand 51, or may input both the business operator / brand 51 and the product name or service name into the text classifier and predict the category.

[0056] The attribute extraction unit 13 may predict the category of a business operator / brand 51 from the business operator / brand 51 by using, for example, a natural language processing model that implements embedding by Transformer or BERT (Bidirectional Encoder Representation from Transformers) as a text classifier. BERT is a pre-training model that uses bidirectional encoding representation by Transformer, and can be pre-trained using a large amount of unlabeled data.

[0057] FIG. 6 is a conceptual diagram illustrating an example of a process executed by the attribute extraction unit 13 of the attribute information extraction device 1 to predict a category to which a business / brand 51 described in a transaction text 22 belongs using a text classifier.

[0058] Referring to FIG. 6, the transaction text 22 has five transaction text 22 entries, each corresponding to five card transactions. Each transaction text 22 includes a key segment of the store name (business / brand 51). Each business / brand 51 entry is input to the text classifier 8, which predicts a category 52 and a subcategory 53. For example, the first entry, business / brand 51 "Rakuten Travel Domestic Accommodation" is classified into category 52 "Travel" and subcategory 53 "Rakuten Travel (domestic)". The second and subsequent entries are similarly classified into their respective categories 52 and subcategories 53. For each entry, a confidence score 57 indicates the confidence of the predicted category 52 and subcategory 53.

[0059] From the transaction text 22 “ABC Tunes” of the fourth entry, it can be seen that “Tunes” is extracted in S31 of Figure 5 as a key segment of business / brand 51, and that this extracted “Tunes” is classified in S33 into category 52 “Hobby & Entertainment” and subcategory 53 “Digital Content.”

[0060] Normally, to identify the category to which a store belongs from its store name, it is necessary to refer to the Merchant Category Code (MCC) assigned to the store name. In contrast, in this embodiment, a text classifier 8 that implements a natural language processing model 3 is used to predict the category 52 and subcategory 53 from the key segment of the store name included in the transaction text 22, so there is no need to associate the MCC with the store name in advance, and the degree of freedom in describing the card transaction on the card statement 2 is not restricted.

[0061] Returning to FIG. 5, in S34, the attribute extraction unit 13 identifies position information corresponding to the key segment recognized in S31. Specifically, when the store name or the like in the transaction text 22 includes information on the location where the card transaction occurred (for example, entry 311 in FIG. 3 ), the attribute extraction unit 13 may directly identify the location information as the transaction location 54 of attribute 5. On the other hand, when the transaction text 22 does not include information on the location where the card transaction occurred, the attribute extraction unit 13 may directly derive information on the location of the store from the store name recognized from the transaction text 22 as the transaction location 54 of attribute 5, for example, by accessing a corpus using NLTK.

[0062] In S 34 , the attribute extraction unit 13 further identifies location information of state (prefecture or province) / country 55 and latitude / longitude 56 corresponding to the transaction location 54 of the direct attribute 5 . Specifically, the attribute extraction unit 13 may identify the state / country 55 and latitude / longitude 56 corresponding to the transaction location 54, for example, by accessing a corpus, map information, or the like via NLTK.

[0063] In S35, the attribute extraction unit 13 extracts the direct attributes 5 identified or derived in S31 to S34. With reference to Figure 2, the direct attributes 5 extracted in S35 include business / brand 51, category 52, subcategory 53, transaction location 54, state / country 55, and latitude / longitude 56.

[0064] At S36, the attribute extraction unit 13 generates derived attributes 6 from the key segments recognised at S31. Specifically, the attribute extraction unit 13 may dynamically generate the derived attribute 6 using, for example, web scraping software that automatically extracts specific information from websites on the Internet. Web scraping has a web crawler that crawls websites on the Internet and a web scraper that removes unnecessary information from the HTML of the website and extracts necessary information, and can obtain information that is updated in real time from the crawled website.

[0065] The derived attributes 6 generated in S36 include a branch 61 and a competitor 62, with reference to FIG. 2. The branch 61 is information on the branch of the business / brand 51, and includes, for example, information on the location of the branch, the size of the branch, etc. The competitor 62 includes information on a competitor competing with the business / brand 51, such as the location, size, competing products and services, and the degree of competition of the competitor. These derived attributes 6 can obtain corresponding information from an external website by appropriately including the business / brand 51, category 52, subcategory 53, transaction location 54, state / country 55, and latitude / longitude 56 extracted as the direct attributes 5 in S35 in the web scraping query.

