Information processing device, information processing method, and program
The information processing device addresses the inefficiency of pre-created question-and-answer collections by directly acquiring and vectorizing reference information to generate accurate and fluent answers, reducing time and cost.
Patent Information
- Application Number
- JP2023097733
- Authority / Receiving Office
- JP · JP
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2023-06-14
- Publication Date
- 2024-11-28
- Estimated Expiration
- 2043-06-14
AI Technical Summary
Existing methods for automatically obtaining answers to questions require the creation of a question-and-answer collection in advance, which is time-consuming and costly.
An information processing device that acquires reference information related to a question and generates answers without pre-creating a question-and-answer collection, using a first acquisition unit to obtain reference information, a second acquisition unit to input it into an answering unit, and a generation unit to generate vectors for comparison and selection of similar information.
Enables the generation of accurate and fluent answers to questions without the need for a pre-existing collection, improving answer quality and reducing time and cost.
Smart Images

Figure 0007591619000001_ABST
Abstract
Description
[Technical field]
[0001] The present invention relates to an information processing device, an information processing method, and a program for obtaining an answer to a received question. [Background technology]
[0002] Conventionally, attempts have been made to automatically obtain answers to accepted questions. For example, Patent Document 1 proposes a method for automatically creating a collection of questions and answers using a computer. Patent Document 1 proposes automatically creating a collection of questions and answers using a method including a voice analysis step in which a computer analyzes voice related to a conversation to obtain spoken words that are words included in the conversation, a word group analysis step in which the spoken words are analyzed to determine which of a plurality of keywords the spoken words correspond to, a question and answer creation step in which a question and an answer to the question are created using the spoken words, and a question and answer acquisition step in which a question and an answer related to the keyword that is determined to correspond to the classified question and the answer to the question are obtained to be recorded in the collection of questions and answers. By creating a collection of questions and answers in this way, it becomes possible to automatically create an answer to a certain question. [Prior art documents] [Patent documents]
[0003] [Patent Document 1] Patent No. 7271019 Summary of the Invention [Problem to be solved by the invention]
[0004] Although the invention disclosed in Patent Document 1 can automatically obtain answers, it is necessary to prepare a collection of questions and answers in advance. Creating such a collection of questions and answers requires time and money.
[0005] The present invention provides an information processing device and the like that can obtain answers to accepted questions without necessarily creating a collection of questions and answers as in Patent Document 1. [Means for solving the problem]
[0006] [Concept 1] The information processing device according to the present invention comprises: a reception section for receiving questions; a first acquisition unit that acquires reference information related to the question from a search unit or generates reference information based on information acquired from the search unit; a second acquisition unit that performs input to an answer unit using the reference information acquired by the first acquisition unit and the question, and acquires an answer to the input; The present invention may also include:
[0007] [Concept 2] In the information processing device according to concept 1, The first acquisition unit acquires or generates a plurality of pieces of reference information related to the question, The second acquisition unit may acquire an answer corresponding to each piece of reference information from the answer unit, using each piece of reference information and the question.
[0008] [Concept 3] In the information processing device according to concept 2, The plurality of pieces of reference information may be obtained by decomposing the original information obtained by the search performed by the search unit.
[0009] [Concept 4] The information processing device according to concept 2 or 3 is a generation unit that generates a question vector for the question and generates a reference information vector for each of a plurality of pieces of reference information; The reference information vector and the question vector may be compared to select a predetermined number of reference information vectors similar to the question vector, and the second acquisition unit may use the reference information corresponding to the reference information vectors to acquire an answer from the answer unit.
[0010] [Concept 5] An information processing device according to any one of concepts 1 to 4, The apparatus may further include a first evaluation unit that evaluates accuracy of the answer acquired by the second acquisition unit.
[0011] [Concept 6] In the information processing device according to concept 5, A generation unit that generates an answer vector for the answer and an evaluation vector for information to be evaluated, The first evaluation unit may evaluate accuracy of the answer by using a similarity between the answer vector and the evaluation vector.
[0012] [Concept 7] In the information processing device according to any one of concepts 1 to 6, The answer may be input from the answer section to a second evaluation section, whereby an evaluation result regarding the fluency of the answer may be received.
