Information processing device and information processing program

Abstract

To present, for a question sentence that includes a plurality of intents, as many answers as the number of intents included in the question sentence.SOLUTION: An information processing device 1 classifies a plurality of answers to a question sentence that includes a plurality of intents into a plurality of groups on the basis of the first degree of similarity that indicates the degree of similarity between answers, and presents, for each of the classified groups, an answer that corresponds to each intent included in the question sentence from among the plurality of answers, using a divided expression of the question sentence that is obtained by combining clauses of the question sentence and a second degree of similarity that indicates the degree of similarity to each answer included in the group.SELECTED DRAWING: Figure 3

Description

Technical Field

[0001] The present disclosure relates to an information processing apparatus and an information processing program.

Background Art

[0002] Patent Document 1 discloses a question-and-answer system including a storage unit that stores question-and-answer patterns each of which is a pair of a question sentence and a corresponding answer, a reception unit that receives an input of the question sentence from a user, a calculation unit that calculates a similarity between the question sentence received by the reception unit and the question sentence stored in the storage unit, an acquisition unit that acquires the question-and-answer pattern from the storage unit based on the similarity calculated by the calculation unit, and a presentation unit that presents the question-and-answer pattern acquired by the acquisition unit to the user. When there are a plurality of question-and-answer patterns including the same answer, the acquisition unit acquires only the question-and-answer pattern having the highest similarity among them from the storage unit. When there are a plurality of question-and-answer patterns including the same answer, the acquisition unit performs a positive correction on the similarity of the question-and-answer pattern having the highest similarity among them, and acquires the question-and-answer patterns from the storage unit in descending order of the corrected similarity.

[0003] Patent Document 2 describes a question answering program that has a comment text storage unit that stores a large number of comment texts and an answer storage unit that stores a large number of answers, and causes the computer to function to extract answers to user questions, comprising: a question text input means for inputting a question text; a question keyword extraction means for extracting multiple question keywords contained in the question text; a comment text search means for searching for comment texts containing the question keywords using the comment text storage unit; a topic classification means for classifying the searched multiple comment texts into multiple topic groups based on the distribution of predicate-argument structures by predicate-argument structure analysis; a group of comment texts included in each topic group; and each answer. A question answering program is disclosed that causes a computer to function as an answer output means, which calculates the similarity between a given text and a given text and associates answers with a predetermined threshold for each topic group; a representative predicate-argument structure extraction means for extracting a representative predicate-argument structure that characterizes the topic group from among the predicate-argument structures included in the associated answers for each topic group; a differential predicate-argument structure extraction means for extracting a representative predicate-argument structure that appears only in that topic group as a differential predicate-argument structure for each topic group; and an answer output means for explicitly displaying the answers detected by the answer detection means together with one or more sentences based on the corresponding differential predicate-argument structures.

[0004] Patent Document 3 discloses an automated response system that includes an answer inference unit that uses a learning model to infer a group of answers corresponding to the query in a ranked format, where each combination of a query and a group of answers is given a feature quantity, and the answer inference unit notifies the user of at least a portion of the ranked group of answers inferred by the learning model. [Prior art documents] [Patent Documents]

[0005] [Patent Document 1] Patent No. 6770283 [Patent Document 2] Patent No. 6007088 [Patent Document 3] Japanese Patent Publication No. 2021-56580 [Overview of the Initiative] [Problems that the invention aims to solve]

[0006] For example, in an FAQ (Frequently Asked Questions) search, if you search for answers to a question that contains multiple possible meanings, several answers may be presented. Some of these answers may simply contain words used in the question, so the user needs to find the answer they are looking for from among the multiple answers presented.

[0007] This disclosure aims to provide an information processing device and an information processing program that can provide as many answers as there are intentions in a question that contains multiple intentions. [Means for solving the problem]

[0008] The information processing device according to the first embodiment includes a processor, which classifies multiple answers to a question sentence containing multiple intentions into multiple groups based on a first similarity score indicating the similarity between each answer, and for each group, uses a second similarity score indicating the similarity between the question sentence and each answer included in the group, to present an answer from among the multiple answers that corresponds to each intention contained in the question sentence.

[0009] In the information processing device according to the second embodiment, the processor presents the classification results of the multiple answers along with the answers corresponding to each intention contained in the question.

