Knowledge Q&A Method, Device, Equipment and Storage Medium

Abstract

The present disclosure provides a knowledge Q&A method, apparatus, device, and storage medium, relating to the field of artificial intelligence technologies, particularly to deep learning and natural language processing technologies, and especially to large model technologies. The specific implementation solution is as follows: obtain an inquiry statement, historical associated statements of the inquiry statement, and a knowledge base; wherein, the knowledge base stores Q&A data generated based on reference text content; generate an inquiry derivative statement of the inquiry statement according to the inquiry statement and the historical associated statements; query the knowledge base for Q&A data matching the inquiry derivative statement; and generate target answer data for the inquiry statement according to the matching result. According to the technology of the present disclosure, the accuracy of knowledge Q&A results is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence technologies, particularly to deep learning and natural language processing technologies, and especially to large model technologies. Background Art

[0002] With the continuous development of artificial intelligence, intelligent question answering has been applied to service fields such as voice assistants, intelligent customer service, and online consultations. Intelligent question answering refers to taking natural language understanding as the core, performing semantic analysis on the input inquiry statement, then matching relevant questions through semantic retrieval or dialogue management and other technologies in a large-scale knowledge base, and finally generating and replying answers through natural language generation technology. Summary of the Invention

[0003] The present disclosure provides a knowledge question answering method, apparatus, device, and storage medium with better accuracy.

[0004] According to one aspect of the present disclosure, there is provided a knowledge question answering method, including:

[0005] Obtaining an inquiry statement, historical associated statements of the inquiry statement, and a knowledge base; wherein, the knowledge base stores question and answer data generated based on reference text content;

[0006] Generating an inquiry derivative statement of the inquiry statement according to the inquiry statement and the historical associated statements;

[0007] Querying the knowledge base for question and answer data matching the inquiry derivative statement;

[0008] Generating target answer data for the inquiry statement according to the matching result.

[0009] According to another aspect of the present disclosure, there is also provided an electronic device, including:

[0010] At least one processor; and

[0011] A memory communicatively connected to the at least one processor; wherein,

[0012] The memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute any one of the knowledge question answering methods provided by the embodiments of the present disclosure.

[0013] According to another aspect of the present disclosure, there is also provided a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute any one of the knowledge question answering methods provided by the embodiments of the present disclosure.

[0014] According to the technology of the present disclosure, the accuracy of the knowledge Q&A results is improved.

[0015] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings are used to better understand the solution and do not constitute a limitation to the present disclosure. Among them:

[0017] Figure 1 is a flowchart of a knowledge Q&A method provided by an embodiment of the present disclosure;

[0018] Figure 2 is a flowchart of another knowledge Q&A method provided by an embodiment of the present disclosure;

[0019] Figure 3 is a flowchart of another knowledge Q&A method provided by an embodiment of the present disclosure;

[0020] Figure 4 is a structural diagram of a knowledge Q&A device provided by an embodiment of the present disclosure;

[0021] Figure 5 is a block diagram of an electronic device for implementing the knowledge Q&A method of the embodiment of the present disclosure. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0022] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to facilitate understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0023] The knowledge Q&A method and the knowledge Q&A device provided by the embodiments of the present disclosure are applicable to the application scenario of knowledge Q&A in the process of human-computer interaction. Each knowledge Q&A method provided by the embodiments of the present disclosure can be executed by a knowledge Q&A device. The device can be implemented by software and / or hardware and is specifically configured in an electronic device. The electronic device can be a smart terminal, such as a mobile phone, a tablet, a smart speaker, a smart wearable device, or a customer service robot, etc. The present disclosure makes no limitation thereto.

[0024] For ease of understanding, the knowledge Q&A method will be described in detail first.

[0025] See Figure 1 The knowledge Q&A method shown includes:

[0026] S101. Obtain an inquiry statement, historical associated statements of the inquiry statement, and a knowledge base; wherein, the knowledge base stores Q&A data generated based on reference text content.

[0027] Among them, the inquiry statement can be a statement carrying an inquiry question input during knowledge Q&A. Among them, the inquiry statement can be text data, or the text data conversion result corresponding to other forms of data (such as voice, etc.), and the present disclosure does not make any limitation thereto.

[0028] Among them, the historical associated statements of the inquiry statement can be historical inquiry statements related to the inquiry statement in the same context. For example, they can be historical inquiry statements adjacent to the input time and / or input times of the inquiry statement. Among them, the historical inquiry statements adjacent to the input time can be inquiry statements input within a historical preset time period before the input time of the inquiry statement; the historical inquiry statements adjacent to the input times can be inquiry statements input within a historical preset number threshold before the input times of the inquiry statement. Among them, the historical preset time period or historical preset number threshold can be set by technicians according to needs or empirical values, or determined through a large number of experiments, and the present disclosure does not make any limitation thereto.