[0013] [Concept 8] An information processing device according to any one of concepts 1 to 7, an evaluation unit that evaluates the accuracy or fluency of the answer acquired by the second acquisition unit; When the evaluation result by the evaluation unit is not equal to or greater than a threshold, the second acquisition unit inputs reference information related to the question, which is different from the reference information already used, and the question to the answer unit to acquire another answer to the input; The evaluation unit may evaluate the another answer for accuracy or fluency.
[0014] [Concept 9] An information processing device according to any one of concepts 1 to 7, an evaluation unit that evaluates the accuracy or fluency of the answer acquired by the second acquisition unit; When the evaluation result by the evaluation unit is not equal to or greater than a threshold, a plurality of pieces of reference information are generated by decomposing information that is related to the question, is obtained from the search unit, and is different from information that has already been used, and the second acquisition unit inputs the plurality of pieces of reference information and the question to the answer unit, and acquires a plurality of different answers to the input. The assessor may assess the plurality of alternative answers for accuracy or fluency.
[0015] [Concept 10] The information processing method according to the present invention comprises: accepting a question by a reception unit; acquiring, by a first acquisition unit, reference information related to the question from a search unit; a step of, by a second acquisition unit, inputting into an answer unit using the reference information acquired by the first acquisition unit and the question, and acquiring an answer to the input; The present invention may also include:
[0016] [Concept 11] The program according to the present invention comprises: A program to be installed on an information processing device, An information processing device in which the program is installed A reception function to accept questions, A first acquisition function for acquiring reference information related to the question from a search unit; a second acquisition function that performs input to an answer unit using the reference information acquired by the first acquisition unit and the question, and acquires an answer to the input; may be executed. Effect of the Invention
[0017] According to the present invention, it is possible to provide an information processing device and the like that can obtain answers to accepted questions without the need to separately create a collection of questions and answers as in Patent Document 1. [Brief description of the drawings]
[0018] [Figure 1] 1 is a block diagram of an information processing apparatus according to an embodiment of the present invention; [Diagram 2] FIG. 11 is a block diagram of an information processing apparatus according to another example of an embodiment of the present invention. [Diagram 3] FIG. 11 is a block diagram of an information processing apparatus according to still another example of an embodiment of the present invention. [Figure 4] FIG. 2 is a diagram showing an example of a processing flow by the information processing device according to the embodiment of the present invention. [Diagram 5] 11A and 11B are diagrams showing examples of answers and their evaluations outputted from the information processing device according to the embodiment of the present invention. [Figure 6] FIG. 11 is a diagram showing another example of the flow of processing by the information processing device according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0019] Embodiment "composition" The information processing device of this embodiment may be composed of one device or multiple devices. In this embodiment, an information processing method using the information processing device of this embodiment, a program (server program) installed to generate the information processing device, and a storage medium such as a USB or DVD storing the program are also provided. The information processing device of this embodiment may be a server, and a mode using a cloud environment can also be adopted. In addition, in this embodiment, a program (user program) installed in the user terminal 200 and a storage medium such as a USB or DVD storing the program are also provided. In addition, the information processing device of this embodiment has a processor, and the information processing method may be realized by processing by the processor.
[0020] As shown in FIG. 1, the information processing device 100 may include a reception unit 10 that receives a question, a first acquisition unit 20 that acquires reference information such as cases related to the question from a search unit 110 or generates the reference information based on information acquired from the search unit 110, and a second acquisition unit 30 that uses the reference information acquired by the first acquisition unit 20 and the question accepted by the acceptance unit 10 to input to an answering unit 120 and acquire an answer to the input. The search unit 110 may be provided in a device different from the information processing device 100, and may use an external search engine such as Bing. The answering unit 120 may also be provided in a device different from the information processing device 100, and may use an external chatbot having an artificial intelligence function such as ChatGPT. However, the present invention is not limited to such an embodiment, and the information processing device 100 may include the search unit 110, and the search unit 110 may search for information stored in a storage unit 60 that constitutes a part of the information processing device 100 (see FIG. 2). Also, the information processing device 100 may have an answering unit 120, and may acquire an answer by inputting a question and reference information to the answering unit 120 (see FIG. 3). The answering unit 120 may have a machine learning function, and may acquire an answer by inputting the input question and reference information to an answer model. The answer model may be generated in advance using a question for learning, reference information, and an answer for learning. The answer model may be stored in the storage unit 60, and may be updated periodically. When a question and reference information are actually input, and the answer obtained is incorrect, the answering unit 120 may update the answer model by inputting an answer that is thought to be correct by an administrator or a user.
