Question classification and answer creation processor system and answer creation method
The processor system improves ESG evaluation efficiency by decomposing and classifying questions, automating answer creation, and enhancing accuracy, addressing the burden and inconsistency issues in supplier responses.
Patent Information
- Application Number
- JP2023079967
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2023-05-15
- Publication Date
- 2024-11-27
AI Technical Summary
Suppliers face a heavy burden in responding to numerous and varied ESG evaluations, leading to low response rates due to the complexity and inconsistency of questionnaires from multiple evaluation agencies, which often require evidence data and differ in format and underlying standards.
A processor system that decomposes questions into smaller units, performs natural language processing to classify similar questions, and automatically generates answers by combining linked answers, reducing the burden and improving accuracy in corporate evaluations.
Enhances the accuracy of question classification and answer creation in ESG evaluations, thereby increasing supplier response rates and simplifying the evaluation process.
Smart Images

Figure 2024164473000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to a question classification / answer creation processor system and an answer creation method. [Background technology]
[0002] In recent years, the ESG (Environment, Social, Governance) investment market has been expanding. In the ESG investment market, investors evaluate companies based on ESG factors and decide where to invest. This is because companies that place importance on ESG are expected to have stable management in the future and have high growth potential.
[0003] In recent years, ESG evaluations, including environmental and child labor issues, are being conducted not only within a company but throughout its entire supply chain. For this reason, buyers are increasingly requesting ESG evaluations of business entities (also called suppliers) participating in the supply chain system.
[0004] The mainstream approach for buyers to conduct ESG evaluations of suppliers is to obtain information by conducting questionnaire surveys of suppliers.
[0005] On the other hand, answering questionnaire surveys places a heavy burden on suppliers. This is because the number of questions can reach dozens or even hundreds, and some questions require suppliers to attach evidence data to support their answers. In addition, there are multiple evaluation agencies, and each agency sets questions that are similar but not completely identical. Furthermore, in addition to the questionnaires from the evaluation agencies, buyers may create their own questions and ask suppliers to answer them. As a result, the response rate from suppliers tends to be low.
[0006] Buyers can enter into a contract with a specific evaluation organization to use the evaluation results of suppliers they do business with. However, it is not often the case that all of the suppliers they do business with participate in the evaluation scheme of the evaluation organization. Therefore, buyers evaluate suppliers by using multiple evaluation organizations in combination, or by using evaluation results in combination with questions that the buyer has created himself.
[0007] In addition, the tendencies of the questions in the surveys are such that the underlying standards are often the same for many standards such as ISO (International Organization for Standardization) 26000, ISO 14000 series, and the United Nations Global Compact, and the differences in the questions are often due to differences in resolution. Alternatively, while the underlying standards are common between surveys, there are often differences in the form of the questions. For example, depending on the survey, there may be differences between those asking about the presence or absence of goal setting and those asking about the results of the same goal.
[0008] In the field of natural language processing, feature extraction methods such as Bagof Words, TF-IDF, BM-25, and N-gram are commonly known as techniques for recognizing commonality in sentences.
[0009] There are many machine learning techniques, but commonly known techniques that are often used as classifiers in natural language processing include support vector machines, decision trees, and k-nearest neighbor methods.
[0010] A prior art technique for natural language processing is a question type learning device shown in Patent Document 1, which configures a highly accurate classifier for identifying question types using N-gram, a natural language processing technique, and Support Vector Machine, a machine learning technique.
[0011] In recent years, techniques such as Word2vec and BERT (Bidirectional Encoder Representations from Transformers) and combinations with artificial intelligence such as neural networks have also become widely known. [Prior art documents] [Patent documents]
[0012] [Patent Document 1] JP 2004-094521 A Summary of the Invention [Problem to be solved by the invention]
[0013] In this context, in order to facilitate the ESG evaluation of suppliers, it is necessary to provide methods to reduce the burden on suppliers to respond to corporate evaluations and to improve the response rate of suppliers. For example, assistance with preparing answers to questions in corporate evaluations could be provided.
[0014] An object of the present invention is to improve the accuracy of question classification and answer creation in corporate evaluation. [Means for solving the problem]
[0015] The present application includes a number of means for solving at least a part of the above problems, examples of which are as follows: A processor system according to one aspect of the present invention for solving the above problems includes a question decomposition processing unit that, upon receiving question information including at least one or more questions, decomposes an answer unit included in the question information into a number of first questions if the answer unit includes a number of questions, a question classification unit that classifies a first question as a similar question to a second question included in master data including at least one or more questions if the first question is similar to the second question, and an answer creation unit that generates display information including a result classified by the question classification unit.
[0016] In other words, the question is broken down into multiple single questions, natural language processing is performed, and answers to compound questions are automatically created by combining answers linked to the multiple single questions, thereby improving the accuracy of answer creation for compound questions. Effect of the Invention
[0017] According to the present invention, it is possible to provide a technique for improving the accuracy of categorizing questions and creating answers in corporate evaluations. Problems, configurations, and effects other than those described above will become apparent from the following description of the embodiments of the invention. [Brief description of the drawings]
[0018] [Figure 1] FIG. 1 is a diagram showing an example of the configuration of a question classification / answer creation processor system. [Diagram 2] FIG. 2 is a diagram illustrating a detailed configuration example of a processing unit of the question classification / answer creation processor system. [Diagram 3] FIG. 13 is a diagram illustrating an example of a process flow of question classification processing by categorization. [Figure 4] 13 is an example of a process flow of a question allocation process. [Diagram 5] A diagram showing an example of a detailed flow of question allocation processing using documentary evidence. [Figure 6] FIG. 13 is a diagram illustrating an example of a flow of a question decomposition process. [Figure 7] FIG. 13 is a diagram illustrating an example of a flow of a question and answer column merging process. [Figure 8] FIG. 11 is a diagram illustrating an example of a flow of resolution adaptation processing. [Figure 9] FIG. 13 is a diagram illustrating an example of a flow of a multiple-choice question answer process. [Figure 10] FIG. 13 is a diagram illustrating an example of a flow of an integrated evaluation question allocation process. [Figure 11] FIG. 13 is a diagram showing an example of a screen for presenting and correcting a proposed answer. [Figure 12] FIG. 2 illustrates an example of a hardware configuration of a processor system. [Figure 13] 13 is a diagram illustrating an example of a data structure of a question material storage area. [Figure 14] FIG. 4 is a diagram illustrating an example of a master data storage area. [Figure 15] FIG. 4 is a diagram illustrating an example of a data configuration of master data. [Figure 16] FIG. 13 is a diagram illustrating an example of a data structure of an answer history storage area. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0019] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The examples are illustrative for explaining the present invention, and are omitted and simplified as appropriate for clarity of explanation. The present invention can be implemented in various other forms. Unless otherwise specified, each component may be singular or plural.
[0020] In order to facilitate understanding of the invention, the position, size, shape, range, etc. of each component shown in the drawings may not represent the actual position, size, shape, range, etc. Therefore, the present invention is not necessarily limited to the position, size, shape, range, etc. disclosed in the drawings.
[0021] As examples of various information, the various information may be described using expressions such as "table", "list", and "queue", but the various information may be expressed using data structures other than these. For example, various information such as "XX table", "XX list", and "XX queue" may be expressed as "XX information". When describing identification information, expressions such as "identification information", "identifier", "name", "ID", and "number" are used, but these are mutually interchangeable. Furthermore, in the embodiments, the identification information described using these expressions is represented using symbols, numbers, natural language, or a combination thereof, but the identification information may be in a form other than these.
[0022] When there are multiple components having the same or similar functions, they may be described by using the same reference numerals with different subscripts, or when there is no need to distinguish between these multiple components, the subscripts may be omitted.