[0029] It should be noted that, in order to ensure the accuracy of the knowledge Q&A result, generally, the inquiry statement and the historical associated statements of the inquiry statement are inquiry statements input by the same account or the same inquirer.

[0030] Among them, at least one Q&A data is pre-stored in the knowledge base for use as the basis for determining answer data during the knowledge Q&A process. Among the Q&A data, reference answer data can be included; or, optionally, the Q&A data can further include reference question data corresponding to the reference answer data. The reference text content can be the text data used to generate the Q&A data.

[0031] Exemplarily, the reference text content can be directly obtained, or rich media data can be obtained, the text content of the rich media data can be extracted, and the extraction result can be used as the reference text content; Q&A data is generated based on the reference text content; the generated Q&A data is stored in a pre-constructed knowledge base for subsequent query matching of the Q&A data. Among them, the rich media data can carry information in at least one media form such as text, picture, voice, video, and file.

[0032] Optionally, the data carried in the reference text content can be directly used as the Q&A data; or optionally, the reference text content can be processed, and the processing result can be used as the Q&A data.

[0033] It should be noted that the execution device for building the knowledge base and the execution device for performing knowledge answering can be the same or different, and the present disclosure does not make any limitation thereto.

[0034] It is worth noting that the acquisition of the inquiry statement and the historical associated statements of the inquiry statement, and the acquisition of the knowledge base can be executed successively, simultaneously or crosswise. The present disclosure does not make any limitation on the specific acquisition timing of different data, and only needs to ensure that the corresponding data can be acquired before performing the corresponding operations.

[0035] S102. Generate an inquiry derivative statement of the inquiry statement according to the inquiry statement and the historical associated statements.

[0036] Among them, the inquiry derivative statement can be an inquiry statement without ambiguity and without content omission.

[0037] Since the inquiry statement and the historical associated statements have the same context, the historical associated statements can supplement the omitted content in the inquiry statement to complete the content of the inquiry statement; and / or can supplement the referential content in the inquiry statement to resolve the reference in the inquiry statement. Therefore, supplementing the content of the inquiry statement and / or resolving the reference according to the historical associated statements can generate an inquiry derivative statement without ambiguity and without content omission, laying a foundation for improving the accuracy of subsequent knowledge answering results.

[0038] In an alternative embodiment, semantic analysis can be performed on the inquiry statement and the historical associated statements, and content supplementation and / or reference resolution can be performed on the inquiry statement according to the semantic analysis results to obtain the inquiry derivative statement. Among them, the semantic analysis can include at least one of syntactic analysis and sentence component analysis, etc.

[0039] In another alternative embodiment, an inquiry derivative statement can be generated based on a large language model according to the inquiry statement and the historical associated statements. Among them, the large language model is learned based on training text data under different natural language tasks.

[0040] A large model can be understood as a neural network model with a large number of parameters (such as hundreds of millions). Among them, a large language model is a deep learning model trained with a large amount of training text data. Since the training text data used in training the large language model corresponds to at least one natural language task, the trained large language model has good natural language processing capabilities and can realize the generation of natural language texts or the understanding of language texts, etc. Among them, natural language tasks can include at least one of text generation tasks, knowledge answering tasks, text parsing tasks, dialogue tasks, text continuation tasks, etc. The present disclosure does not make any limitation on the specific network structure of the large language model used, which can be a knowledge-enhanced large language model that can continuously learn by integrating from massive data and large-scale knowledge, has technical features such as knowledge enhancement, retrieval enhancement or dialogue enhancement, and has information extraction and knowledge generation capabilities.

[0041] Exemplarily, in order to further improve the accuracy of the generated result corresponding to the inquiry-derived statement in the process of generating the inquiry-derived statement by the large language model, the training text data under the text generation task and the text-derived result corresponding to the pre-annotated training text data can also be used to fine-tune the parameters of the pre-trained large language model, so that the adjusted large language model is more suitable for the text generation task. Correspondingly, the fine-tuned large language model is used to generate inquiry-derived data.

[0042] It can be understood that, based on the method of using the large language model to replace traditional semantic analysis to generate inquiry-derived statements, since the large language model has good natural language processing capabilities, the accuracy of the generated inquiry-derived statements is higher, which helps to improve the accuracy of knowledge answering results.

[0043] S103. Query the question-and-answer data in the knowledge base that matches the inquiry-derived statement.

[0044] Use an inquiry-derived statement that is unambiguous and carries more comprehensive information to replace the inquiry statement, and query and match the question-and-answer data in the knowledge base, so that the matching result is more accurate.

[0045] Exemplarily, the inquiry-derived statement can be searched and matched with each question-and-answer data in the knowledge base by means of vector similarity matching, etc. Among them, the vector construction method of the question-and-answer data and the inquiry-derived statement, and the determination method of the similarity between the two can be implemented by at least one of the existing technologies, and the present disclosure does not make any limitation on this.