[0021] In this embodiment, when a user inputs a question from the operation unit 210 (see step Sa in FIG. 4), the first acquisition unit 20 automatically performs a search by the search unit 110 using the question, and acquires a search result (see step Sb in FIG. 4). Then, the second acquisition unit 30 automatically inputs the acquired search result (reference information) or reference information generated from the acquired search result and the question into the answering unit 120, and automatically acquires an answer to the input result (see step Sc in FIG. 4). As in this embodiment, by inputting reference information to the answering unit 120 in addition to the question, the accuracy of the answer by the answering unit 120 can be improved. The reference information is, for example, information related to the question, text information containing correct content, and may be information stored in advance in the storage unit 60 or information acquired by searching the WEB.
[0022] The information processing device 100 may be capable of communicating with a plurality of user terminals 200. The user terminal 200 may be capable of communicating with the information processing device 100 by inputting user identification information such as a login ID and a password from the user terminal 200. Information input from the user terminal 200 (e.g., a question) and information obtained by the information input from the user terminal 200 (e.g., an answer) may be stored in the storage unit 60 in association with the user identification information. Such information may be readable by the user later. The user terminal 200 may have an operation unit 210 for a user to operate, and a display unit 220 consisting of a display screen or the like for displaying various information. In the case of a smartphone or a tablet terminal, the touch panel screen serves as both the operation unit 210 and the display unit 220.
[0023] The first acquisition unit 20 may acquire or generate multiple pieces of reference information related to the question. As an example, a predetermined number (one or more) of pieces of information on a website obtained by inputting a question into the search unit 110 or pieces of information stored in the storage unit 60 may be acquired in descending order of relevance. When searching with the search unit 110, pieces of information with high relevance are often displayed in descending order from the top, so by selecting pieces of information in ascending order from the top, reference information with high relevance to the question can be automatically acquired.
[0024] In another embodiment, the multiple pieces of reference information acquired by the first acquisition unit 20 may be obtained by decomposing the original information stored in the storage unit 60 or the result of a search on the WEB. As an example, the case information or WEB information stored in the storage unit 60 acquired by the first acquisition unit 20 may be divided into chunks of a certain size (chunks). When dividing into chunks of a certain size, the structure of HTML, a paper, or the like may be used to adjust the chunks to be in paragraph units or sentence units. As an example, when a paper published on the WEB is used as reference information, the reference information may be divided into paragraphs such as introductions and explanations, and the questions and the reference information may be input in sequence to obtain multiple answers. By adopting such an embodiment, the reference information considered to be related to the question can be made to an appropriate length before the answer is obtained. This embodiment is particularly useful when the number of characters that can be input to the answer unit 120 is limited. The original information may be information that is listed at the top as being highly related in the result of searching the storage unit 60 or the result of a search on the WEB, or a predetermined number of pieces of information may be used in order from the top as the original information.
[0025] Whether to use the information stored in the storage unit 60 or the search results on the Web as the search target may be input from the operation unit 210. In this case, it is also possible to use both the information in the storage unit 60 and the search results on the Web. When using information from the storage unit 60, highly accurate reference information such as case studies can be used. When using search results on the Web, it is advantageous in that a large amount of information can be used as candidates. The information in the storage unit 60 may be periodically updated by an administrator to increase the amount and improve the accuracy.
[0026] A generating unit 50 for generating a vector may be provided. When generating a vector, the sentence may be vectorized using the number of occurrences of words that appear in a given sentence. Also, the sentence may be vectorized by focusing on the meaning of each word used in the sentence. The generating unit 50 may generate a question vector for a question, and generate a reference information vector for each of a plurality of pieces of reference information. Then, the second acquiring unit 30 may compare the reference information vector based on the reference information with the question vector based on the question, select a reference information vector similar to the question vector, and acquire an answer from the answering unit 120 using the reference information corresponding to the reference information vector. By adopting such an embodiment, it is possible to select and use reference information that matches the question. At this time, only reference information that matches a predetermined first threshold or more may be used as information to be input together with the question. By adopting such an embodiment, it is beneficial because an answer that matches the question can be obtained with high accuracy.