[0023] In the embodiments, a process performed by executing a program may be described. Here, a computer executes the program using a processor (e.g., CPU, GPU), and performs the process defined by the program while using a storage resource (e.g., memory) and an interface device (e.g., communication port), etc. Therefore, the subject of the process performed by executing the program may be the processor. Similarly, the subject of the process performed by executing the program may be a controller, device, system, computer, or node having a processor. The subject of the process performed by executing the program may be a calculation unit, and may include a dedicated circuit that performs specific processing. Here, the dedicated circuit is, for example, an FPGA (Field Programmable Gate Array), an ASIC (Application Specific Integrated Circuit), or a CPLD (Complex Programmable Logic Device), etc.
[0024] The program may be installed in the computer from a program source. The program source may be, for example, a program distribution server or a computer-readable storage medium. When the program source is a program distribution server, the program distribution server may include a processor and a storage resource that stores the program to be distributed, and the processor of the program distribution server may distribute the program to be distributed to other computers. In addition, in the embodiment, two or more programs may be realized as one program, and one program may be realized as two or more programs.
[0025] Although the present invention is a processor system, it may be realized as a platform having the functions of the present invention.
[0026] Specifically, possible methods for assisting in creating answers to questions in corporate evaluations include using the similarities between questions to reduce the burden of answering similar questions, and similarly using the similarities between questions to consolidate questionnaires from multiple different evaluation agencies, allowing a variety of questionnaires to be used in supplier evaluations.
[0027] For these methods to work correctly, the questions need to be classified correctly, but simply applying existing natural language processing technology may result in low classification accuracy. For example, sentences may be classified as similar if the number of words or word order in the sentences are the same. The following two sentences are classified as similar in natural language processing, but they are questions in different categories, "environment" and "labor," respectively, and it would be desirable to classify them as having different question contents. -What efforts does your company make to combat air pollution? -What is your company's approach to child labor?
[0028] Furthermore, the number of question elements included in one question does not necessarily match depending on the questionnaire distributor, and there are questions that ask about multiple contents (compound questions) and questions that ask about one content, so there is a problem that there is not necessarily a one-to-one correspondence. In this case, it is difficult to assist with answers or integrate questions if the questions remain in their original format.
[0029] Furthermore, it is necessary to have a function that improves the accuracy of question classification, appropriately processes questions, and creates answers, including cases where there are differences in the type of question, such as when asking about whether or not goals have been set as described above and when asking about the results of achieving those goals.
[0030] 1 is a diagram showing an example of the configuration of a question classification / answer creation processor system. For example, the question classification / answer creation processor system 10 is a question classification / answer creation processor system including a processor system 100, a network 50, an evaluation institution D computer 300, an evaluation institution E computer 310, a supplier A computer 400, a supplier B computer 410, a supplier C computer 420, a buyer F computer 800, a buyer G computer 810, and an evaluation requester H computer 850.
[0031] The network 50 is, for example, any one of a LAN (Local Area Network), a WAN (Wide Area Network), a VPN (Virtual Private Network), a communication network using a general public line such as the Internet in part or in whole, a mobile phone communication network, etc., or a combination of these networks. Note that the network 50 may be a wireless communication network such as Wi-Fi (registered trademark) or 5G (Generation).
[0032] Supplier A, Supplier B, and Supplier C are examples of suppliers that are entities that provide parts and products in a parts supply network such as a supply chain network. The number of suppliers is not limited to only three organizations, and there are usually many more. However, in the example of this embodiment, these three organizations are considered as suppliers for the sake of simplicity.
[0033] Evaluation institutions D and E are examples of institutions that evaluate suppliers, which are entities that provide parts and products, in a parts supply network such as a supply chain network. The number of evaluation institutions is not limited to two institutions, and there are usually many more. However, in this embodiment, these two institutions are considered as evaluation institutions for the sake of simplicity.
[0034] Buyer F and Buyer G are examples of buyers that purchase parts and products in a parts supply network such as a supply chain network. The number of buyers is not limited to only two organizations, and there are usually many more. However, in the example of this embodiment, these two organizations are considered as buyers to simplify the explanation. When managing and selecting suppliers, the buyer may use the evaluation organization, or may evaluate the supplier himself without using the evaluation organization. In the example of this embodiment, the supplier and the buyer are described separately to simplify the explanation, but when a supplier purchases a part or product, the supplier may become the buyer, and when another buyer purchases a part or product provided by the buyer, the buyer may become the supplier.
[0035] The evaluation requester H is an example of an evaluation requester that requests the processor system to evaluate a supplier for the purpose of managing and selecting suppliers. The evaluation requester may be, for example, a buyer, but the evaluation requester may also be someone other than a buyer, such as an investor. In the example of this embodiment, the evaluation requester and the buyer are described separately to simplify the explanation, but as described above, the evaluation requester may be a buyer.
[0036] Furthermore, by using some or all of the functions of this processor system, a company with multiple affiliated companies can evaluate and manage its own sustainability. In that case, the company in question can be treated as the evaluation requester H and its affiliated companies as suppliers, allowing for an internal ESG evaluation.
[0037] The processor system 100 is a system having one or more processors. The processor system 100 can also be called a question classification / answer creation processor system. The processor system 100 includes a memory unit 110, a processing unit 120, an input / output interface 130, and a transmission interface 140. The memory unit 110 includes a question material storage area 111, a master data storage area 112, and an answer history storage area 113. The processing unit 120 includes a question material receiving unit 121, a question classification / answer creation processing unit 122, and an answer receiving unit 127.
[0038] The question material receiving unit 121 receives and stores question materials, which are question information including at least one question, from an evaluation institution or a buyer. Specifically, the question material receiving unit 121 receives question materials from the evaluation institution D computer 300, the evaluation institution E computer 310, the buyer F computer 800, and the buyer G computer 810 via the transmission interface 140. In addition, the question material receiving unit 121 breaks down the received question materials into question units and stores them in the question material storage area 111 and the master data storage area 112.
[0039] The question classification and answer creation processing unit 122 includes a question preprocessing unit 123 , a question classification unit 124 , an answer creation unit 125 , and a learning and optimization unit 126 .
[0040] When the question information includes a hierarchical question, the question preprocessing unit 123 formats the question information so that one question sentence is set for each step of tracing the question structure down to the lowest level. Specifically, the question preprocessing unit 123 breaks down the question material, divides it into candidates for units that are formally considered questions, identifies sentences that are question units, and determines whether each sentence contains a combination of multiple questions.
[0041] The question classification unit 124 classifies the preprocessed first question into a predetermined category, identifies questions in the classified category that are similar to the first question and second questions in the master data, and classifies the first question as a similar question to the second question.
[0042] The answer creation unit 125 generates display information including the results classified by the question classification unit 124. Specifically, the answer creation unit 125 creates an answer proposal by transcribing the answer to the first question from the answer to the second question that is a similar question, i.e., the answer in the master data, and generates the display information. Note that, when the question is a multiple-choice question in which one option is selected from a finite number of options, the answer creation unit 125 selects the option that is closest to the answer in the master data as the answer proposal. In particular, when the question on the questionnaire to be supplemented is a question about non-financial information related to management and conforms to a predetermined international standard, the answer creation unit 125 tends to be able to present an appropriate answer proposal by transcribing from the answer in the master data.
[0043] When the learning and optimization unit 126 receives answers to question materials and evidence data that guarantees the answers, it learns information such as the data format and data type of the evidence data attached by the answerer to the questions of each evaluation agency, and updates the learning area of the answer creation unit 125.