[0010] The information processing device according to the third embodiment is an information processing device according to the second embodiment in which the processor further presents the number of intentions contained in the question.

[0011] The information processing device according to the fourth embodiment, in the information processing device according to the first embodiment, the processor identifies for each group a first answer which is the answer to which the second similarity of the segmented expression of the question sentence that has the largest decrease difference in the second similarity of the question sentence has been associated, or a second answer which is the answer to which the highest second similarity among the second similarities of each segmented expression of the question sentence has been associated, and presents them as answers corresponding to each intention contained in the question sentence.

[0012] The information processing device according to the fifth embodiment, in the information processing device according to the fourth embodiment, if the redundancy of the question statement, which indicates that the value increases as the expression becomes more redundant, is greater than or equal to a predetermined value, the processor presents the second answer as an answer corresponding to each of the intentions contained in the question statement.

[0013] The information processing program according to the sixth embodiment is a program that causes a computer to classify multiple answers to a question sentence containing multiple intentions into multiple groups based on a first similarity score indicating the similarity between each answer, and for each group, to present an answer from among the multiple answers that corresponds to each intention contained in the question sentence, using the question sentence and a second similarity score indicating the similarity between the segmented representation of the question sentence obtained by combining two or more phrases of the question sentence and each answer included in the group. [Effects of the Invention]

[0014] According to the first and sixth embodiments, the effect is that, for a question containing multiple intentions, it is possible to provide as many answers as there are intentions contained in the question.

[0015] According to the second embodiment, the effect is that the user can be notified of the reason for the estimation of the number of intentions contained in the question.

[0016] According to the third aspect, there is an effect that the number of intentions included in the question sentence recognized by the information processing apparatus can be notified to the user.

[0017] According to the fourth aspect, there is an effect that a method for specifying an answer to be presented to the user can be selected according to the characteristics of the question sentence.

[0018] According to the fifth aspect, there is an effect that an answer closer to the intention of the question sentence can be presented as compared with the case of presenting the first answer as the answer to the question sentence.

Brief Description of Drawings

[0019] [Figure 1] It is a diagram showing an example of the functional configuration of an information processing apparatus. [Figure 2] It is a diagram showing an example of the main part configuration of the electrical system of an information processing apparatus. [Figure 3] It is a flowchart showing an example of the flow of answer search processing. [Figure 4] It is a diagram showing an example of classification of answers to a question sentence. [Figure 5] It is a diagram showing an example of division of a question sentence. [Figure 6] It is a diagram showing an example of calculation for calculating the similarity between the answer to a question sentence, the question sentence, and the divided expression of the question sentence. [Figure 7] It is a diagram showing an example of selecting an answer according to a selection rule. [Figure 8] It is a diagram showing an example of selecting an answer according to another selection rule. [Figure 9] It is a diagram showing an example of display of an answer to a question sentence. [Figure 10] It is a diagram showing an example of display showing the number of intentions included in a question sentence. [Figure 11] It is a diagram showing an example of a classification screen showing the classification result of an answer to a question sentence.

Modes for Carrying Out the Invention

[0020] The embodiments of the disclosure will be described below with reference to the drawings. The same reference numerals will be used throughout the drawings for the same components and processes, and redundant explanations will be omitted.

[0021] Figure 1 shows an example of the functional configuration of the information processing device 1 according to this embodiment. The information processing device 1 is a device that takes a question sentence expressed in natural language as input and outputs an answer to the input question sentence in natural language. The question sentence according to this embodiment contains multiple intentions.

[0022] The information processing device 1 includes functional units such as a search unit 2, a classification unit 3, a division unit 4, a similarity determination unit 5, a selection unit 6, and an output unit 7, as well as a response DB 8, which is an example of a database (hereinafter referred to as "DB").

[0023] The question input to the information processing device 1 does not necessarily have to be in text form; it may be represented in a form other than text, such as audio data or image data. If audio data is input, the information processing device 1 can convert the audio data into a question in text form using a known speech recognition method, such as LSTM (Long Short Term Memory), which is an example of a recurrent neural network. If image data is input, the information processing device 1 can convert the image data into a question in text form using a known image recognition method, such as ResNet (Residual Neural Network), which is an example of a convolutional neural network.