[0066] Once the derived attribute 6 is generated in S36, the process proceeds to S4 in FIG. 4, and the attribute output unit 14 outputs the direct attribute 5 extracted in S35 in FIG. 5 and the derived attribute 6 generated in S36 to the attribute information database 4.

[0067] FIG. 7 is a conceptual diagram illustrating an example of direct attributes 5 and derived attributes 6 respectively extracted from a Japanese transaction text 22. As shown in FIG. Referring to FIG. 7, direct attribute 5 and derived attribute 6 are extracted from Japanese transaction text 22 "Green Line Shin-Shizuoka Store." As direct attributes 5, the following are extracted: business / brand 51 “Green Line”, category 52 “Clothing Store”, subcategory 53 “Apparels”, transaction location 54 “Shizuoka”, state / country 55 “Japan”, and latitude / longitude 56 “34.9756” and “138.3827”. As the derived attributes 6, the branch offices 61 "within the same state: 10" and "within the same country: 25" and the competitors 62 "Blue Line", "Red Line", and "Yellow Line" are extracted.

[0068] On the other hand, FIG. 8 is a conceptual diagram illustrating an example of direct attributes 5 and derived attributes 6 extracted from an English transaction text 22. As shown in FIG. Referring to FIG. 8, direct attributes 5 and derived attributes 6 are extracted from the English transaction text 22 “DEFG bangalore 8”. As direct attributes 5, the following have been extracted: business / brand 51 “DEFG”, category 52 “ToyStore”, subcategory 53 “Children Games”, transaction location 54 “Bangalore”, state / country 55 “Karnataka”, “India”, latitude / longitude 56 “12.9716”, “77.5946” respectively. As the derived attributes 6, branch offices 61 "within the same state: 15" and "within the same country: 30" and competitors 62 "Toys Store" and "Pearl Toys" are extracted.

[0069] In addition, in Figure 8, the direct attribute 5 includes, as card transaction identification information 7, the transaction amount 72 ``420'' extracted from the transaction amount 23 field contained in the transaction information, and the transaction date and time 73 ``20th August 2022'' extracted from the transaction date and time 24 field. As can be seen from Figures 7 and 8, according to this embodiment, by applying multiple natural language processing processes using multiple trained natural language processing models 3 to the transaction text 22, which is a small text string, it is possible to easily and centrally extract a large number of types of attributes. 7 and 8, the direct attributes 5 and the derived attributes 6 are extracted in English, but this is not limited to this embodiment. In this embodiment, the attribute extraction unit 13 uses a natural language processing model 3 that is multilingual, so that both the direct attributes 5 and the derived attributes 6 can be output in any language that the natural language processing model 3 can support, regardless of the description language of the transaction text 22.

[0070] <Use case of attribute extraction processing> An example of post-processing as a use case executed by the post-processing unit 15 of the attribute information extraction device 1 will be described with reference to FIGS. FIG. 9 is a diagram showing an example of a display output of the processing result of user profiling processing based on the attribute information directly accumulated in the attribute information database 4, which is executed by the post-processing unit 15 of the attribute information extraction device 1. As shown in FIG. The post-processing unit 15 can read and analyze the attributes of a specific user who is the card owner over a specified period of time that are stored in the attribute information database 4, thereby visualizing the purchasing behavior of the user during that period.

[0071] Referring to FIG. 9 , the percentage of purchasing behaviors performed by the user through card transactions within a given period of time, categorized based on categories 52 and subcategories 53 of direct attributes 5 extracted from transaction text 22, is shown in a pie chart according to transaction amount 72. By selecting the Merchants tab in the upper right corner of Figure 9, it is also possible to visualize the proportion of purchasing behavior according to transaction amount 72 for each merchant / brand 51 of the direct attribute 5 extracted from the transaction text 22. In this way, according to this embodiment, it is possible to grasp the trends and preferences in the user's purchasing behavior in detail by profiling the purchasing behavior of the user based on the attribute information accumulated in the attribute information database 4. Therefore, it is possible to optimize advertisement delivery by, for example, giving priority to the recommendation display of businesses / brands and product / service categories that have a higher appeal to the user.