[0044] The answer receiving unit 127 receives and stores the answers and evidential data of the answerer to each evaluation institution. Specifically, the answer receiving unit 127 receives answers to the question materials and evidential data that guarantees the answers from the answerer's computers such as the supplier A computer 400, the supplier B computer 410, and the supplier C computer 420 via the transmission interface 140. The answer receiving unit 127 breaks down the received answers and evidential data that guarantees the answers into question units and passes them to the learning / optimization unit 126 and the answer history storage area 113, respectively.
[0045] 2 is a diagram showing a detailed configuration example of the processing unit of the question classification / answer creation processor system. The question preprocessing unit 123 includes a question decomposition processing unit 1231 and a question / answer column combination processing unit 1232.
[0046] When an answer unit included in question information includes multiple questions, the question decomposition processing unit 1231 decomposes the answer unit into multiple questions. When the answer field of the question information is in a free description format and further includes sentences in the answer field, the question and answer field merging processing unit 1232 reconstructs the sentence obtained by merging the question and the sentence in the answer field as a question if an answer field is provided for each sentence. When an answer field is not provided for each sentence, the question and answer field merging processing unit 1232 reconstructs the sentence obtained by merging the question and all sentences in the answer field as a question.
[0047] The question classification unit 124 includes a classifier 1241 and an evidence data comparison unit 1247. The classifier 1241 includes a question categorization processing unit 1242, a plurality of sub-classifiers 1244, and integrated teacher data 1246. The question categorization processing unit 1242 includes a keyword dictionary 1243. Each of the sub-classifiers 1244 includes sub-teacher data 1245 according to a category and a model trained using the sub-teacher data 1245.
[0048] In the present embodiment, the number of sub-classifiers 1244 and the number of sub-teacher data 1245 are three for ease of explanation, but there may be two or more. Alternatively, the sub-classifiers 1244 and the sub-teacher data 1245 may be a single classifier and teacher data including multiple categories.
[0049] Also, in this figure, the sub-teacher data 1245 is shown taking supervised learning as an example, but there are cases where there is no sub-teacher data 1245. Cases where there is no sub-teacher data 1245 include cases where the model of the sub-classifier 1244 is formed by unsupervised learning, where learning is performed using only the integrated teacher data 1246, and where external trained data is used. The sub-teacher data 1245 may contain data used to create teacher data, or may be trained data that has already been trained using a group of questions linked to a specific category.
[0050] The evidence data comparison unit 1247 performs a process (described later in detail) of allocating questions of the current year to answer information of the master data by using documentary evidence of the answer of the respondent of the previous year or earlier.
[0051] The input / output interface 130 receives input and output of various data. Specifically, the input / output interface 130 receives input from the supplier regarding the question materials.
[0052] The transmission interface 140 communicates with other devices via the network 50. The other devices include an evaluation agency D computer 300, an evaluation agency E computer 310, a supplier A computer 400, a supplier B computer 410, a supplier C computer 420, a buyer F computer 800, a buyer G computer 810, and an evaluation requester H computer 850.
[0053] Fig. 3 is an example of a process flow of question classification processing by categorization. The question classification processing by categorization S100 is started when the processor system 100 receives a start instruction from the evaluation requester, supplier, etc., and there is a question that needs to be classified. Alternatively, the question classification processing by categorization may be started at a predetermined date and time (e.g., 6:00 a.m. every day) or at a predetermined interval (e.g., every 12 hours). The process flows shown in Fig. 4 and subsequent figures are also started by the same instruction as in Fig. 3, or at a predetermined timing, etc.
[0054] First, as a preliminary step of the question classification process S100 by categorization, the question material receiving unit 121 receives and stores question materials from the evaluation institution or the buyer. Specifically, the question material receiving unit 121 receives question materials from the evaluation institution D computer 300, the evaluation institution E computer 310, the buyer F computer 800, and the buyer G computer 810 via the transmission interface 140. The question material receiving unit 121 breaks down the received question materials into question units and passes the questions to the question classification / answer creation processing unit 122.
[0055] When question classification / answer creation processing unit 122 receives a question, it performs a predetermined preprocessing in question preprocessing unit 123 as necessary, and then passes the question to question classification unit 124.
[0056] Then, the question classification unit 124 accepts and reads a question (step S101). Then, the question categorization processing unit 1242 determines whether the question includes a keyword registered in the keyword dictionary 1243, and if the keyword is included, the question is categorized by passing it to a sub-classifier 1244 linked to the keyword (step S102). Each sub-classifier 1244 is linked to a keyword group related to a specific category. Therefore, each sub-classifier 1244 collects questions classified into a specific category. For example, a sub-classifier 1244 linked to a keyword group such as "forced labor" and "child labor" collects questions related to the category "labor."
[0057] The sub-classifier 1244 is linked to the master data stored in the master data storage area 112 in the storage unit 110. The sub-classifier 1244 classifies the received question. Specifically, the sub-classifier 1244 determines the similarity between the received question and a group of questions in the master data stored in the master data storage area 112 linked to the sub-classifier 1244, and classifies the question within the category of the group of questions with the highest similarity (step S103). At this time, if the sub-classifier 1244 has sub-teacher data 1245, the sub-teacher data 1245 is used as teacher data to classify the question.
[0058] Then, the answer creating unit 125 transcribes the answer information in the master data as the answer to the question (step S104). Specifically, the answer information of the respondent is linked to the questions in the master data in advance, and when the question received by the sub-classifier 1244 is classified into one of the questions in the master data, the answer creating unit 125 transcribes the answer information linked to the question as the answer information of the received question. Then, the answer creating unit 125 updates the master data storage area 112, and causes the learning and optimization unit 126 to use the success or failure of the category classification as teacher data to train the sub-classifier 1244.
[0059] At this time, each sub-classifier 1244 is configured to collect questions classified into a specific category by the question categorization processing unit 1242. Therefore, the questions of the master data stored in the master data storage area 112 linked to each sub-classifier 1244 are questions related to each category. In this embodiment, the category is determined depending on whether the question contains a keyword or not, but depending on the question material (questionnaire), each question may be created by dividing it into categories such as "environment" and "labor and human rights" in advance. In such a case, the questions may be classified into the sub-classifier 1244 collectively using the category name included in the questionnaire instead of each question.
[0060] The above is an example of the process flow of the question classification process S100 by categorization. According to the question classification process S100 by categorization, the questions in the question document are classified into a predetermined category, and questions similar to each question in the master data are identified, and answer information of the master data is transcribed. In addition, by classifying into categories, the accuracy of answers can be improved. For example, a question such as "Please tell us about your company's efforts against air pollution" and a question such as "Please tell us about your company's efforts against child labor" are similar in question text, but require different answers. Therefore, by classifying the questions into categories in advance, the answers to be transcribed are appropriate. Conversely, if the categories are not clearly defined, there is a risk that answers to the same master data questions will be transcribed to both questions.
[0061] 4 shows an example of the process flow of the question allocation process. In the question allocation process S200, the evidence data comparison unit 1247 is used to allocate questions using evidence documents to provide answer support. First, learning processing for the previous year or earlier is performed, and then answer support processing for the current year is performed.
[0062] As a learning process for the previous year or earlier, the answer receiving unit 127 receives and stores answers and evidential data from registered users (the respondent and all respondents) such as suppliers to each evaluation agency (step S201). Specifically, the answer receiving unit 127 receives answers to question materials and evidential data that guarantees the answers from the supplier A computer 400, the supplier B computer 410, and the supplier C computer 420 via the transmission interface 140. The answer receiving unit 127 breaks down the received answers and evidential data that guarantees the answers into question units, and passes them to the learning / optimization unit 126 and the answer history storage area 113, respectively. The answer history storage area 113 stores the passed answers and evidential data as answer history.