[0024] Search unit 2 searches the answer database 8 for the input question. The answer database 8 is a database that stores answers to various questions in advance. In the answer database 8, answers to questions are managed in an FAQ format, which is a question-and-answer format. Hereafter, the question that is pre-associated with the answer in FAQ format, that is, the part corresponding to the "Q" in FAQ, will be referred to as the "question title".

[0025] There are no restrictions on the search method used in the search unit 2; for example, publicly known search methods such as keyword search or natural language search can be used.

[0026] For example, in keyword search, answers containing the same words as those in the question, or answers associated with question titles containing the same words as those in the question, are searched as answers to the question, so multiple answers may be found for a given question. Similarly, in natural language search, answers with a similar context to the question, or answers associated with question titles with a similar context to the question, are searched as answers to the question, so multiple answers may be found for a given question. Thus, the answers found by search unit 2 are not necessarily unique.

[0027] The classification unit 3 classifies each response into multiple groups based on the similarity between each response retrieved by the search unit 2.

[0028] Specifically, the classification unit 3 calculates the similarity between each answer found by the search unit 2 and classifies similar answers into the same group. Answers classified into the same group have a higher similarity than answers classified into other groups, so they form a set of answers to similar questions, such as those asking the same thing but with different wording. In other words, if a question contains multiple intentions, a group is formed for each intention contained in the question. To put it another way, the number of answer groups represents the number of intentions contained in the question. The similarity between each of these answers is an example of the first similarity. The specific classification method of the found answers will be explained later.

[0029] Meanwhile, the splitting unit 4 performs syntactic analysis on the input question sentence and divides the question sentence into phrases.

[0030] A phrase is the smallest unit obtained by dividing a question sentence into the shortest possible segments that do not sound unnatural in terms of meaning or pronunciation. In this embodiment, the question sentence is divided into phrases according to predetermined division rules regarding the parts of speech of the question sentence.

[0031] The similarity determination unit 5 calculates the similarity between each answer in the same group of answers, for each group of answers classified by the classification unit 3, and the segmented expression of the question obtained by combining the input question and the segmented phrases of the question divided by the segmentation unit 4.

[0032] A question sentence splitting expression refers to a combination of two or more phrases included in a question sentence, in which at least one phrase has been removed from the question sentence. Furthermore, since the question title of an answer managed in FAQ format is information paired with the answer, the answer in this embodiment also includes the question title.

[0033] The selection unit 6, for each group of answers classified by the classification unit 3, selects one answer from among multiple answers classified in the same group that best matches the intent associated with the group, by referring to the selection rules and the similarity determined by the similarity determination unit 5. The selection rules are rules that define how to select the desired answer based on the similarity between the question, the segmented expressions of the question, and the answers. The specific details will be explained later. Note that the similarity between the question, the segmented expressions of the question, and the answers is an example of the second similarity metric.

[0034] The output unit 7 presents the answer selected by the selection unit 6 to the user as the answer to the input question.

[0035] Presenting the answer to a question to the user means making the user able to recognize the answer to the question. Therefore, the output unit 7 may output the answer in any form as long as the user can recognize the answer to the question. For example, the output unit 7 may display the answer to the question on the display unit 19 (see Figure 2), which will be described later, or it may transmit the answer to the question to an external device (not shown) connected to a communication line via the communication unit 17 (see Figure 2), which will be described later. The output unit 7 may also print the answer to the question on paper using an image forming device (not shown), such as a printer, or it may output the answer to the question as audio. The following describes an example in which the output unit 7 displays the answer to a question on the display unit 19.

[0036] Note that the information processing device 1 does not necessarily need to have the answer DB8; for example, the answer DB8 may be provided in an external device. In this case, the information processing device 1 can connect to the external device via the communication unit 17 described later and search for the answer to the input question.

[0037] The information processing device 1 shown in Figure 1 can be configured using a computer 10. Figure 2 shows an example of the main electrical system configuration of the information processing device 1 configured using a computer 10.

[0038] Computer 10 includes a CPU (Central Processing Unit) 11, which is an example of a processor responsible for executing each functional unit of the information processing device 1 shown in Figure 1; a ROM (Read Only Memory) 12 that stores the startup program (Basic Input Output System: BIOS) that performs the startup process of computer 10; a RAM (Random Access Memory) 13 used as a temporary workspace for the CPU 11; a non-volatile memory 14; and an input / output interface (I / O) 15. The CPU 11, ROM 12, RAM 13, non-volatile memory 14, and I / O 15 are connected to each other via a bus 16.