[0027] The second acquisition unit 30 may acquire an answer corresponding to each piece of reference information from the answering unit 120 using each piece of reference information acquired by the first acquisition unit 20 and a question. In this case, the second acquisition unit 30 may input a certain question and one piece of reference information related to the certain question to the answering unit 120 and acquire the input result as an answer to the certain question. In this case, the second acquisition unit 30 may acquire an answer by sequentially changing the combination of a certain question and reference information. When such an embodiment is adopted, a plurality of answers corresponding to each piece of reference information used will be acquired. The answer acquired in this manner may be output by the output unit 80 or may be stored in the storage unit 60 as an answer to the question.
[0028] If there is no appropriate reference information, a method may be used in which only the question is first input to the answering unit 120 to make the answering unit 120 create an expected answer, and the vector of the expected answer generated by the generating unit 50 is used to search for reference information (reference information that matches at least a predetermined first threshold). The vector of the reference information is also generated by the generating unit 50. At this time, it is also useful to remove noise that has a strong effect on the vector from the expected answer and the reference information, such as by deleting proper nouns. Then, the obtained reference information and the question may be input to the answering unit 120 to obtain an answer. As another aspect, a question and one or more pieces of reference information may be input in advance to the answering unit 120, and the obtained answer may be used as the expected answer. Then, another reference information (reference information that matches at least a predetermined first threshold) similar to the vector of the expected answer generated by the generating unit 50 may be searched for. In this case, the answer is obtained from the answering unit 120 using the obtained other reference information and the question.
[0029] The expected responses described above may also be evaluated by the evaluation unit 70, and only expected responses that are equal to or greater than a second threshold value, which will be described later, may be used.
[0030] An evaluation unit 70 that judges the accuracy (right or wrong) of the answer acquired by the second acquisition unit 30 may be provided. The evaluation unit 70 may be provided in the information processing device 100 (see Figs. 2 and 3), or may be provided in an external device different from the information processing device 100, or a part of the evaluation unit 70 (first evaluation unit 71 in Fig. 1) may be provided in the information processing device 100, and another evaluation unit 70 (second evaluation unit 72 in Fig. 1) may be provided in an external device different from the information processing device 100 (see Fig. 1). As an example, the first evaluation unit 71 may evaluate the accuracy of the answer, and the second evaluation unit 72 may evaluate the fluency.
[0031] The evaluation unit 70 may determine the accuracy (correctness) by evaluating whether the answer obtained from the answering unit 120 is consistent with the information stored in the storage unit 60 or the information searched by the search unit 110 (see step Sd in FIG. 4). At this time, if even one example of information consistent with the answer is found, the evaluation unit 70 may give a high evaluation. For example, if even one piece of information matching the answer is found in the information stored in the storage unit 60, the information searched by the search unit 110, the reference information input together with the question, etc., the evaluation unit 70 may give a high evaluation, and if no information matching the answer is found, the evaluation unit 70 may give a low evaluation. Also, if information inconsistent with the answer is found in a plurality of pieces of information, the evaluation unit 70 may give a low evaluation. Also, the evaluation unit 70 may use the similarity between the vector of the information stored in the storage unit 60, the information searched by the search unit 110, the reference information input together with the question, etc., and the vector of the obtained answer to perform the evaluation. At this time, the highest similarity value may be the evaluation value.
[0032] The vector (evaluation vector) of the information to be evaluated (information stored in the storage unit 60, information searched by the search unit 110, reference information input together with the question, etc.) and the vector (answer vector) of the answer are generated by the generation unit 50. The information stored in the storage unit 60 and the information searched by the search unit 110 used when the evaluation unit 70 performs the evaluation may be delimited into vectors of a certain size (chunk). In this case, as described above, the structure of HTML, a paper, etc. may be used to adjust the units to paragraphs or sentences. Note that the reference information may also be delimited into vectors of a certain size (chunk) if it is long. This aspect is particularly useful when the number of characters that can be input to the evaluation unit 70 is limited. When dividing the information to be evaluated (information stored in the storage unit 60, information searched by the search unit 110, reference information input together with the question, etc.), a certain range of overlap may be provided to avoid destroying semantic consistency, and the information may be divided into units that form sentences.
[0033] Furthermore, the information to be evaluated (information stored in the storage unit 60, information searched by the search unit 110, reference information entered together with the question, etc.) may be summarized, and the evaluation unit 70 may use the summarized information and the answer to evaluate the accuracy (correctness). Since proper nouns and numbers may be important points in the evaluation when summarizing, proper nouns and numbers contained in the answer may be left as they are. In this case, the generation unit 50 may also generate a vector for the summarized content, and the evaluation unit 70 may evaluate whether the evaluation using the similarity between the vector and the vector from the answer is equal to or greater than the second threshold value.