[0046] In an alternative embodiment, a confidence level between the query-derived statement and the Q&A data may be introduced to measure the semantic matching between the query-derived statement and different Q&A data; the Q&A data with a confidence level exceeding a preset confidence threshold is selected as the matching result of the query-derived statement. Among them, the preset confidence threshold can be set or adjusted by a technician according to needs or empirical values, or determined through a large number of experiments; the confidence level can be implemented by at least one confidence level determination method in the prior art, and the present disclosure does not make any limitation thereto.

[0047] S104. Generate target answer data for the query statement according to the matching result.

[0048] If the matching is successful, that is, there is Q&A data in the knowledge base that matches the query-derived statement, the answer data in the matching result is directly used as the target answer data for the query statement; alternatively, the matching result is processed so that the matching result can conform to the context of the query statement, and the processing result is used as the target answer data for the query statement.

[0049] Furthermore, if the matching fails, the standard answer data of the query-derived statement input manually is obtained, and the obtained result is used as the target answer data. To avoid subsequent matching failures, the query-derived statement and the standard answer data can be stored in the knowledge base as a Q&A pair for subsequent query use.

[0050] In the embodiment of the present disclosure, by introducing the historical association data of the query statement and combining the query statement to generate a query-derived statement, since the historical association data carries the referential information or omitted information in the query statement, etc., the query-derived statement generated is more informative than the query statement and eliminates referential ambiguity, etc. Therefore, the present disclosure uses the query-derived statement to replace the query statement to search and match the Q&A data in the knowledge base, and the matching result has higher accuracy; correspondingly, according to the matching result, the target answer data of the query statement is generated, which improves the accuracy of the generated target answer data and further improves the accuracy of the knowledge Q&A result.

[0051] Based on the above technical solutions, the present disclosure also provides an alternative embodiment, in which the content included in the knowledge base is optimized and improved. It should be noted that for the parts not detailed in the embodiment of the present disclosure, reference may be made to the relevant descriptions of other embodiments.

[0052] See Figure 2 A knowledge Q&A method shown in

[0053] S201. Obtain an inquiry statement, historical associated statements of the inquiry statement, and a knowledge base; the knowledge base includes a first knowledge base and / or a second knowledge base; the first knowledge base stores Q&A pairs included in the reference text content; the second knowledge base stores knowledge generation data of the reference text content.

[0054] Among them, a Q&A pair can be understood as a data pair constructed by question data and corresponding answer data of the question data; knowledge generation data can be understood as data generated by performing knowledge understanding and in-depth mining on the reference text content.

[0055] S202. Generate an inquiry derivative statement of the inquiry statement according to the inquiry statement and historical associated statements.

[0056] S203. Query Q&A data matching the inquiry derivative statement in the knowledge base.

[0057] S204. Generate target answer data of the inquiry statement according to the matching result.

[0058] In an optional embodiment, the knowledge base may only include the first knowledge base; correspondingly, when querying Q&A data matching the inquiry derivative statement in the knowledge base, Q&A data matching the inquiry derivative statement can be directly queried in the first knowledge base.

[0059] Exemplarily, for reference text content containing Q&A pairs, the Q&A pairs included in the reference text content are stored in a pre-constructed first knowledge base as the basis for subsequent querying and matching of Q&A data.

[0060] Optionally, the reference text content can be classified and marked manually in advance according to whether it contains Q&A pairs; according to the classification identifier, it is determined whether the reference text content contains Q&A pairs. Among them, the specific presentation method of the classification identifier in the present disclosure is not limited in any way, and it is only necessary to ensure that the text categories containing Q&A pairs are different from the classification identifiers of the text pairs not included.

[0061] Alternatively, optionally, it is also possible to automatically identify whether the reference text content contains Q&A pairs, so as to distinguish between reference text content containing and not containing Q&A pairs, and at the same time realize the automatic identification of Q&A pairs.

[0062] Exemplarily, the Q&A pairs in the reference text content can be identified according to at least one of the document structure template used in the reference content document, the document type of the reference content document, and the use of preset delimiters in the reference content document. Among them, the reference text document carries the reference content document.

[0063] In an alternative embodiment, different document structure templates can be set in advance. Among them, the reference content documents containing Q&A pairs and the reference content documents not containing Q&A pairs adopt different document structure templates. Therefore, it is possible to identify whether the reference text content contains Q&A pairs based on the document structure template adopted by the reference content document.

[0064] Optionally, the reference content documents can include two categories: those containing only Q&A pairs and those not containing Q&A pairs; correspondingly, the Q&A pairs in the reference text content can be directly identified based on the document structure template adopted by the reference content document.