[0039] Non-volatile memory 14 is an example of a storage device that retains stored information even when the power supplied to it is cut off. For example, semiconductor memory is used, but a hard disk may also be used. Therefore, non-volatile memory 14 stores, for example, an information processing program that makes computer 10 function as information processing device 1. As in answer DB8, information that needs to be retained even when the power to information processing device 1 is cut off is stored in non-volatile memory 14.

[0040] For example, a communication unit 17, an input unit 18, and a display unit 19 are connected to I / O 15.

[0041] The communication unit 17 is connected to a communication line and is equipped with a communication protocol for data communication with external devices via the communication line.

[0042] The input unit 18 is an example of a unit that receives user operations on the information processing device 1 and notifies the CPU 11 of them, and includes, for example, buttons, touch panels, keyboards, mice, and pointing devices.

[0043] The display unit 19 is an example of a unit that visually displays information processed by the CPU 11, and includes, for example, a liquid crystal display and an organic EL (Electro-Luminescence) display.

[0044] Furthermore, when remote control of the information processing device 1 is performed from an external device, the information processing device 1 receives the user's operations via the communication unit 17 and transmits the information processed by the information processing device 1 to the external device via the communication unit 17. Therefore, it is not always necessary to connect the input unit 18 and the display unit 19 to I / O 15.

[0045] Next, we will explain the answer search process performed by the information processing device 1.

[0046] Figure 3 is a flowchart showing an example of the flow of the answer retrieval process executed by the CPU 11 of the information processing device 1 when a question is received from a user.

[0047] The information processing program that defines the answer search process is pre-stored, for example, in the non-volatile memory 14 of the information processing device 1. The CPU 11 of the information processing device 1 reads the information processing program stored in the non-volatile memory 14 and executes the answer search process.

[0048] In step S10, the CPU 11 searches the answer DB 8 for the answer to the input question. The CPU 11 may also use, for example, the search function of a portal site to search for the answer to the question from various databases published on the internet.

[0049] In step S20, the CPU 11 obtains the number of intents contained in the question from the answers retrieved in step S10.

[0050] Specifically, CPU11 performs cluster analysis on each of the responses found in step S10 and classifies each response into a cluster.

[0051] Figure 4 shows an example of classification where answers to an input question are searched from the answer database 8 and the retrieved answers are classified into clusters.

[0052] The example in Figure 4 shows how the seven answers, represented by IDs 1 through 7, found in response to the question "Scan an image and send it by fax," are classified into two clusters. Frame 20 represents the first cluster, and frame 21 represents the second cluster. The numbers within each cluster represent the answer IDs. Therefore, the example in Figure 4 shows that the answers represented by IDs 1, 2, 5, and 7 are classified into the first cluster ("Group G1"), and the answers represented by IDs 3, 4, and 6 are classified into the second cluster ("Group G2"). Since each answer is classified into two clusters, CPU 11 obtains "2" as the number of intentions contained in the question.

[0053] As an example, in this embodiment, each answer is classified into groups based on the similarity of the question titles of the answers. However, each answer may also be classified into groups based on the similarity of the answer text, or the similarity of the answer text and the question title combined.

[0054] Furthermore, while cluster analysis was used to classify the responses in this embodiment, the method of classifying responses is not limited to cluster analysis, as long as each response can be classified into a group. For example, responses may be classified using known unsupervised learning methods such as the k-means method or Linear Discriminant Analysis (LDA). Alternatively, responses may be classified using an analysis method such as the tf-idf method, which measures the frequency of occurrence of each word included in the responses and classifies the responses by focusing on the similarity of responses obtained from the word frequency.

[0055] In step S30 of Figure 3, the CPU 11 divides the input question sentence into phrases according to predetermined division rules regarding the parts of speech of the question sentence. Specifically, the CPU 11 performs morphological analysis on the question sentence and identifies the part of speech for each morpheme. Then, the CPU 11 reads the question sentence from the beginning to the end and divides the question sentence into multiple phrases by treating the range up to just before the appearance of the next noun as one phrase. Note that nouns include compound nouns formed by combining two or more nouns, such as "economic recovery".