[0063] Then, when the learning / optimization unit 126 receives the answers to the question materials and the evidence data that guarantees the answers, it learns information such as the data format and data type of the evidence data attached by the registered user to the question of each evaluation institution, and updates the learning area (step S202).
[0064] Then, the question classification unit 124 causes the classifier 1241 to link the questions of each evaluation organization with the questions of the master data. The evidence data comparison unit 1247 assigns information such as the format and data type of the evidence data to the questions of the master data (step S203). The above is an example of the learning process for the previous year or earlier.
[0065] Next, an example of the current year response support process is shown. First, information on the evaluation agencies that have already responded in the current year is received from the supplier. Here, an example in which information on the evaluation agency D is received is used for explanation. The response receiving unit 127 receives information on the evaluation agency D that has already responded in the current year from the supplier, specifically, the question materials of the evaluation agency D for the current year, the supplier's answers to each question in the question materials of the evaluation agency D, and the attached evidence data that guarantees the answers to each question, via the transmission interface 140 (step S204). The response receiving unit 127 breaks down the received answers and the evidence data that guarantees the answers into question units, and transfers them to the learning and optimization unit 126, the master data storage area 112, and the response history storage area 113, respectively. The response history storage area 113 stores the transferred answers and evidence data as response history.
[0066] Then, the question categorizing unit 124 causes the classifier 1241 to link the question of the evaluation institution D with the question of the master data (step S205). At this time, the question categorizing unit 124 similarly links the answer and the attachment as well as the question of the evaluation institution D.
[0067] Then, the question material receiving unit 121 receives the unanswered questions from evaluation institution E for the current year via the transmission interface 140 (step S206). The question material receiving unit 121 breaks down the received question material into question units and transfers them to the question material storage area 111 and the master data storage area 112. At this time, there are two possible cases for the questions from evaluation institution E: they may be received directly from evaluation institution E, or they may be received from a supplier. An example will be shown in which the questions are received from evaluation institution E. However, even when they are received from a supplier, there is no difference in the basic operation.
[0068] Then, the question classification unit 124 causes the classifier 1241 to link each of the questions of the evaluation agency E to each question in the master data as a similar question, and causes the answer creation unit 125 to create an answer proposal in which the answers to the evaluation agency D for the current year and the attached documents in the master data are assigned as answer information to the questions of the evaluation agency E (step S207). At this time, the attachment document assignment process is performed by the evidence data comparison unit 1247.
[0069] Then, the response receiving unit 127 transmits the proposed response to the supplier via the transmission interface 140 (step S208).
[0070] Then, the answer receiving unit 127 receives the corrections to the proposed answer, the corrected answer, and evidence data from the supplier via the transmission interface 140 (step S209). The answer receiving unit 127 stores the corrected answer and evidence data as an answer history in the answer history storage area 113, and updates the master data storage area 112 and the learning and optimization unit 126 (step S210).
[0071] Then, the answer receiving unit 127 transmits the answer and evidence data to the evaluation agency E via the transmission interface 140 (step S211). The above is an example of the flow of the answer support process for the current year. The question allocation process S200 makes it possible to use documentary evidence with the evidence data comparison unit 1247 to allocate questions and provide answer support. Next, an example of the detailed flow of step S207 will be described with reference to FIG. 5.
[0072] 5 is a diagram showing an example of a detailed flow of the question allocation process using documentary evidence. The detailed question allocation process using documentary evidence is an example of a process flow in step S207 of the question allocation process S200.
[0073] First, the question material receiving unit 121 breaks down the question material (questionnaire) of the evaluation institution E into question units, and transfers them to the question material storage area 111 and the master data storage area 112 (step S2071).
[0074] Then, the question classification unit 124 causes the classifier 1241 to determine the similarity between each question of the evaluation organization E and each question of the master data, and extracts a predetermined number of questions from the master data in descending order of similarity (step S2072).
[0075] Then, the answer creation unit 125 determines whether there is a question similar to the question to be classified among the questions from the previous year or earlier stored in the answer history memory area 113 of the evaluation institution E, and whether the question is linked to evidential data (step S2073).
[0076] If there is a question similar to the question to be classified and evidence data is linked to that question (if "Yes" in step S2073), the evidence data comparison unit 1247 compares the format and type of evidence data attached to the answers to each highly similar question extracted from the master data with the format and type of evidence data attached to similar questions from evaluation institution E in the previous year or earlier, and extracts from the master data the question with the most similar format and type of evidence data (step S2074).
[0077] On the other hand, if there are no questions similar to the question to be classified, or if there are no questions linked to evidence data ("No" in step S2073), the evidence data comparison unit 1247 extracts the question that was determined to have a high similarity in step S2072 as the most likely candidate (step S2075).
[0078] Then, the evidence data comparison unit 1247 assigns the answer information of evaluation institution D for the current year, which is a similar question of the extracted master data, to the question of evaluation institution E for the current year (step S2076). Note that, although the most similar question is extracted in steps S2074 and S2075, and the answer information is assigned to the question in step S2076, it is also possible to arrange multiple candidate questions in order of similarity and present multiple proposals to the supplier.
[0079] The above is a diagram showing an example of a flow of detailed question allocation processing using documentary evidence. According to the detailed question allocation processing using documentary evidence, questions similar in format and type to evidence data can be allocated as similar questions.
[0080] 6 is a diagram showing an example of the flow of the question decomposition process. The question decomposition process S300 is a process in which, after receiving question materials, if a question is a compound question in which multiple questions are combined, the compound question is decomposed, answers to each question are acquired by transcription from master data or the like, and the answers to each question are merged again to answer the compound question sentence before decomposition. First, the question preprocessing unit 123 reads the question (step S301). Specifically, the question preprocessing unit 123 decomposes the question materials, divides them into candidates for units that are formally considered to be questions, and identifies sentences that are units of questions.
[0081] Then, the question pre-processing unit 123 determines whether or not each of the identified question units has a complex content of multiple questions (step S302). If the question does not have a complex content ("No" in step S302), the question pre-processing unit 123 advances the control to step S304.
[0082] If the question has complex content (if "Yes" in step S302), the question decomposition processor 1231 decomposes the identified question unit sentence into multiple questions (step S303). In this case, to determine whether the question has complex content and to decompose the complex question into single questions, for example, a method of learning division rules by periods or specific particles in advance, or a method using a technique such as syntax analysis may be used.
[0083] Then, the question classification unit 124 judges the similarity between the single question and each question in the master data, and assigns the single question to a question having a high similarity (step S304).
[0084] Then, the answer creation unit 125 links the answer information to one compound question according to the number of simple questions that are decomposed (step S305). Specifically, since each question in the master data has a unique ID that is a combination of alphanumeric characters and symbols and is managed, the answer creation unit 125 assigns the master data ID to each simple question, and the answer creation unit 125 assigns multiple master data IDs to compound questions, thereby linking the answer information.
[0085] Then, the answer creation unit 125 combines the multiple linked answer information to automatically create an answer proposal for the compound question (step S306). If the question is not a compound question (if "No" in step S302), after performing step S304, the answer creation unit 125 creates an answer proposal using the answer information linked to the question as an answer (step S307).
[0086] After creating the proposed reply, the reply receiving unit 127 transmits the proposed reply to the supplier via the transmission interface 140 (step S308).
[0087] Then, the response receiving unit 127 receives the corrected response, the content of the correction to the proposed response, and evidence data from the supplier via the transmission interface 140 (step S309).
[0088] The answer receiving unit 127 stores the received corrected answers, corrections to the proposed answers, and evidence data as answer history in the answer history storage area 113, and also stores them in the master data storage area 112. The answer receiving unit 127 instructs the learning / optimization unit 126 to update the learning area of the answer creating unit 125 with information such as the data format and data type of the evidence data attached by the answerer to the question of each evaluation institution.