[0065] Or optionally, the reference content documents can include two categories: those containing Q&A pairs and those not containing Q&A pairs; correspondingly, it is possible to identify whether the reference text content contains Q&A pairs from the document structure template adopted by the reference content document; for the reference content documents of the category containing Q&A pairs, the Q&A pairs corresponding to the reference text content can be identified according to the Q&A pair distribution area in the adopted document structure template.

[0066] In another alternative embodiment, during the process of generating the reference content document, a unique preset delimiter can be set at the position or area containing Q&A pairs; correspondingly, it is possible to identify whether the reference document content contains Q&A pairs based on the presence or absence of the preset delimiter in the reference content document. Further, the distribution area of Q&A pairs can also be determined according to the specific position of the preset delimiter, and the Q&A pairs corresponding to the reference text content can be identified according to the Q&A pair distribution area; or, the Q&A pairs corresponding to the reference text content can be identified according to the Q&A pair distribution area corresponding to the document structure template adopted by the reference content document.

[0067] In yet another alternative embodiment, in the execution device for performing the knowledge base construction process, a document upload component can be preset for uploading the reference content document. During the process of uploading the reference content document, the document type of the reference content document needs to be set as required. Among them, the document types include Q&A pair documents and ordinary documents. Among them, the reference content documents of the Q&A pair document category must contain Q&A pairs; ordinary documents are prohibited from containing Q&A pairs. Therefore, it is possible to identify whether the reference text content contains Q&A pairs based on the document type of the reference content document.

[0068] Optionally, if the Q&A pair document contains only Q&A pairs, the Q&A pair data in the reference text document can also be directly obtained.

[0069] Alternatively, if the Q&A pair document includes Q&A pairs and other content, the Q&A pair distribution area corresponding to the document structure template adopted by the reference content document retrieval can be used to identify the reference text content corresponding to the Q&A pair; or the Q&A pair distribution area can be determined through the positions with preset delimiters added in the reference content document, and the reference text content corresponding to the Q&A pair can be identified according to the Q&A pair distribution area.

[0070] It can be understood that by introducing a reference content document carrying a reference content document and identifying Q&A pairs in the reference text content according to at least one of the document structure template used in the reference content document, the document type of the reference content document, and the use of preset delimiters in the reference content document, the richness and diversity of the identification of Q&A pairs in the reference text content are improved, thereby improving the diversity and flexibility of the first knowledge base construction process.

[0071] It can be understood that since the Q&A data in the first knowledge base are Q&A pairs included in the reference text content itself, the Q&A data results matched by querying the first knowledge base are more accurate, and thus the accuracy of the target answer data of the inquiry statement generated based on the matching result is improved.

[0072] In another alternative embodiment, the knowledge base may only include a second knowledge base; correspondingly, when querying the Q&A data matching the inquiry derivative statement in the knowledge base, the Q&A data matching the inquiry derivative statement can be directly queried in the second knowledge base.

[0073] Exemplarily, knowledge generation can be performed on the reference text content based on a large language model to obtain knowledge generation data including Q&A data; wherein, the large language model is learned based on training text data under different natural language tasks.

[0074] The so-called large model can be understood as a neural network model with a large number of parameters (such as hundreds of millions of scales). Among them, the large language model is a deep learning model trained with a large amount of training text data. Since the training text data used in training the large language model corresponds to at least one natural language task, the trained large language model has good natural language processing capabilities and can realize the generation of natural language text or the understanding of language text, etc. Among them, the natural language tasks can include at least one of text generation tasks, knowledge Q&A tasks, text parsing tasks, dialogue tasks, and text continuation tasks, etc. The present disclosure does not make any limitation on the specific network structure of the large language model used, and it can be a knowledge-enhanced large language model that can continuously learn from massive data and large-scale knowledge, has technical features such as knowledge enhancement, retrieval enhancement, or dialogue enhancement, and has information extraction and knowledge generation capabilities.

[0075] It should be noted that the large language model used in the knowledge generation process and the large language model used in the aforementioned query-derived statement generation process may be the same or different, and the present disclosure does not make any limitation in this regard. In order to reduce the number of large language models used and improve the convenience of the knowledge Q&A process, in a specific implementation, the same large language model can be used to perform knowledge generation and query-derived statement generation respectively.

[0076] Exemplarily, in order to further improve the accuracy of the generated results corresponding to the knowledge generation data in the knowledge generation data generation process by the large language model, training text data under the text parsing task and the text parsing results corresponding to the pre-annotated training text data can also be used to fine-tune the parameters of the pre-trained large language model, so that the adjusted large language model is more suitable for the text parsing task. Correspondingly, the fine-tuned large language model is used to generate knowledge generation data.