[0056] Figure 5 shows an example of how a question can be divided. The question "Scan an image and send it by fax" shown in [No. 1] of Figure 5 can be divided into four phrases: "Scan," "Scan an image," "By fax," and "Send." The question shown in [No. 1] is an example of a question with a parallel relationship, as the same sentence structure, "Phrase A1 + Phrase A2" and "Phrase B1 + Phrase B2," is repeated to express different questions.

[0057] The question "How do I replace and purchase toner?" shown in [No.2] of Figure 5 can be divided into three phrases: "how to replace and purchase toner." The question in [No.2] is a question with a coordinate relationship because phrases A1 and B1 modify phrases A2 and B2 respectively, expressing different questions.

[0058] Furthermore, some questions combine parallel and equal relationships, such as, "Please tell me how to replace and purchase toner cartridges, and the payment deadline and payment method for the purchased toner cartridges."

[0059] In this way, the CPU 11 divides the question sentence, which is composed of parallel relationships, equivalence relationships, or a combination of parallel and equivalence relationships, into phrases.

[0060] In step S40 of Figure 3, the CPU 11 generates segmented representations of the question. For example, if the question is "Scan an image and send it by fax" as shown in [No. 1] of Figure 5, the following 10 segmented representations can be obtained by combining two or more phrases contained in the question. Note that "(N): N is a natural number" associated with each segmented representation is the segmented representation ID representing that segmented representation.

[0061] (1) “Import the image and send it by fax.” (2) “Send a fax by scanning” (3) “Scan and send images.” (4) “Scan the image and send it by fax” (5) "Send a fax" (6) “Scan to fax” (7) “Send by scanning” (8) “Import the image and send it by fax” (9) “Import and send images” (10) “Scan the image and”

[0062] After generating segmented representations of the question, the CPU 11 calculates the similarity between each answer in the group classified in step S20 and the question and its segmented representations using a known similarity calculation method.

[0063] Figure 6 shows an example of how to calculate the similarity between each answer included in the group shown in Figure 4 and the question and its segmented form.

[0064] In this embodiment, the similarity score increases as each answer in the group becomes more similar to the question and its sub-expressions. Specifically, a similarity score of "0" indicates that the two are not similar at all, and a similarity score of "1" indicates that the two are identical. However, it goes without saying that a similarity score that decreases as each answer in the group becomes more similar to the question and its sub-expressions may also be used.

[0065] In the example shown in Figure 6, the answer "How to operate the scanner," which is ID 1 and classified as group G1 in Figure 4, is found as an answer to the question "Scan an image and send it by fax" shown in [No. 1] in Figure 5, and the similarity between the question and its segmented expressions is shown. According to the example shown in Figure 6, the similarity between the question and the answer is "0.90", the similarity between segmented expression ID (1) and the answer is "0.78", the similarity between segmented expression ID (2) and the answer is "0.56", and the similarity between segmented expression ID (5) and the answer is "0.30".

[0066] In this way, the CPU 11 calculates the similarity between the answers ID1, ID2, ID5, and ID7, which are classified as group G1 in Figure 4, and the question and its segmented representations, as well as the similarity between the answers ID3, ID4, and ID6, which are classified as group G2 in Figure 4, and the question and its segmented representations.

[0067] In step S50 of Figure 3, the CPU 11 selects one response from among multiple responses classified in the same group for each group of responses classified in step S20 that best matches the intention associated with the group.

[0068] Specifically, CPU 11, following the selection rules, selects one answer that best matches the intent associated with the group, based on the similarity between each answer included in the group calculated in step S40 and the question and its segmented form.

[0069] For example, suppose a selection rule A is set up to select the answer that best matches the group and the associated intent from among multiple answers classified into the same group, the answer whose segmented expression of the question has the largest decrease in similarity between the answer and the question (referred to as the "criteria similarity").

[0070] The reason for setting these selection rules is that the answer selected by Selection Rule A has the highest probability of containing the intent of the question expressed by the clause that is not included in the segmentation expression of the question, i.e., the deleted clause, among the multiple answers classified in the same group. The answer selected by Selection Rule A is an example of the first answer.

[0071] When selecting an answer according to selection rule A, CPU11 calculates the maximum difference, which is the largest decrease in similarity relative to the baseline, for each answer classified into the same group.