[0034] After the evaluation by the evaluation unit 70 is completed, the answer may be output by the output unit 80 and displayed on the user terminal 200 (see step Se in FIG. 4).
[0035] An example of the evaluation result by the evaluation unit 70 is shown in Fig. 5. The "prompt" in Fig. 5 is the question, and the "obtained response" is the answer. The comments "It is not Mr. Mori Mamoru" and "The API is public, so it is inaccurate" in Fig. 5 explain the reason why the accuracy of each was evaluated low.
[0036] When only the question entered by the user is input to answer section 120, information with low accuracy is often obtained as the answer. However, as in this embodiment, by inputting reference information together with the question into answer section 120, highly accurate information can be obtained.
[0037] 5, the evaluation unit 70 may indicate the evaluation result as a score, and the score may be output by the output unit 80 and displayed on the user terminal 200. When such an embodiment is adopted, the user can understand how reliable the answer content is. As an example of a method of calculating the score, as described above, the similarity between the vector of the information to be evaluated (information stored in the storage unit 60, information searched by the search unit 110, reference information entered together with the question, etc.) and the vector of the obtained answer may be used.
[0038] A plurality of answers output by the answering unit 120 may be displayed on the user terminal 200. In this case, as described above, when a score is displayed on the user terminal 200, the user can grasp each answer after grasping the reliability of each answer. When a plurality of answers are displayed on the user terminal 200, the answers may be displayed in a list, or the answers may be displayed sequentially by swiping or flicking the display screen.
[0039] Furthermore, if the evaluation of one or more answers obtained by the evaluation unit 70 does not reach the second threshold or more, another piece of reference information may be obtained, or one or more pieces of reference information may be newly created from another piece of original information, and one or more answers may be obtained again. Then, such an operation may be automatically repeated until the evaluation of the answer by the evaluation unit 70 reaches the second threshold or more (see step S30 in FIG. 6). By adopting such an embodiment, it is advantageous in that an answer with high accuracy can be automatically obtained.
[0040] The evaluation unit 70 may also evaluate the fluency of the answer acquired by the second acquisition unit 30. In this case, in addition to the accuracy (reliability) of the answer, the fluency of the answer is also evaluated. In this case, as described above, another piece of reference information may be acquired, or one or more pieces of reference information may be newly created from another piece of original information, and then one or more answers may be acquired again. Then, such an operation may be automatically repeated until the evaluation of the answer by the evaluation unit 70 becomes equal to or greater than the second threshold (see step S30 in FIG. 6). By adopting such an embodiment, a highly accurate and fluent answer can be automatically acquired.
[0041] The evaluation unit 70 may have a machine learning function, and may generate an evaluation model using the answer for learning and the evaluation of the answer. Regarding the accuracy of the answer, an accuracy model may be generated using the input answer for learning and the evaluation of the accuracy of the answer. Regarding the fluency of the answer, a fluency model may be generated using the input answer for learning and the evaluation of the fluency of the answer. The accuracy model and the fluency model may be stored in the storage unit 60 and may be updated periodically. The accuracy model and the fluency model may be actually used, and the output evaluation value may be rewritten by the administrator, so that the evaluation unit 70 updates the accuracy model and the fluency model. In some cases, the accuracy model and the fluency model may be actually used by a user, and the output evaluation value may be rewritten by the user, so that the evaluation unit 70 updates the accuracy model and the fluency model.
[0042] When the accuracy and fluency of the answer are to be evaluated, the accuracy and fluency of the answer may be weighted and evaluated by the evaluation unit 70. The weights may be determined by an administrator or may be freely set by a user from the user terminal 200.
[0043] For example, UniEval may be used as one of the externally provided evaluation units 70 (second evaluation unit 72). By using UniEval, it is possible to evaluate whether the sentence is fluent, whether the sentence is consistent, whether the sentence is consistent with the input question or reference information, whether the sentence is easy to understand, etc. When using an externally provided evaluation unit 70 in this way, the output unit 80 may input the answer from the answering unit 120 to the second evaluation unit 72, and the receiving unit 10 may receive the evaluation result by the second evaluation unit 72 regarding the fluency of the answer.