[0077] It can be understood that knowledge generation is performed on the reference text content based on the large language model, so that the Q&A-related data in the reference text content can be extracted from the obtained knowledge generation data, and the semantic information hidden in the reference text content can be mined, improving the richness and accuracy of the knowledge generation data, and thus improving the richness and accuracy of the data stored in the second knowledge base. Correspondingly, query matching of Q&A data is performed based on the second knowledge base, improving the accuracy of the Q&A data matching result, and further helping to improve the accuracy of the knowledge Q&A result.

[0078] Since the knowledge generation data is generated based on the large language model, there may be certain differences in the sentence structures of the knowledge generation data generated from different reference text contents. Moreover, the target answer data generated based on the matched Q&A data may also not match the context of the query statement. To overcome the above problems, in an alternative embodiment, the target answer data for the query statement can also be generated based on the large language model according to the matching result and the query-derived statement. Among them, the large language model is learned based on the training text data under different natural language tasks.

[0079] A large model can be understood as a neural network model with a large number of parameters (such as in the order of hundreds of millions). Among them, a large language model is a deep learning model trained using a large amount of training text data. Since the training text data used in training the large language model corresponds to at least one natural language task, the trained large language model has good natural language processing capabilities and can realize the generation of natural language texts or the understanding of language texts, etc. Among them, natural language tasks can include at least one of text generation tasks, knowledge answering tasks, text parsing tasks, dialogue tasks, and text continuation tasks, etc. The present disclosure does not make any limitations on the specific network structure of the large language model used, and it can be a knowledge-enhanced large language model that can continuously learn by integrating from massive data and large-scale knowledge, has technical features such as knowledge enhancement, retrieval enhancement, or dialogue enhancement, and has information extraction and knowledge generation capabilities.

[0080] It should be noted that the large language model used in the inquiry-derived statement generation process, the large language model used in the knowledge generation process, and the large language model used in the target answer data generation process can be the same or at least partially different, and the present disclosure does not make any limitations on this. In order to reduce the number of large language models used and improve the convenience of the knowledge answering process, in a specific implementation, the same large language model can be used to perform knowledge generation, inquiry-derived statement generation, and target answer data generation respectively.

[0081] Exemplarily, in order to further improve the accuracy of the generated result corresponding to the target answer data in the generation process of the large language model, the training text data under the text generation task and the answer generation results corresponding to the pre-annotated training text data can also be used to fine-tune the parameters of the pre-trained large language model, so that the adjusted large language model is more suitable for the text generation task. Correspondingly, the fine-tuned large language model is used to generate the target answer data.

[0082] It can be understood that since the inquiry-derived statement can reflect the real context of the inquiry statement to a certain extent, the generation of the target answer data based on the inquiry-derived statement combined with the matching result (that is, the knowledge generation data including the question-and-answer data) makes the generated target answer data more in line with the context of the inquiry statement. At the same time, using the large language model to generate the target answer data can effectively avoid the situation of unclear reference or ambiguity in the target answer data, and further improve the accuracy of the target answer data.

[0083] In a specific embodiment, based on a large language model, according to the matching result and the query-derived statement, the target answer data of the query statement can be generated, which may include: generating answer template data according to the matching result and the query-derived statement; based on the large language model, generating the target answer data of the query statement according to the answer template data.

[0084] Among them, the answer template data can be a standard template determined according to the syntactic category of the query-derived statement, and is used to restrict the syntactic structure of the subsequent generated target answer data; among them, the standard templates corresponding to different syntactic categories are different, and can be set by technicians according to needs or empirical values, or determined through a large number of experiments. The present disclosure does not make any limitations in this regard.

[0085] It can be understood that by restricting the syntactic structure of the generated target answer data through the answer template data, the situation of generating target answer data with chaotic structures can be avoided. In addition, since the answer template data is generated based on the matching result and the query-derived statement, the generated answer template data also carries answer semantic information. Therefore, the target answer data generated based on the answer template data can avoid the situation of semantic errors in the target answer data and improve the accuracy of the target answer data.

[0086] It should be noted that the execution device for constructing the first knowledge base and the execution device for constructing the second knowledge base can be the same or different, and the present disclosure does not make any limitations in this regard.

[0087] In another alternative embodiment, the knowledge base may include a first knowledge base and a second knowledge base; correspondingly, when querying the Q&A data matching the query-derived statement in the knowledge base, the Q&A data matching the query-derived statement can be directly queried in the first knowledge base and / or the second knowledge base.

[0088] For the reference text data containing Q&A pairs, since the Q&A pairs in the first knowledge base and the knowledge generation data corresponding to the Q&A pairs in the second knowledge base usually have the same content or the same semantics, there is a situation where at least part of the data stored in the first knowledge base and the second knowledge base overlaps. For this part of the content, the first knowledge base or the second knowledge base can be used for query matching of Q&A data.