[0072] FIG. 10 is a diagram showing an example of a display output of the processing result of the profiling process of a store (business operator / brand) based on attribute information, which is executed by the post-processing unit 15 of the attribute information extraction device 1. The post-processing unit 15 reads out and analyzes the attributes of all users over a given period stored in the attribute information database 4, thereby making it possible to evaluate a plurality of businesses during that period. Referring to FIG. 10, based on the direct attribute 5 of the business / brand 51 extracted from the transaction text 22, the amount of purchasing behavior per business performed by all users via card transactions within a specified period is shown along with the number of card transactions. Furthermore, the branch office 61 and competitor 62 of the derived attribute 6 may be used to display a comparison between the business and its competitors.

[0073] In this way, according to this embodiment, it is possible to easily identify the business operators, brands, and store names that are most used in card transactions by profiling business operators and brands based on the attribute information accumulated in the attribute information database 4. Therefore, for example, by displaying the identified brand, etc. as a recommendation on the website of the credit card to be analyzed or on a related EC site, it is expected that the conversion rate (CVR) for the brand will be improved. Alternatively, the identified brand, etc. may be displayed as a banner on the website of the analyzed credit card, and the interest of card users may be increased by offering a credit card in collaboration with the identified brand, etc.

[0074] As described above, according to this embodiment, the attribute information extraction device reads out card usage details stored in a storage device, obtains transaction information from the read card usage details data, and separates transaction text from the obtained card transaction information. The attribute information extraction device further performs natural language processing on the separated transaction text using a plurality of natural language processing models to extract a plurality of corresponding attributes from the transaction text, and outputs the extracted plurality of attributes to an attribute information database. The output attributes are subjected to various post-processing processes, such as profiling of users and stores, by the attribute information extraction device.

[0075] This makes it possible to easily and accurately extract attribute information related to the purchasing behavior of a user, thereby enabling more accurate profiling of the purchasing behavior of users across a variety of domains.

[0076] <Hardware configuration of the attribute information extraction device> FIG. 11 is a diagram showing a non-limiting example of the hardware configuration of the attribute information extraction device 1 according to this embodiment. The attribute information extraction device 1 according to this embodiment can be implemented on a single or multiple computers, mobile devices, or any other processing platform. 11, an example in which the attribute information extraction device 1 is implemented in a single computer is shown, but the attribute information extraction device 1 according to this embodiment may be implemented in a computer system including multiple computers. The multiple computers may be connected to each other so as to be able to communicate with each other via a wired or wireless network.

[0077] 15, the attribute information extraction device 1 may include a CPU 111, a ROM 112, a RAM 113, a HDD 114, an input unit 115, a display unit 116, a communication I / F 117, and a system bus 118. The attribute information extraction device 1 may also include an external memory. The CPU (Central Processing Unit) 111 centrally controls the operation of the attribute information extraction device 1, and controls each component (112 to 117) via a system bus 118, which is a data transmission path. Instead of or in addition to the CPU 111, the attribute information extraction device 1 may include a GPU (Graphics Processing Unit), which may execute pre-learning and inference processing of the natural language processing model 3, etc.

[0078] The ROM (Read Only Memory) 112 is a non-volatile memory that stores control programs and the like necessary for the CPU 111 to execute processing. Note that the programs may be stored in a non-volatile memory such as an HDD (Hard Disk Drive) 114 or an SSD (Solid State Drive) or an external memory such as a removable storage medium (not shown). The RAM (Random Access Memory) 113 is a volatile memory and functions as a main memory, a work area, etc. of the CPU 111. That is, when executing a process, the CPU 111 loads necessary programs, etc. from the ROM 112 into the RAM 113 and executes the programs, etc. to realize various functional operations.

[0079] The HDD 114 stores, for example, various data and various information required when the CPU 111 performs processing using a program. The HDD 114 also stores, for example, various data and various information obtained when the CPU 111 performs processing using a program. The input unit 115 is composed of a keyboard and a pointing device such as a mouse. The display unit 116 is configured with a monitor such as a liquid crystal display (LCD). The display unit 116 may provide a GUI (Graphical User Interface) that is a user interface for inputting various parameters used in machine learning processing, communication parameters used in communication with other devices, and the like, to the attribute information extraction device 1.