[0089] The above is an example of the flow of the question decomposition process. According to the question decomposition process, for a compound question, answers to the decomposed questions are retrieved and merged to create an answer proposal.
[0090] 7 is a diagram showing an example of the flow of the question and answer section merging process. The question and answer section merging process S400 is a process that assumes that multiple answer sections are prepared for one question in the question material (questionnaire), and questions (hierarchical questions) are also included in the answer section of the questionnaire. In such a case, classification of questions using only the content of the formal question portion makes it impossible to provide an appropriate answer in each answer section. Therefore, the question and answer section merging process S400 restructures the questions according to the answer section.
[0091] First, the question / answer box merge processor 1232 reads a question, and determines whether or not there are multiple answer boxes for each of the identified sentences in the question (step S401).
[0092] If there are multiple answer fields (if "Yes" in step S401), the question / answer field merger processor 1232 separates the answer fields by the number of answer fields, and merges the items entered in each answer field with the questions to create multiple questions (step S402).
[0093] If there are not multiple answer fields (No in step S401), the question and answer field merger processor 1232 merges the answer field entry and the question to create a question (step S403).
[0094] Then, the answer creating unit 125 determines the similarity of each question in the master data for each of the questions generated by merging, and classifies questions with high similarity as similar questions (step S404).
[0095] Then, the answer creating unit 125 extracts and transcribes answer information associated with questions with high similarity in the master data as answer proposals for each question generated by merging (step S405). At this time, the same number of answer information as the number of answer columns is extracted.
[0096] Then, for questions that do not have multiple answer fields, the answer creating unit 125 extracts and transcribes answer information linked to questions with high similarity in the master data as answer suggestions (step S406).
[0097] After creating the proposed reply, the reply receiving unit 127 transmits the proposed reply to the supplier via the transmission interface 140 (step S407).
[0098] Then, the response receiving unit 127 receives the corrected response, the content of the correction to the proposed response, and evidence data from the supplier via the transmission interface 140 (step S408).
[0099] The answer receiving unit 127 stores the received corrected answers, corrections to the proposed answers, and evidence data as answer history in the answer history storage area 113, and also stores them in the master data storage area 112. The answer receiving unit 127 instructs the learning / optimization unit 126 to update the learning area of the answer creating unit 125 with information such as the data format and data type of the evidence data attached by the answerer to the question of each evaluation institution.
[0100] The above is an example of the flow of the question and answer column merging process. According to the question and answer column merging process S400, when multiple answer columns are prepared for one question in the question material (questionnaire) and questions (hierarchical questions) are also included in the answer columns of the questionnaire, the questions can be reconfigured according to the answer columns. For example, when the answer columns of the question information are in a free description format and further sentences are included in the answer columns, if an answer column is provided for each sentence, the sentences obtained by merging the question and the sentences in the answer columns can be regarded as questions and reconfigured to create a proposed answer.
[0101] 8 is a diagram showing an example of the flow of resolution adaptation processing. The resolution adaptation processing S500 is a process for creating an answer proposal by inferring from the answer to a question of a different resolution when an answer is not linked to the question of the master data classified as the question with the highest similarity.
[0102] First, an example will be shown in which learning processing is performed for the previous year or earlier, and then answer creation processing is performed for related questions for the current year.
[0103] As a learning process for the previous year or earlier, the learning and optimization unit 126 registers questions with different resolutions in the master data with different IDs (step S501).Then, the learning and optimization unit 126 links questions with different resolutions based on the meaning of the question, and registers questions with higher concepts as higher-level questions and questions with lower concepts as lower-level questions (step S502).
[0104] The above is an example of the learning process flow from the previous year or earlier. For example, a question of a lower concept to a question of a higher concept, "Are you taking steps to reduce CO2 emissions?" could be "Please tell us the results of your efforts to reduce CO2 emissions." If the answer to such a question of a lower concept includes a description of the results, it can be assumed that the efforts themselves are being carried out, and the answer to the question of the above concept will be an affirmative answer. In this way, the learning process from the previous year or earlier can link questions with different resolutions, which can improve the rate at which answers can be created for questions.
[0105] Next, an example of the process of creating answers from related questions is shown. First, the question material receiving unit 121 receives questions from evaluation institution E that have not been answered in the current year via the transmission interface 140. The question material receiving unit 121 breaks down the received question material into question units and transfers them to the question material storage area 111 and the master data storage area 112 (step S503). At this time, there are two possible cases for the questions from evaluation institution E: they may be received directly from evaluation institution E, or they may be received from a supplier. An example will be shown in which the questions are received from evaluation institution E. However, even if they are received from a supplier, there is no difference in the basic operation.
[0106] Then, the question classification unit 124 causes the classifier 1241 to link each of the questions of the evaluation institution E to each question of the master data as a similar question (step S504). Specifically, the classifier 1241 determines the similarity between the questions of the evaluation institution E and each question of the master data, and classifies questions with high similarity as similar questions.
[0107] Then, the answer creation unit 125 determines whether or not an answer has been entered in the similar question in the master data for each question of the evaluation institution E by referring to the master data storage area 112 and the answer history storage area 113 (step S505). If an answer has been entered in the similar question in the master data ("No" in step S505), the answer creation unit 125 transcribes the answer in the master data as a proposed answer for the question (step S510).
[0108] If no answer is entered for the similar question in the master data (if "Yes" in step S505), the answer creation unit 125 judges whether the question is a higher-level question (step S506). Specifically, if a lower-level question is registered for the question, the answer creation unit 125 judges that the question is a higher-level question.
[0109] If the question is not a higher-level question (step S506: No), the answer creating unit 125 determines that the answer information is not available in the processor system and requests the supplier to provide an additional answer (step S509). Specifically, the answer creating unit 125 does not complement the proposed answer.
[0110] If the question is a higher-level question (step S506: Yes), the answer creation unit 125 determines whether or not the answer to the related lower-level question is in the master data storage area 112 and the answer history storage area 113 (step S507). If there is no answer to the related lower-level question (step S507: No), the answer creation unit 125 advances control to step S509. If there is an answer to the related lower-level question (step S507: Yes), the answer creation unit 125 uses the answer to the lower-level question to automatically create a proposed answer to the higher-level question (step S508).
[0111] Then, the reply receiving unit 127 transmits the proposed reply to the supplier via the transmission interface 140 (step S511).
[0112] Then, the answer receiving unit 127 receives the corrections to the proposed answer, the corrected answer, and evidence data from the supplier via the transmission interface 140 (step S512). The answer receiving unit 127 stores the corrected answer and evidence data as an answer history in the answer history storage area 113, and updates the master data storage area 112 and the learning and optimization unit 126 (step S513).
[0113] The above is an example of the flow of the process of creating an answer from a related question. The process of creating an answer from a related question makes it possible to create a proposed answer using answers to questions at different resolutions and provide answering support.
[0114] 9 is a diagram showing an example of the flow of multiple choice question answer processing. Multiple choice question answer processing S600 is an example of processing for creating an answer proposal for a multiple choice question in which the answer to the question is selected from a finite number of options. Multiple choice question answer processing S600 is processing for appropriately selecting an answer from the options when a similar question in the master data has an answer in free text, i.e., when the answer cannot be transcribed as is, for a multiple choice question in which the answer to the question is selected from a finite number of options.
[0115] First, the question material receiving unit 121 receives unanswered questions from evaluation institution E for the current year via the transmission interface 140. The question material receiving unit 121 breaks down the received question material into question units and transfers them to the question material storage area 111 and the master data storage area 112 (step S601). At this time, there are two possible cases for the questions from evaluation institution E: they may be received directly from evaluation institution E, or they may be received from a supplier. An example will be shown in which the questions are received from evaluation institution E. However, even when they are received from a supplier, there is no difference in the basic operation.