[0089] During the use of the second knowledge base, in order to further improve the accuracy of the target Q&A data and enhance the interaction experience, a large language model is usually introduced to generate the target answer data according to the matching result and the query-derived statement, which will inevitably bring a certain waiting time and reduce the knowledge Q&A efficiency.

[0090] As for the reference text data that does not contain question-answer pairs, since there is no relevant content stored in the first knowledge base, the second knowledge base stores some question-answer data that is not stored in the second knowledge base. When the first knowledge base is used for query matching, no matching results can be obtained at all, which affects the feedback of the knowledge question-answering results.

[0091] In view of this, when the knowledge base includes a first knowledge base and a second knowledge base, the question and answer data that matches the query derivative statement can be first queried in the first knowledge base; if the match is successful in the first knowledge base, the matching question and answer data is directly used as the target answer data for feedback, thereby improving the efficiency of knowledge question and answer.

[0092] If the match fails in the first knowledge base, the second knowledge base is searched for question and answer data that matches the query derivative statement, and the target answer data is subsequently generated based on the matched question and answer data and the query derivative statement. The advantage of this is that the question and answer data is first matched in the first knowledge base, thereby improving the efficiency of knowledge question and answer. When the first knowledge base cannot guarantee the smooth execution of knowledge question and answer, the second knowledge base with richer and more comprehensive question and answer data is used as a substitute. On the basis of sacrificing a certain degree of knowledge question and answer efficiency, the accuracy of the knowledge question and answer results is guaranteed, thus achieving a balance between the efficiency of knowledge question and answer and the accuracy of knowledge question and answer results.

[0093] Exemplarily, if the match fails in the second knowledge base, the standard answer data of the manually input query derivative statement is obtained, and the obtained result is used as the target answer data; the query derivative statement and the standard answer data are used as a new question-answer pair and stored in the first knowledge base.

[0094] When the first knowledge base and the second knowledge base cannot be matched successfully, rashly using the question and answer data in the first knowledge base or the question and answer data in the second knowledge base to generate the target answer data will most likely result in factual errors. The method of manually inputting standard answer data is used to perform manual intervention to achieve a knowledge question and answer guarantee, thereby improving the accuracy of the knowledge question and answer results and the knowledge question and answer experience. At the same time, the query-derived data and the standard answer data are stored in the first knowledge base as new question and answer pairs, so that in the subsequent same or similar knowledge question and answer processes, no manual intervention is required, and content replies can be made in a timely, efficient and accurate manner, which helps to improve the accuracy of the corresponding results of the subsequent knowledge question and answer process and the efficiency of question and answer. In addition, the manual intervention method adopted by the present disclosure is convenient, fast and highly operational.

[0095] In the embodiments of the present disclosure, by refining the knowledge base into a first knowledge base storing the question-and-answer pairs included in the reference text content and / or a second knowledge base storing the knowledge generation data of the reference text content, the richness and diversity of the knowledge base are improved, and the diversity and flexibility of the knowledge question-and-answer process are improved by using the first knowledge base and / or the second knowledge base.

[0096] On the basis of the above technical solutions, the present disclosure also provides a preferred embodiment in which the knowledge question-and-answer process based on the large language model is described in detail. It should be noted that for the parts not detailed in the embodiments of the present disclosure, reference may be made to the relevant descriptions of other embodiments.

[0097] See Figure 3 The knowledge question-and-answer method shown includes: a knowledge base construction stage and a knowledge question-and-answer stage.

[0098] Among them, the knowledge base construction stage includes:

[0099] S301. File acquisition: Acquire the file to be processed transmitted through the preset upload interface of the terminal device.

[0100] S302. File parsing: Parse the file to be processed to obtain the reference text content; continue to execute S303A and S303B.

[0101] S303A. First knowledge base generation: Add the question-and-answer pairs included in the reference text content as question-and-answer data to the first knowledge base.

[0102] S303B. Knowledge generation: Perform knowledge generation on the reference file content based on the large language model to obtain knowledge generation data including question-and-answer data; continue to execute S304.

[0103] S304. Second knowledge base generation: Add the knowledge generation data to the second knowledge base.

[0104] Among them, the knowledge question-and-answer stage includes:

[0105] S305. Inquiry statement acquisition: Acquire the inquiry statement transmitted through the preset inquiry interface of the terminal device.

[0106] S306. Inquiry statement completion: Based on the large language model, complete the content of the inquiry statement and / or resolve the reference according to the inquiry statement and the historical inquiry statement associated with the inquiry statement to obtain an inquiry derivative statement.

[0107] S307. Knowledge Retrieval: Retrieve in the first knowledge base to recall Q&A data that matches the query-derived statement. If there is Q&A data in the recall result with a confidence level greater than the preset confidence threshold, then use the answer data in this Q&A data as the target answer data for the query statement. If there is no Q&A data in the recall result with a confidence level greater than the preset confidence threshold, then based on the preset retrieval and recall model, recall Q&A data that matches the query-derived statement in the second knowledge base.