[0044] An example of the information processing method will be described with reference to FIG.
[0045] When a question is inputted by text or voice from the operation unit 210 of the user terminal 200, the question is received by the reception unit 10 (step S1).
[0046] When the reception unit 10 receives a question, the first acquisition unit 20 inputs the question to the search unit 110 (step S2) and obtains search results (step S3). At this time, the search unit 110 may use a search engine different from that of the information processing device 100, or may search for information stored in the storage unit 60 (see FIG. 1). Alternatively, the search unit 110 may be provided in the information processing device 100, and the search unit 110 may search for information stored in the storage unit 60 that constitutes a part of the information processing device 100 (see FIGS. 2 and 3).
[0047] Next, the first acquisition unit 20 acquires a plurality of pieces of reference information related to the question, and generates a plurality of pieces of reference information (case information) by decomposing the original information that is the search result (step S4).
[0048] Next, the generating unit 50 generates a question vector and a reference information vector generated for each of the plurality of reference information. Then, the second acquiring unit 30 compares the question vector with the reference information vector generated for each of the plurality of reference information, and selects one or more pieces of reference information that match at a level equal to or greater than a predetermined first threshold (step S10).
[0049] Next, the second acquisition unit 30 inputs the selected one or more pieces of reference information and questions to the answering unit 120 (step S20), and acquires one or more answers from the answering unit 120 (step S21). At this time, the second acquisition unit 30 may input the reference information and the question to the answering unit 120 at the same time, or, if the answering unit 120 is a chatbot, the second acquisition unit 30 may first input the question to the answering unit 120, and then input the reference information to the answering unit 120 at the timing of the next input.
[0050] Next, the evaluation unit 70 evaluates the accuracy and fluency of the answer acquired by the second acquisition unit 30 (step S25). Here, if the evaluation of the answer is equal to or higher than the second threshold, the output unit 80 outputs the answer, and the display unit 220 displays the answer. At this time, if the evaluation of multiple answers is equal to or higher than the second threshold, the output unit 80 may output the answer with the highest evaluation value, and the display unit 220 may display the answer. Alternatively, instead of this mode, multiple answers may be displayed on the display unit 220 together with their scores. On the other hand, if there is no answer with an evaluation of equal to or higher than the second threshold, steps S2 to S30 may be performed again, and the above steps may be repeated until an answer with an evaluation of equal to or higher than the second threshold is obtained. At this time, the first acquisition unit 20 may input the question to the search unit 110, and automatically select reference information that has not yet been used. Also, the storage unit 60 may store the search results when the question is input, and the first acquisition unit 20 may automatically access the search results (reference information) that have not yet been used. In this case, steps S3 to S30 are carried out again, rather than steps S2 to S30.
[0051] Each of the components of the reception unit 10, the output unit 80, the first acquisition unit 20, the second acquisition unit 30, the generation unit 50, etc. of this embodiment may be realized by one or more IC chips or electronic modules, etc., or may be realized by a circuit configuration. In addition, two or more "parts" such as the reception unit 10, the output unit 80, the first acquisition unit 20, the second acquisition unit 30, and the generation unit 50 may be realized by one control unit. In addition, when the search unit 110 and the answer unit 120 are provided in the information processing device 100, these search unit 110 and the answer unit 120 may be realized by one or more IC chips or electronic modules, etc., or may be realized by a circuit configuration, or two or more "parts" may be realized by one control unit.
[0052] The above description of the embodiment and the disclosure of the drawings are merely examples for explaining the invention described in the claims, and the invention described in the claims is not limited by the above description of the embodiment or the disclosure of the drawings. In addition, the description of the claims at the time of filing is merely an example, and the description of the claims may be appropriately changed based on the description of the specification, drawings, etc. [Explanation of symbols]
[0053] 10 Reception 20 First Acquisition Department 30 Second Acquisition Department 50 Generation part 60 Storage section 70 Evaluation Section 80 Output section 100 Information processing device 110 Search Department 120 Answer section
Claims
1. a reception section for receiving questions; a first acquisition unit that acquires cases related to the question from a search unit or generates cases based on information acquired from the search unit; a second acquisition unit that performs input to an answer unit using the case and the question acquired or generated by the first acquisition unit and acquires an answer to the input; An information processing device comprising:
2. A reception unit for receiving questions; a first acquisition unit that acquires reference information related to the question from a search unit or generates reference information based on information acquired from the search unit; a second acquisition unit that performs input to an answer unit using the reference information acquired or generated by the first acquisition unit and the question, and acquires an answer to the input; Equipped with The first acquisition unit acquires or generates a plurality of pieces of reference information related to the question, The second acquisition unit acquires an answer corresponding to each piece of reference information from the answer unit using each piece of reference information and the question.