[0116] Then, the question classification unit 124 causes the classifier 1241 to link each of the questions of the evaluation institution E to each question of the master data as a similar question (step S602). Specifically, the classifier 1241 determines the similarity between the questions of the evaluation institution E and each question of the master data, and classifies questions with high similarity as similar questions.
[0117] Then, the answer creation unit 125 determines whether the answer field is a multiple-choice question such as a binary YES / NO answer or a 1 to N scale answer for each question of the evaluation organization E (step S603). If the question is not multiple-choice ("No" in step S603), the answer creation unit 125 transcribes the answer from the master data as a proposed answer for the question (step S606).
[0118] If the question is a multiple choice question (Yes in step S603), the answer creating unit 125 merges each option of the question with the question text and regards it as a question for each option (step S604).
[0119] Then, the classifier 1241 judges the similarity between the sentences of the answers linked to the similar questions in the master data and the sentences of the questions obtained by merging each option of the question with the question. The answer creation unit 125 selects an option with a high similarity as an answer proposal and writes it in the answer column (step S605). At this time, the reason for merging the option and the question rather than the option's text itself is to avoid cases where an answer cannot be created with the option alone. For example, the options may be "YES", "NO", "implemented", "not implemented", etc., and the meaning may not be clear by themselves. However, step S604 is not necessarily required, and a method of using the option's text as it is for classification is also conceivable.
[0120] Alternatively, the answer creation unit 125 may compare the answer of the master data with each option and select the most similar option as the answer proposal, or may compare the answer of the master data with a sentence in which each option is merged with the original question text and select the most similar option as the answer proposal. Also, the answer creation unit 125 may compare a sentence in which the question of the master data and the answer of the master data are merged with each option and select the most similar option as the answer proposal, or may compare a sentence in which the question of the master data and the answer of the master data are merged with a sentence in which each option is merged with the original question text and select the most similar option as the answer proposal.
[0121] Then, the reply receiving unit 127 transmits the proposed reply to the supplier via the transmission interface 140 (step S607).
[0122] Then, the answer receiving unit 127 receives the corrections to the proposed answer, the corrected answer, and evidence data from the supplier via the transmission interface 140 (step S608). The answer receiving unit 127 stores the corrected answer and evidence data as an answer history in the answer history storage area 113, and updates the master data storage area 112 and the learning and optimization unit 126 (step S609).
[0123] The above is an example of the flow of the multiple-choice question answer process. The multiple-choice question answer process makes it possible to provide answer support by creating an answer plan for a multiple-choice question in which the answer to the question is selected from a finite number of options.
[0124] The above-described categorization process S100, question allocation process S200, question decomposition process S300, question and answer column merging process S400, resolution adaptation process S500, and multiple choice answer process S600 can each function independently to create answer proposals, but by combining some or all of them, it is possible to handle question materials that include a variety of question combinations.
[0125] For example, the question material receiving unit 121 may break down the received question material into question units, and when each broken down question has multiple answer fields, apply the processes of steps S401 to S406 of the question answer field combining process S400, and when the questions are compound, apply the processes of steps S301 to S308 of the question breaking down process S300. The question material receiving unit 121 may further apply steps S101 to S104 of the question classification process S100 by categorization, and apply the processes of steps S201 to S207 of the question allocation process S200. In addition, the question material receiving unit 121 may apply the processes of steps S601 to S606 of the multiple choice question answer processing S600 if the question is a multiple choice question, and may apply the processes of steps S501 to S510 of the resolution correspondence processing S500 inferred from the answers to the questions at different resolutions if the answers are not linked to the questions in the master data.
[0126] The above is an example of the processing flow of each function of the question classification / answer creation processor system 10. The question classification / answer creation processor system 10 can reduce the burden on respondents to answer questions on ESG questionnaires and the like. Therefore, the burden on suppliers to answer questions in corporate evaluations can be reduced. In this case, the similarity of sentences in each flowchart may be determined by, for example, methods such as Bagof Words, TF-IDF, BM-25, N-gram, Word2vec, BERT (Bidirectional Encoder Representations from Transformers), machine learning methods such as Support Vector Machine, decision tree, k-nearest neighbor method, and artificial intelligence such as neural network.
[0127] In addition, each function of the above-mentioned question classification / answer creation processor system 10 is assumed to be used as a method of reducing the burden of answering similar questions by utilizing the similarities between questions, as an example to succinctly and simply explain the technical features of the invention, but is not limited to this.
[0128] For example, it can be used in a method (integrated evaluation) in which multiple questionnaires answered by suppliers are integrated to utilize a variety of responses in supplier evaluation. Note that when multiple questionnaires are integrated to utilize a variety of questionnaires in supplier evaluation, the flow may differ in part from the above-mentioned flowchart.
[0129] Fig. 10 is a diagram showing an example of the flow of the integrated evaluation question allocation process. The integrated evaluation question allocation process S220 is an example of the flow when used in a method (integrated evaluation) in which a variety of questionnaires are utilized for supplier evaluation by integrating multiple questionnaires from a viewpoint similar to the question allocation process using the evidence documents shown in Fig. 4 and Fig. 5.
[0130] First, the response receiving unit 127 receives information on the evaluation agency D that has already responded this year from the supplier A (step S221). Specifically, the response receiving unit 127 receives, as the information on the evaluation agency D that has responded this year, the question materials of the evaluation agency D this year, the answers of the supplier A to each question in the question materials of the evaluation agency D, and evidence data that guarantees the answers to each question, via the transmission interface 140. Similarly, the response receiving unit 127 receives information on the evaluation agency E (different from the evaluation agency D) that has already responded this year from the supplier B (step S222). Specifically, the response receiving unit 127 receives, as the information on the evaluation agency E that has responded this year, the question materials of the evaluation agency E this year, the answers of the supplier B to each question in the question materials of the evaluation agency E, and evidence data that guarantees the answers to each question, via the transmission interface 140.
[0131] Then, the question categorization unit 124 causes the classifier 1241 to link the questions of the evaluation institution D and the questions of the evaluation institution E according to the similarity (step S223). At this time, the question categorization unit 124 similarly links the answers and attachments in addition to the questions of the evaluation institution D and the evaluation institution E.
[0132] Then, the evidence data comparison unit 1247 compares the format and type of evidence data of supplier B attached to the answers to the highly similar questions extracted from evaluation institution E with the format and type of evidence data of supplier A attached to the similar questions from evaluation institution D, and extracts the most similar questions (step S224). Note that, although this example shows an example of classifying questions without using master data, the evidence data may be compared after the answers of supplier A and supplier B are respectively assigned to the master data.
[0133] Alternatively, the format and type of evidential data may be machine-learned for each question of the master data, and when the answers of supplier A and supplier B are assigned to the master data, the assignment may be made according to the similarity of the evidential data. In this case, data from other companies may be used to learn the format of the evidential data to be linked to the master data. Usually, the evidential data itself is confidential information and may not be used as learning data. However, in the present invention, the learning content is not the contents of the data but only the format and type of the evidential data, so that it can be used as learning data without using confidential information. The above is an example of a flow when the question assignment process using the evidential documents using the evidential data comparison unit 1247 is used for the integrated evaluation. By performing question assignment in this manner, it is possible to align the evaluation axis in the integrated evaluation, making it easier to compare multiple suppliers.
[0134] 11 is a diagram showing an example of an answer proposal presentation / modification screen. In the above example, an answer proposal is sent to the supplier in each of steps S208, S308, S407, S511, and S607. However, the answer proposal for a question is not limited to one, and it is also possible to present multiple answer proposals when there are multiple similar questions. An example of an answer proposal presentation / modification screen 500 shows an example of a screen with a function for presenting one or more answer proposals, a function for correcting the presented answers, and a function for extracting questions that need to be corrected depending on the question similarity score.