[0108] Among them, the preset recall model can be implemented by using at least one machine learning model with data retrieval functions in the prior art. The present disclosure does not make any limitations on the specific network structure and training method of the preset retrieval and recall model. Among them, the preset confidence threshold can be set or adjusted by those skilled in the art according to needs or empirical values, or set through a large number of experiments. The present disclosure does not make any limitations on this.

[0109] Exemplarily, when retrieving using the first knowledge base, only the question data in the Q&A data can be searched and matched. When retrieving using the second knowledge base, a full-text retrieval method can be used to search and match the entire knowledge generation data.

[0110] S308. Semantic Fine Ranking: Rank each recall result according to the confidence level of each Q&A data in the recall result; then continue to execute S309.

[0111] Among them, the confidence level is used to characterize the semantic matching situation between the semantic information of the recalled Q&A statement and the query-derived statement.

[0112] S309. Answer Generation: If there is Q&A data in the recall result with a confidence level greater than the preset confidence threshold, then generate answer template data based on the Q&A data and the query-derived statement. Based on the large language model, generate the target answer data for the query statement according to the answer template data.

[0113] S310. Manual Intervention Backup: If there is no Q&A data in the recall result with a confidence level greater than the preset confidence threshold, then obtain the standard answer data input manually, and use the query-derived statement and the standard answer data as a new Q&A pair, and add it to the first knowledge base for subsequent use.

[0114] Embodiments of the present disclosure can greatly reduce factual errors in the knowledge Q&A process and improve the accuracy of knowledge Q&A results by means of a first knowledge base, a second knowledge base, and manual intervention as a fallback. At the same time, adopting the technical solution of the present disclosure does not require a large amount of human resources for model training and data annotation, saving labor costs. Further, first retrieve in the first knowledge base to ensure the efficiency of knowledge Q&A. In the case where the first knowledge base cannot meet the requirements, use the second knowledge base for supplementary retrieval to ensure the accuracy of knowledge Q&A results, achieving both the efficiency of knowledge Q&A and the accuracy of knowledge Q&A results. Further, a large language model is introduced for data processing in multiple links of the knowledge Q&A process. Since the large language model has high natural language text generation ability and natural language text understanding ability, the accuracy of the large language model processing results is improved, which in turn contributes to the accuracy of knowledge Q&A results.

[0115] As an implementation of the above knowledge Q&A methods, the present disclosure also provides an optional embodiment of an execution device for implementing the above knowledge Q&A methods.

[0116] See Figure 4 The knowledge Q&A device 400 shown in the figure includes: a data acquisition module 401, an inquiry-derived statement generation module 402, a query matching module 403, and a target answer data generation module 404.

[0117] Among them,

[0118] The data acquisition module 401 is configured to acquire an inquiry statement, historical associated statements of the inquiry statement, and a knowledge base; wherein, the knowledge base stores Q&A data generated based on reference text content;

[0119] The inquiry-derived statement generation module 402 is configured to generate an inquiry-derived statement of the inquiry statement according to the inquiry statement and the historical associated statements;

[0120] The query matching module 403 is configured to query Q&A data matching the inquiry-derived statement in the knowledge base;

[0121] The target answer data generation module 404 is configured to generate target answer data of the inquiry statement according to the matching result.

[0122] In an embodiment of the present disclosure, by introducing historical associated data of an inquiry statement and combining it with the inquiry statement, an inquiry-derived statement is generated. Since the historical associated data carries referential information or omitted information in the inquiry statement, etc., the inquiry-derived statement generated carries more information than the inquiry statement and eliminates referential ambiguity, etc. Therefore, the present disclosure uses the inquiry-derived statement to replace the inquiry statement to search and match question-and-answer data in the knowledge base, and the matching result has higher accuracy. Based on this matching result, the target answer data of the inquiry statement is generated, improving the accuracy of the generated target answer data and further improving the accuracy of the knowledge question-and-answer result.

[0123] In an alternative embodiment, the inquiry-derived statement generation module 402 is specifically configured to:

[0124] Based on a large language model, generate the inquiry-derived statement according to the inquiry statement and the historical associated statement;

[0125] wherein, the large language model is learned based on training text data under different natural language tasks.

[0126] In an alternative embodiment, the knowledge base includes a first knowledge base and / or a second knowledge base;

[0127] The first knowledge base stores question-and-answer pairs included in the reference text content;

[0128] The second knowledge base stores knowledge generation data of the reference text content.

[0129] In an alternative embodiment, the apparatus 400 further includes a knowledge generation data generation module, specifically configured to:

[0130] Based on a large language model, perform knowledge generation on the reference text content to obtain knowledge generation data including question-and-answer data;

[0131] wherein, the large language model is learned based on training text data under different natural language tasks.