[0080] The communication I / F 117 is an interface that controls communication between the attribute information extraction device 1 and an external device. The communication I / F 117 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 117. In this embodiment, the communication I / F 117 may execute communication via a wired LAN (Local Area Network) or a dedicated line conforming to a communication standard such as Ethernet (registered trademark). However, the network available in this embodiment is not limited to this, and may be configured as a wireless network. This wireless network includes wireless PAN (Personal Area Network) such as Bluetooth (registered trademark), ZigBee (registered trademark), and UWB (Ultra Wide Band). It also includes wireless LAN (Local Area Network) such as Wi-Fi (Wireless Fidelity) (registered trademark), and wireless MAN (Metropolitan Area Network) such as WiMAX (registered trademark). It also includes wireless WAN (Wide Area Network) such as LTE / 3G, 4G, and 5G. Note that the network only needs to connect each device so that they can communicate with each other and be capable of communication, and the communication standard, scale, and configuration are not limited to those described above.

[0081] At least some of the functions of each element of the attribute information extraction device 1 shown in Fig. 1 can be realized by the CPU 111 executing a program. However, at least some of the functions of each element of the attribute information extraction device 1 shown in Fig. 1 may be operated as dedicated hardware. In this case, the dedicated hardware operates under the control of the CPU 111.

[0082] Although specific embodiments have been described above, the embodiments are merely illustrative and are 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 appropriately omitted, substituted, and modified without departing from the scope of the present invention. Such omitted, substituted, and modified forms are included in the scope of the claims and their equivalents, and belong to the technical scope of the present invention.

[0083] (Embodiments of the present disclosure) The present disclosure includes the following embodiments. [1] An information processing device comprising: a transaction information acquisition unit that acquires transaction information including entries describing transactions at a store; a text separation unit that separates transaction text from the transaction information acquired by the transaction information acquisition unit; an attribute extraction unit that inputs the transaction text separated by the text separation unit into a plurality of natural language processing models and extracts a plurality of attributes corresponding to the transaction text; and an output unit that outputs the plurality of attributes extracted by the attribute extraction unit.

[0084] [2] The information processing device according to [1], wherein the attribute extraction unit identifies a key segment from the transaction text and extracts the multiple attributes from the identified key segment.

[0085] [3] The information processing device described in [2], wherein the attribute extraction unit recognizes the key segments by processing the transaction text using a trained natural language processing model that implements a natural language processing library among the plurality of natural language processing models.

[0086] [4] The information processing device described in [2] or [3], wherein the attribute extraction unit predicts a category of the identified key segment using a trained natural language processing model that implements a text classifier among the plurality of natural language processing models, and extracts the predicted category as one of the plurality of attributes.

[0087] [5] The information processing device described in any of [2] to [4], characterized in that the attribute extraction unit generates derived attributes derived from the key segment by accessing an external information source using the identified key segment as a query.

[0088] [6] The information processing device described in any one of [2] to [5], wherein the attribute extraction unit uses a trained natural language processing model among the plurality of natural language processing models that implements a fuzzy text matching algorithm to search for a text string that is semantically similar to the identified key segment, and extracts the plurality of attributes from the searched text string and the key segment.

[0089] [7] The information processing device described in any of [2] to [6], characterized in that the attribute extraction unit uses a trained natural language processing model that implements a phonetic analysis algorithm among the plurality of natural language processing models to search for a text string that is phonetically similar to the identified key segment, and derives the plurality of attributes from the searched text string and the key segment.

[0090] [8] The information processing device described in any one of [1] to [7], characterized in that the transaction information acquisition unit acquires the transaction information from a single transaction detail for a specified period.

[0091] [9] An information processing method executed by an information processing device, comprising: a step of acquiring transaction information including an entry describing a transaction at a store; a step of separating transaction text from the acquired transaction information; a step of inputting the separated transaction text into a plurality of natural language processing models to extract a plurality of attributes corresponding to the transaction text; and a step of outputting the extracted plurality of attributes.