[0135] The classification result list screen 501 includes a master data question 501 a , an original question 501 b , an answer 501 c , a score 501 d , a selection button 502 , an answer candidate output button 503 , and a score range selection area 504 .
[0136] The master data question 501a is a question in the master data that is most similar to the original question 501b. The original question 501b is an unanswered question that requires an answer. The answer 501c is an answer linked to the master data question 501a. The score 501d is a predetermined index value that indicates the similarity between the master data question 501a and the original question 501b. In FIG. 11, the score 501d shows an example in which the larger the value, the higher the similarity, that is, the master data question 501a and the original question 501b are similar, but depending on the definition of the similarity, it is also possible that the smaller the value, the more similar the similarity, or the closer to a specific value, the more similar the similarity.
[0137] The selection button 502 accepts the selection of an answer that needs to be corrected. The answer candidate output button 503 outputs other answer candidates for the answer related to the selection accepted by the selection button 502. The score range selection area 504 accepts the range of index values indicated in the score 501d. Questions having a score within the range accepted by the score range selection area 504 can be excluded from the questions to be corrected. FIG. 11 shows an example in which questions having a score within the range accepted by the score range selection area 504 are excluded from the questions to be corrected, and only questions whose answers need to be corrected are displayed.
[0138] On the classification result list screen 501, the respondent, such as a supplier, selects and inputs the answer that needs to be corrected from the questions for which suggested answers are provided, using selection button 502, and when he or she inputs an answer candidate output button 503, the screen transitions to answer candidate output screen 505.
[0139] The answer candidate output screen 505 includes answer candidates 501e, a similarity score 501d, and a selection button 506 for receiving an answer selection from the answer candidates 501e.
[0140] When input is received, the select button 506 transitions the screen to an answer confirmation / correction screen 507. The answer confirmation / correction screen 507 includes an answer correction field 508 and a confirm button 509. The answer content previously written in the answer candidate 501e is transcribed into the answer correction field 508. When input is received, the confirm button 509 accepts the content entered in the answer correction field 508 as answer correction information, reflects it in the answer 501c, and closes the answer candidate output screen 505 and the answer confirmation / correction screen 507.
[0141] Here, the score range selection area 504 shows an example of extracting questions that need to be corrected, but is not limited to this. For example, for questions that are outside the range of values received in the score range selection area 504, the answer may not be automatically transcribed (the proposed answer may be discarded).
[0142] 12 is a diagram showing an example of the hardware configuration of a processor system. The question classification / answer creation processor system 10 can be realized by a general computer 900 having a processor (e.g., a central processing unit (CPU) or a graphics processing unit (GPU)) 901, a hardware memory 902 such as a random access memory (RAM), an external storage device 903 such as a hard disk drive (HDD) or a solid state drive (SSD), a reading device 905 for reading information from a portable storage medium 904 such as a compact disk (CD) or a digital versatile disk (DVD), an input device 906 such as a keyboard, a mouse, a barcode reader, or a touch panel, an output device 907 such as a display, and a communication device 908 for communicating with other computers via a communication network such as a LAN or the Internet, or a network system having a plurality of such computers 900. The reading device 905 may be capable of not only reading but also writing to the portable storage medium 904.
[0143] The processor 901 executes various processes by executing various predetermined programs loaded from the external storage device 903 to the memory 902. The programs are, for example, application programs that can be executed on an OS (Operating System) program. The programs may be installed in the external storage device 903 from a portable storage medium 904 via a reading device 905, or may be downloaded from a network via a communication device 908 and executed by the processor 901.
[0144] For example, the question material receiving unit 121, the question classification / answer creation processing unit 122, and the answer receiving unit 127 can be realized by loading programs stored in the external storage device 903 into the memory 902 and executing them with the processor 901. The input / output interface 130 can be realized by the processor 901 using the input device 906, the output device 907, and the communication device 908. The storage unit 110 can be realized by the processor 901 using the memory 902 or the external storage device 903. The transmission interface 140 can be realized by the processor 901 using the communication device 908.
[0145] 13 is a diagram showing an example of a data structure of a question material storage area. The question material storage area 111 stores information on questions for suppliers. Specifically, the question material storage area 111 has an issuing agency ID 111a, a material name 111b, a response target period 111c, a responding supplier ID 111d, and a question 111e. The issuing agency ID 111a, the material name 111b, the response target period 111c, the responding supplier ID 111d, and the question 111e are associated with each other.
[0146] The issuing agency ID 111a stores information for identifying the issuing agency of the question, which is identification information for identifying the issuing agency of the question. In this embodiment, the issuing agencies are the evaluation agencies D and E for non-financial information questionnaires and CSR questionnaires, and the issuing agency is the buyer F for SAQs (self-assessment questionnaires). However, the self-assessment questionnaire may be provided by the evaluation agency, and the buyer may issue the non-financial information questionnaire and CSR questionnaire.
[0147] The document name 111b stores the document name of the document in which the questions are written. The document name may differ depending on the issuing organization, but in this embodiment, it is a "non-financial information questionnaire," a "CSR (Corporate Social Responsibility) questionnaire," a "self-assessment questionnaire," etc. However, it is not limited to these, and generally, it is sufficient if it requests responses about non-financial information such as ISO26000, ISO14000 series, etc. Furthermore, many of these documents include questions issued by evaluation organizations or buyers with the supplier as the respondent.
[0148] The answer target period 111c includes information that specifies the period to which the document containing the questions is to be answered. Since many evaluation institutions require suppliers to answer the results for the previous year once a year, information that specifies the previous year is stored in the answer target period 111c. However, if question documents are issued with a different frequency, information that specifies the period (first half, second half, first quarter, etc.) corresponding to the answer target period is stored in the answer target period 111c.
[0149] The replying supplier ID 111d stores information for identifying the entity that will reply to the material identified by the material name 111b.
[0150] The question 111e stores a question sentence (natural language, index, or formula) included in the material specified by the material name 111b. In Fig. 13, there are two questions for the sake of simplicity, but in reality there are several to several hundred questions.
[0151] 14 is a diagram showing an example of a master data storage area. The master data storage area 112 includes a responding supplier ID 112a, a responding period 112b, an issuing institution ID 112c, and a master data ID 112d.
[0152] The answering supplier ID 112a stores information for identifying the entity providing the answer. The answering period 112b includes information for identifying the period to be answered. The issuing institution ID 112c stores information for identifying the issuing institution of the question, which is identification information for identifying the issuing institution of the question. The master data ID 112d includes information for identifying the master data, described later, created for each supplier in each answering period 112b.
[0153] The master data storage area 112 stores, for each supplier, the relationship between the master data ID 112d created in each response period 112b and the issuing institution ID 112c of the evaluation institution or buyer that set the questions used when creating the master data ID. For example, the master data ID 112d in the response period 112b of the record in which the responding supplier ID 112a is "Supplier B" is "2019", is created by the answer to the question created by the institution in which the issuing institution ID 112c is "Evaluation institution D".
[0154] FIG. 15 is a diagram showing an example of the data configuration of the master data. The master data 114 exists for each supplier and for each response period. The master data 114 is data in which a category 114a, a reference ID 114b, a question 114c, a response 114d, evidence data 114e, and a response data ID 114f are associated with each other. The category 114a indicates the category to which the question 114c belongs. For example, the category "E" is a category related to the environment. The reference ID 114b is information that is associated one-to-one with the feature vector of the reference question. The question 114c is a question from an evaluation institution or a buyer that has a feature vector corresponding to the reference ID 114b. The response 114d is a response from a supplier that is logarithmic to the question identified by the question 114c.