[0072] Figure 7 shows an example of selecting the most appropriate answer from each of the responses included in Group G1 shown in Figure 4, according to selection rule A, to match the intention associated with Group G1.

[0073] As shown in Figure 4, Group G1 includes responses ID1, ID2, ID5, and ID7. Therefore, CPU 11 calculates the maximum difference for each response in Group G1 by referring to the similarity between each response and the question and the segmented representation of the question, which was calculated in step S40. In the example shown in Figure 7, for each response represented by response IDs ID1, ID2, ID5, and ID7, the segmented representation represented by segmented representation ID (5) had the lowest similarity among all the segmented representations of the question.

[0074] In this case, the maximum difference for ID1's answer is "0.60", the maximum difference for ID2's answer is "0.43", the maximum difference for ID5's answer is "0.50", and the maximum difference for ID7's answer is "0.40". Therefore, CPU11 selects ID1's answer as the answer that best matches the intent associated with group G1.

[0075] Similarly, CPU 11 selects the response with the largest maximum difference from among the responses ID3, ID4, and ID6 included in group G2 shown in Figure 4 as the response that best matches the intent associated with group G2. In this way, from among the multiple responses retrieved for the question "Scan an image and send it by fax," the number of responses that best match the intent contained in the question (in this case, two) is selected.

[0076] Note that selection rule A is just one example of a selection rule, and CPU 11 may choose an answer according to a selection rule other than selection rule A.

[0077] For example, CPU11 may select an answer according to selection rule B, which selects the answer with the highest similarity to the segmented expression of the question from among multiple answers classified in the same group, as the answer that best matches the intent associated with the group.

[0078] The segmented form of the question is an expression from which redundant phrases (also called "noise") unrelated to the intent contained in the question have been removed. Therefore, the segmented form of the question with the highest similarity score can be considered to represent the intent contained in the question. Accordingly, the answer with the highest similarity score among the segmented forms of the question is the answer with the question title that is most similar to the intent associated with the group. The answer selected by selection rule B is an example of a second answer.

[0079] Figure 8 shows an example of selecting the most appropriate answer from each of the responses included in Group G1 shown in Figure 4, according to selection rule B, to match the intention associated with Group G1.

[0080] Suppose that the highest similarity score between each answer and the segmented expression of the question, as shown in Figure 8, is "0.92". In this case, CPU 11 selects the answer with the highest similarity score, i.e., answer ID 1, as the answer that best matches the intent associated with group G1. Note that if the similarity score between each answer and the question and the segmented expression of the question is highest, it means that the question originally contained only one intent.

[0081] Since question sentences may contain redundant expressions, CPU11 expresses the redundancy of the expression in the question sentence using a redundancy score, where a higher value indicates more redundant expressions. If the redundancy score of the question sentence is greater than or equal to a predetermined value, CPU11 may select an answer according to selection rule B. This is because selection rule B removes noise from the question sentence and selects the answer that best matches the intent associated with the group from among multiple answers classified in the same group, making it more likely to select an answer that aligns with the intent associated with the group than selecting an answer according to selection rule A.

[0082] Furthermore, if multiple selection rules are defined, such as selection rule A and selection rule B, the CPU 11 may allow the user to choose which selection rule to follow when selecting an answer.

[0083] In step S60 of Figure 3, the CPU 11 displays the answer selected in step S50, that is, the answer corresponding to each intention contained in the question, on the display unit 19.

[0084] Figure 9 shows an example of how answers are displayed when, for example, the information processing device 1 functions as a chatbot that answers questions received from users via chat.

[0085] In screen 22, the message associated with the robot icon 23, which represents the shape of a robot, is a message output by the CPU 11, and the message associated with the user icon 24, which represents the shape of a human, is a question entered by the user into the information processing device 1.

[0086] As shown in Figure 9, if the question entered by the user into the information processing device 1 is "Scan an image and send it by fax," then, as explained earlier, it is presumed that the question contains two intentions. Therefore, from among the multiple answers found, the question titles that best suit each intention, such as "How to operate the scanner" and "How to operate the fax machine," are displayed on screen 22. Displaying these answer question titles is one example of how answers are presented. Naturally, the CPU 11 may also display the answer itself on screen 22 along with the answer question title.

[0087] When a user selects a question title for an answer displayed on screen 22 using a mouse or other means, the CPU 11 displays the answer associated with the selected question title on screen 22.