[0092]

[10] An information processing program for causing a computer to execute information processing, the program causing the computer to execute processes including a transaction information acquisition process for acquiring transaction information including an entry describing a transaction at a store, a text separation process for separating transaction text from the transaction information acquired by the transaction information acquisition process, an attribute extraction process for inputting the transaction text separated by the text separation process into a plurality of natural language processing models and extracting a plurality of attributes corresponding to the transaction text, and an output process for outputting the plurality of attributes extracted by the attribute extraction process. [Explanation of symbols]

[0093] 1...attribute information extraction device, 2...card usage details, 3...natural language processing model, 4...attribute information database, 11...transaction information acquisition unit, 12...transaction text separation unit, 13...attribute extraction unit, 14...attribute output unit, 15...post-processing unit, 111...CPU, 112...ROM, 113...RAM, 114...HDD, 115...input unit, 116...display unit, 117...communication I / F, 118...system bus

Claims

1. a transaction information acquisition unit that acquires transaction information including entries describing transactions at a store; a text separation unit that separates a transaction text from the transaction information acquired by the transaction information acquisition unit; an attribute extraction unit that identifies a key segment including information on a business or brand indicating the name of the store where the transaction occurred from the transaction text separated by the text separation unit, and inputs the identified key segment into a plurality of natural language processing models to extract a plurality of attributes corresponding to the transaction text; an output unit that outputs the plurality of attributes extracted by the attribute extraction unit; Equipped with The attribute extraction unit Identifying the key segments as direct attributes by processing the transaction text using a trained natural language processing model that implements a natural language processing library capable of handling multilingual text, among the plurality of natural language processing models; searching for a text string that approximates the identified key segment as the direct attribute using a trained natural language processing model that implements a fuzzy text matching algorithm or a trained natural language processing model that implements a phonetic analysis algorithm among the plurality of natural language processing models; accessing an external information source using the direct attributes as a query, generating derived attributes from the direct attributes, the derived attributes including at least one of information on branches of the business or the brand and information on competitors of the business or the brand; extracting the plurality of attributes from the direct attributes and the derived attributes; 23. An information processing apparatus comprising:

2. The attribute extraction unit predicts a category of the identified key segment using a trained natural language processing model that implements a text classifier among the plurality of natural language processing models, and extracts the predicted category as one of the plurality of attributes.

2. The information processing apparatus according to claim 1,

3. The transaction information acquisition unit acquires the transaction information from a single transaction statement for a predetermined period.

2. The information processing apparatus according to claim 1,

4. An information processing method executed by an information processing device, obtaining transaction information including entries describing transactions at the store; separating transaction text from the obtained transaction information; Identifying a key segment including information on a business or brand indicating a store name where the transaction occurred from the separated transaction text, and inputting the identified key segment into a plurality of natural language processing models to extract a plurality of attributes corresponding to the transaction text, respectively; outputting the extracted attributes; Including, The step of extracting each of the plurality of attributes includes: Identifying the key segments as direct attributes by processing the transaction text using a trained natural language processing model that implements a natural language processing library capable of handling multilingual text, among the plurality of natural language processing models; searching for a text string that approximates the identified key segment as the direct attribute using a trained natural language processing model that implements a fuzzy text matching algorithm or a trained natural language processing model that implements a phonetic analysis algorithm among the plurality of natural language processing models; accessing an external information source using the direct attributes as a query, generating derived attributes from the direct attributes, the derived attributes including at least one of information on branches of the business or the brand and information on competitors of the business or the brand; extracting the plurality of attributes from the direct attributes and the derived attributes; 23. An information processing method comprising:

5. An information processing program for causing a computer to execute information processing, the program comprising: a transaction information acquisition process that acquires transaction information including entries describing transactions at the store; a text separation process for separating a transaction text from the transaction information acquired by the transaction information acquisition process; an attribute extraction process for identifying a key segment including information on a business or brand indicating the name of the store where the transaction occurred from the transaction text separated by the text separation process, and inputting the identified key segment into a plurality of natural language processing models to extract a plurality of attributes corresponding to the transaction text; an output process for outputting the plurality of attributes extracted by the attribute extraction process; The present invention is directed to a process including the steps of: The attribute extraction process includes: Identifying the key segments as direct attributes by processing the transaction text using a trained natural language processing model that implements a natural language processing library capable of handling multilingual text, among the plurality of natural language processing models; searching for a text string that approximates the identified key segment as the direct attribute using a trained natural language processing model that implements a fuzzy text matching algorithm or a trained natural language processing model that implements a phonetic analysis algorithm among the plurality of natural language processing models; accessing an external information source using the direct attributes as a query, generating derived attributes from the direct attributes, the derived attributes including at least one of information on branches of the business or the brand and information on competitors of the business or the brand; extracting the plurality of attributes from the direct attributes and the derived attributes; 2. An information processing program comprising:

Images