[0155] The evidence data 114e is information for identifying data that serves as evidence associated with the answer 114d. The answer data ID 114f is the same as the answer 113d in the answer history storage area 113, and is linked to the question and answer from the evaluation institution or the buyer.
[0156] The master data 114 is a data in which the questions and answers that the supplier answers to questions from one or more evaluation institutions or buyers during the response period are associated with a criterion ID 114b associated with a standard question. In other words, if a question that asks the same content as the standard question is included in the questions from the evaluation institution or buyer, the question and its answer are associated as question 114c and answer 114d, respectively, in the row of the criterion ID 114b associated with the standard question. On the other hand, if a question that asks the same content as the standard question is not included in the questions from the evaluation institution or buyer, the question 114c and answer 114d in the row of the criterion ID 114b are left blank.
[0157] FIG. 16 is a diagram showing an example of the data structure of the response history storage area. The response history storage area 113 stores the response data of the supplier by issuing institution and response period. Specifically, the response history storage area 113 has a responding supplier ID 113a, a question 113b, a response period 113c, and a response 113d. The response data of the response 113d includes the response data finally given to the issuing institution and the attached evidence data. However, in the case where there are several responses, such as resubmissions in the same period, the response 113d may include a history of the responses.
[0158] The above is an example of the question classification / answer creation processor system 10 according to an embodiment of the present invention. Note that the present invention is not limited to the above-mentioned example, and various modified examples are included. For example, the above-mentioned embodiment has been described in detail to clearly explain the present invention, and is not necessarily limited to having all of the configurations described. It is possible to replace part of the configuration of an embodiment with another configuration, and it is also possible to add the configuration of another embodiment to the configuration of an embodiment. It is also possible to delete part of the configuration of an embodiment.
[0159] The above-mentioned units, configurations, functions, processing units, etc. may be realized in part or in whole by hardware, for example by designing them as integrated circuits. Also, the above-mentioned units, configurations, functions, etc. may be realized in software by a processor interpreting and executing a program that realizes each function. Information such as the programs, tables, files, etc. that realize each function can be stored in a memory, a recording device such as a hard disk, or a recording medium such as an IC card, an SD card, or a DVD.
[0160] The control lines and information lines in the above-described embodiment are those considered necessary for the explanation, and do not necessarily show all the control lines and information lines in the product. In reality, it may be considered that almost all the components are connected to each other. The present invention has been described above with a focus on the embodiment. [Explanation of symbols]
[0161] 10: question classification / answer creation processor system, 50: network, 100: processor system, 110: memory unit, 111: question material storage area, 112: master data storage area, 113: answer history storage area, 120: processing unit, 121: question material reception unit, 122: question classification / answer creation processing unit, 123: question preprocessing unit, 124: question classification unit, 125: answer creation unit, 126: learning / optimization unit, 127: answer reception unit, 130: input / output interface, 140: transmission interface, 300: evaluation institution D computer, 310: evaluation institution E computer, 400: supplier A computer, 410: supplier B computer, 420: supplier C computer, 800: buyer F computer, 810: buyer G computer, 850: evaluation requester H computer.
Claims
1. a question decomposition processing unit that, when receiving question information including at least one question, decomposes the answer unit into a plurality of first questions when an answer unit included in the question information includes a plurality of questions; a question classification unit that classifies a first question as a similar question to the second question when a second question included in master data including at least one question and the first question are similar to each other; an answer creation unit that generates display information including the results classified by the question classification unit; A processor system comprising:
2. 2. The processor system of claim 1, the question classification unit includes at least one or more classifiers associated with at least one or more keywords and at least one or more of the second questions for each predetermined category, and causes the classifier for the category in which the keyword is included to determine a similarity between the first question and the second question; 23. A processor system comprising:
3. 3. The processor system according to claim 1, the master data includes an answer associated with each of the second questions; the answer creation unit creates an answer proposal for the first question by using the answer associated with the second question determined to be the similar question, and if the first question is one that has been decomposed by the question decomposition processing unit, further generates an answer proposal by combining the answer proposals for the answer units; 23. A processor system comprising:
4. 3. The processor system according to claim 1, the master data includes an answer associated with each of the second questions; the answer creation unit creates an answer proposal for the first question by using the answer associated with the second question determined as the similar question, and when the first question is a question to be answered by selecting from a finite number of options, sets the option similar to the answer proposal as the answer proposal; 23. A processor system comprising:
5. 3. The processor system according to claim 1, The master data includes answers and associated data sets associated with each of the second questions; the question decomposition processing unit classifies the first question as a similar question to the second question when the first question requires a predetermined dataset to be attached to the answer and the requested dataset is similar to the dataset associated with the second question; 23. A processor system comprising:
6. 6. The processor system according to claim 5, the question decomposition processing unit classifies the first question as a similar question to the second question when a data format or a data type of the requested data set is similar to the data set associated with the second question; 23. A processor system comprising:
7. 3. The processor system according to claim 1, the question decomposition processing unit, when the first question requires a plurality of answers, combines each of the required answers with the question and divides the first question; 23. A processor system comprising:
8. 3. The processor system according to claim 1, the master data includes an answer associated with each of the second questions; the answer creation unit creates a proposed answer to the first question by using the answer associated with the second question determined to be the similar question, and when there is no answer associated with the second question, generates a proposed answer by using the answer associated with another second question related to the second question; 23. A processor system comprising:
9. 3. The processor system according to claim 2, the question classification unit causes the classifier of the category to which the first question belongs in the question information to determine a similarity between the first question and the second question; 23. A processor system comprising:
10. 3. The processor system according to claim 1, a learning and optimization unit that receives input information related to the correctness of the results classified by the question classification unit and performs learning related to the determination of similarity; A processor system comprising:
11. 3. The processor system according to claim 1, the question classification unit, when the first question is classified as a similar question to a plurality of the second questions, presents the plurality of similar second questions in a manner that allows a selective input to be accepted, and accepts any selective input to set the similar question; 23. A processor system comprising:
12. 3. The processor system according to claim 1, the question classification unit calculates and presents a score indicating a degree of similarity between the second question and the first question for each of the second questions, and accepts an input of a range of the scores and classifies questions having scores within the range as the similar questions.
23. A processor system comprising:
13. 3. The processor system according to claim 1, the question classification unit calculates and presents a score indicating a degree of similarity between the second question and the first question for each of the second questions, and receives an input of a range of the scores and excludes questions having a score outside the range from the similar questions.
23. A processor system comprising:
14. 3. The processor system according to claim 1, At least one of the second questions of the master data is a question about non-financial information related to management; 23. A processor system comprising:
15. A response method using a processor system, comprising: The processor system includes: a question decomposition process for decomposing an answer unit into a plurality of first questions when answer information including at least one question is received and an answer unit included in the question information includes a plurality of questions; a question classification step of classifying a first question as a similar question to the second question when a second question included in master data including at least one question and the first question are similar to each other; an answer creation step of generating display information including the results classified by the question classification step; The answering method comprising:
Citation Information
Patent Citations
Method for providing company information
JP2002140615A
Query answering device and method
JP2014229275A
Inquiry responding support apparatus and inquiry responding support system
JP2018181033A
Questionnaire processing device, method and program
JP2019144983A
Question / answer display server, question / answer display method and question / answer display program
JP2021015599A
Cited By
Answer generating system, answer generating device and program
JP2025039443A
Information processing program, information processing device, and information processing method
JP2026090954A
Information processing systems, information processing methods, and programs
JP2026128900A
Information processing program, information processing device, and information processing method
JP7635952B1
Answer generation system, answer generation apparatus, and program
JP7708449B2