[0088] Furthermore, if the CPU 11 receives a request via chat from a user instructing it to display a different answer from the one displayed on screen 22, it may display on screen 22 the next answer that is appropriate to the intent associated with the group, from among the answers classified into the group that includes the answer for which the user was instructed to change the display.

[0089] When CPU 11 displays an answer to a question on screen 22, it may not simply display the answer, but may also display the number of intentions contained in the question obtained by CPU 11 along with the answer to the question. Figure 10 shows an example of a display that shows the number of intentions contained in the question. By displaying "There are 2 intentions contained in the question," the number of intentions contained in the question is presented to the user.

[0090] Furthermore, the reliability of the answer presented on screen 22 is increased if the CPU 11 explains to the user the reasoning behind the determination of the number of intentions contained in the question, rather than simply stating the number of intentions contained in the question. Therefore, the CPU 11 may present to the user the classification results of the answers searched for in relation to the question as the basis for the number of intentions contained in the question.

[0091] In the screen 22 shown in Figure 10, if the user selects the message "There are two intentions included in the question. Click here for the reasoning behind the estimated number of intentions" using the mouse or other means, the CPU 11 displays a classification screen 25, which includes the classification results of the answer, on the display unit 19. Figure 11 shows an example of the classification screen 25.

[0092] The classification screen 25 shown in Figure 11 displays, for example, the results of the cluster analysis for each answer shown in Figure 4. The classification screen 25 also displays the question titles of all answers found in the search for the question text, along with the answer IDs, which were not displayed in screen 22 shown in Figure 10. If the user selects a question title of an answer displayed on the classification screen 25 with a mouse or the like, the CPU 11 may display the content of the answer associated with the selected question title.

[0093] Although one embodiment of the information processing device 1 has been described above using embodiments, the disclosed form of the information processing device 1 is merely an example, and the form of the information processing device 1 is not limited to the scope described in the embodiments. Various changes or improvements can be made to the embodiments without departing from the gist of this disclosure, and such changed or improved forms are also included in the technical scope of the disclosure. For example, the order of processing in the answer search process shown in Figure 3 may be changed without departing from the gist of this disclosure. Specifically, the question sentence division shown in step S30 of Figure 3 may be performed before the search for answers to the question sentence shown in step S10.

[0094] Furthermore, in the above embodiment, an example of a system in which the answer search process is implemented in software was described. However, the same process as the answer search process flowchart shown in Figure 3 may also be performed in hardware. In this case, the processing speed can be increased compared to the case in which the answer search process is implemented in software.

[0095] In the above embodiment, the term "processor" refers to a broad type of processor, including general-purpose processors (e.g., CPU11) and dedicated processors (e.g., GPU: Graphics Processing Unit, ASIC: Application Specific Integrated Circuit, FPGA: Field Programmable Gate Array, programmable logic device, etc.).

[0096] Furthermore, the operation of the processor in the above embodiment may not be performed by a single processor, but may be performed by multiple processors located in physically separate locations working together. Also, the order of the processor operations is not limited to the order described in the above embodiment, but may be changed as appropriate.

[0097] In the above embodiment, an example was described in which the information processing program is stored in the non-volatile memory 14, but the storage location of the information processing program is not limited to the non-volatile memory 14. The information processing program of this disclosure can also be provided in a form recorded on a storage medium readable by the computer 10. For example, the information processing program may be provided in a form recorded on an optical disc such as a CD-ROM (Compact Disk Read Only Memory) or DVD-ROM (Digital Versatile Disk Read Only Memory). Alternatively, the information processing program may be provided in a form recorded on a portable semiconductor memory such as a USB (Universal Serial Bus) memory or a memory card. ROM 12, non-volatile memory 14, CD-ROM, DVD-ROM, USB, and memory card are examples of non-transitory storage media.

[0098] Furthermore, the information processing device 1 may download an information processing program from an external device via a communication line and store the downloaded information processing program in a memory device. In this case, the CPU 11 of the information processing device 1 reads the information processing program downloaded from the external device from the memory device and executes each of its processes.

[0099] The following are additional details regarding this embodiment.