3. The information processing apparatus according to claim 2 , wherein the plurality of pieces of reference information are obtained by decomposing the original information obtained by the search performed by the search unit.
4. a generation unit that generates a question vector for the question and generates a reference information vector for each of a plurality of pieces of reference information; The information processing device according to claim 2 or 3, further comprising: a second acquisition unit that compares the reference information vector with the question vector to select a predetermined number of reference information vectors similar to the question vector, and uses the reference information corresponding to the reference information vectors to acquire an answer from the answer unit.
5. A reception unit for receiving questions; a first acquisition unit that acquires reference information related to the question from a search unit or generates reference information based on information acquired from the search unit; a second acquisition unit that performs input to an answer unit using the reference information acquired or generated by the first acquisition unit and the question, and acquires an answer to the input; a first evaluation unit that evaluates the accuracy of the answer acquired by the second acquisition unit; An information processing device comprising:
6. A generation unit that generates an answer vector for the answer and an evaluation vector for information to be evaluated, The information processing device according to claim 5 , wherein the first evaluation unit evaluates accuracy of the answer by using a similarity between the answer vector and the evaluation vector.
7. A reception unit for receiving questions; a first acquisition unit that acquires reference information related to the question from a search unit or generates reference information based on information acquired from the search unit; a second acquisition unit that performs input to an answer unit using the reference information acquired or generated by the first acquisition unit and the question, and acquires an answer to the input; Equipped with An information processing device that receives an evaluation result regarding the fluency of the answer by inputting the answer from the answer section to a second evaluation section.
8. A reception unit for receiving questions; a first acquisition unit that acquires reference information related to the question from a search unit or generates reference information based on information acquired from the search unit; a second acquisition unit that performs input to an answer unit using the reference information acquired or generated by the first acquisition unit and the question, and acquires an answer to the input; an evaluation unit that evaluates the accuracy or fluency of the answer acquired by the second acquisition unit; Equipped with When the evaluation result by the evaluation unit is not equal to or greater than a threshold, the second acquisition unit inputs reference information related to the question, which is different from the reference information already used, and the question to the answer unit, and acquires another answer to the input; The information processing device, wherein the evaluation unit evaluates the accuracy or fluency of the other answer.
9. A reception unit for receiving questions; a first acquisition unit that acquires reference information related to the question from a search unit or generates reference information based on information acquired from the search unit; a second acquisition unit that performs input to an answer unit using the reference information acquired or generated by the first acquisition unit and the question, and acquires an answer to the input; an evaluation unit that evaluates the accuracy or fluency of the answer acquired by the second acquisition unit; Equipped with When the evaluation result by the evaluation unit is not equal to or greater than a threshold, a plurality of pieces of reference information are generated by decomposing information that is related to the question, is obtained from the search unit, and is different from information that has already been used, and the second acquisition unit inputs the plurality of pieces of reference information and the question to the answer unit, and acquires a plurality of different answers to the input. The information processing device wherein the evaluation unit evaluates the accuracy or fluency of the multiple different answers.
10. accepting a question by a reception unit; a step of acquiring, by a first acquisition unit, cases related to the question from a search unit or generating cases based on information acquired from the search unit; a step of inputting, by a second acquisition unit, the case and the question acquired or generated by the first acquisition unit into an answer unit and acquiring an answer to the input; An information processing method comprising:
11. A program to be installed on an information processing device, An information processing device in which the program is installed A reception function to accept questions, a first acquisition function for acquiring cases related to the query from a search unit or generating cases based on information acquired from the search unit; a second acquisition function that performs input to an answer section using the case and the question acquired or generated by the first acquisition function, and acquires an answer to the input; A program characterized by executing the above.
Citation Information
Patent Citations
Answer generation device, answer generation method, and program
JP2022161375A
Persona chatbot control method and system
JP2022180282A
Method for automatically creating a collection of questions and answers, its program and recording medium
JP7271019B1
JPP7313757B
JPP7325152B
Cited By
Know-how research support device and application program
JP2026095959A