[0100] (((1))) Equipped with a processor, The aforementioned processor, Multiple responses to a question containing multiple intentions are classified into multiple groups based on a first similarity score indicating the degree of similarity between each response. For each group, a second similarity metric is used to indicate the similarity between the question and each answer included in the group, using the question and a segmented expression of the question obtained by combining two or more phrases of the question. Based on this, an answer corresponding to each intention contained in the question is presented from among the multiple answers. Information processing device.

[0101] (((2))) The processor presents the classification results of the multiple answers, along with the answers corresponding to each intent contained in the question. The information processing device described in (((1))).

[0102] (((3))) The processor further presents the number of intentions contained in the question statement. The information processing device described in (((2))).

[0103] (((4))) The processor identifies, for each group, a first answer which is the answer associated with the second similarity of the segmented expression of the question that has the largest decrease difference in the second similarity of the question, or a second answer which is the answer associated with the highest second similarity among the second similarities of each segmented expression of the question, and presents them as answers corresponding to each intent contained in the question. An information processing device as described in any of (((1))) to (((3))).

[0104] (((5))) The processor, if the redundancy of the question statement (which indicates that the more redundant the expression, the larger the value), is greater than or equal to a predetermined value, presents the second answer as an answer corresponding to each of the intentions contained in the question statement. The information processing device described in (((4))).

[0105] (((6))) On the computer, Multiple responses to a question containing multiple intentions are classified into multiple groups based on a first similarity score indicating the degree of similarity between each response. For each of the aforementioned groups, a process is performed to present an answer from among the multiple answers that corresponds to each of the intentions contained in the question, using a second similarity score that indicates the similarity between the question and each answer included in the group, obtained by combining the question and two or more phrases of the question to form a segmented expression of the question. Information processing program.

[0106] According to (((1))) and (((6))), the effect is that for a question that contains multiple intentions, it is possible to provide as many answers as there are intentions contained in the question.

[0107] According to (((2))), this has the effect of informing the user of the reason for the estimation of the number of intentions contained in the question.

[0108] According to (((3))), this has the effect of notifying the user of the number of intentions contained in the question sentence recognized by the information processing device.

[0109] According to (((4))), this has the effect of allowing the user to choose how to identify the answer to be presented, depending on the characteristics of the question.

[0110] According to (((5))), compared to presenting the first answer as the answer to the question, this method has the effect of being able to present an answer that is closer to the intent of the question. [Explanation of Symbols]

[0111] 1. Information Processing Device 2. Search section 3 Classification section 4 Division 5 Similarity determination section 6. Selection Section 7 Output section 8 Answer DB 10 Computers 11 CPU 12 ROM 13 RAM 14 Non-volatile memory 15 I / O 16 bus 17 Communication Unit 18 Input Units 19 Display Unit 20(21) slot 22 screens 23 Robot Icons 24 User Icons 25 Classification screen

Claims

1. Equipped with a processor, The aforementioned processor, Multiple responses to a question containing multiple intentions are classified into multiple groups based on a first similarity score indicating the degree of similarity between each response. For each group, the system uses the question statement, a segmented representation of the question statement obtained by combining two or more phrases of the question statement, and a second similarity score indicating the similarity between each answer included in the group, to present answers from among the multiple answers that correspond to each intention contained in the question statement. Information processing device.

2. The processor presents the classification results of the multiple answers, along with the answers corresponding to each intent contained in the question. The information processing apparatus according to claim 1.

3. The processor further presents the number of intentions contained in the question statement. The information processing apparatus according to claim 2.

4. The processor identifies, for each group, a first answer which is the answer associated with the second similarity of the segmented expression of the question that has the largest decrease difference in the second similarity of the question, or a second answer which is the answer associated with the highest second similarity among the second similarities of each segmented expression of the question, and presents them as answers corresponding to each intention contained in the question. The information processing apparatus according to claim 1.

5. The processor, if the redundancy of the question statement, which indicates that the more redundant the expression, is greater than or equal to a predetermined value, presents the second answer as an answer corresponding to each of the intentions contained in the question statement. The information processing apparatus according to claim 4.

6. On the computer, Multiple responses to a question containing multiple intentions are classified into multiple groups based on a first similarity score indicating the degree of similarity between each response. For each of the aforementioned groups, a process is performed to present an answer from among the multiple answers that corresponds to each of the intentions contained in the question, using a second similarity score that indicates the similarity between the question and each answer included in the group, obtained by combining the question and two or more phrases of the question. Information processing program.

Images