[0132] In an alternative embodiment, the apparatus 400 further includes a question-and-answer pair recognition module, specifically configured to:

[0133] Identify question-and-answer pairs in the reference text content according to at least one of a document structure template used in the reference content document, a document type of the reference content document, and a usage condition of a preset delimiter in the reference content document;

[0134] wherein, the reference text document carries the reference content document.

[0135] In an alternative embodiment, if the knowledge base includes a first knowledge base and a second knowledge base, the query matching module 403 includes:

[0136] A first query matching unit, configured to query the Q&A data that matches the inquiry-derived statement in the first knowledge base;

[0137] A second query matching unit, configured to, if the matching fails in the first knowledge base, query the Q&A data that matches the inquiry-derived statement in the second knowledge base.

[0138] In an alternative embodiment, if the Q&A data is a matching result in the second knowledge base, the target answer data generation module 404 is specifically configured to:

[0139] Based on a large language model, generate the target answer data of the inquiry statement according to the matching result and the inquiry-derived statement;

[0140] Wherein, the large language model is learned based on training text data under different natural language tasks.

[0141] In an alternative embodiment, the target answer data generation module 404 includes:

[0142] An answer template data generation unit, configured to generate answer template data according to the matching result and the inquiry-derived statement;

[0143] A target answer data generation unit, configured to generate the target answer data of the inquiry statement based on the large language model according to the answer template data.

[0144] In an alternative embodiment, the apparatus 400 further includes:

[0145] An artificial intervention module, configured to, if the matching fails in the second knowledge base, obtain the standard answer data of the inquiry-derived statement input manually, and use the obtained result as the target answer data;

[0146] A new Q&A pair storage module, configured to store the inquiry-derived statement and the standard answer data as a new Q&A pair into the first knowledge base.

[0147] The above knowledge Q&A apparatus can execute the knowledge Q&A method provided in any embodiment of the present disclosure, and has corresponding functional modules and beneficial effects for executing each knowledge Q&A method.

[0148] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of data such as inquiry statements, historical association data of inquiry statements, knowledge bases, and reference text contents comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0149] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0150] Figure 5 FIG. shows a schematic block diagram of an exemplary electronic device 500 that can be used to implement the embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, for example, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely exemplary and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0151] As Figure 5 shown, the device 500 includes a computing unit 501 that can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. In the RAM 503, various programs and data required for the operation of the device 500 can also be stored. The computing unit 501, the ROM 502, and the RAM 503 are connected to each other via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0152] A plurality of components in the device 500 are connected to the I / O interface 505, including: an input unit 506, such as a keyboard, a mouse, etc.; an output unit 507, such as various types of displays, speakers, etc.; a storage unit 508, such as a magnetic disk, an optical disk, etc.; and a communication unit 509, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 509 allows the device 500 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0153] The computing unit 501 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 executes the various methods and processes described above, such as the knowledge Q&A method. For example, in some embodiments, the knowledge Q&A method can be implemented as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 508. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 500 via the ROM 502 and / or the communication unit 509. When the computer program is loaded into the RAM 503 and executed by the computing unit 501, one or more steps of the knowledge Q&A method described above can be executed. Alternatively, in other embodiments, the computing unit 501 can be configured to execute the knowledge Q&A method by any other suitable means (e.g., by means of firmware).

[0154] The various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special-purpose or general-purpose programmable processor, receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0155] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to the processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program codes can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0156] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. The machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. The machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of the machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0157] In order to provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, speech input, or tactile input).

[0158] The systems and techniques described herein can be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0159] A computer system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is created by computer programs running on respective computers and having a client-server relationship with each other. The server can be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system, solving the defects of difficult management and weak business scalability existing in traditional physical hosts and VPS services. The server can also be a server of a distributed system or a server combined with blockchain.

[0160] Artificial intelligence is a discipline that studies how to make a computer simulate certain human thinking processes and intelligent behaviors (such as learning, reasoning, thinking, planning, etc.), including both hardware-level technologies and software-level technologies. Artificial intelligence hardware technologies generally include technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, and big data processing; artificial intelligence software technologies mainly include several major directions such as computer vision technology, speech recognition technology, natural language processing technology, machine learning / deep learning technology, big data processing technology, and knowledge graph technology.

[0161] Cloud computing refers to a technical system that accesses an elastic and scalable shared physical or virtual resource pool through a network. The resources can include servers, operating systems, networks, software, applications, and storage devices, etc., and the resources can be deployed and managed in a demand-based and self-service manner. Through cloud computing technology, it can provide efficient and powerful data processing capabilities for the application and model training of technologies such as artificial intelligence and blockchain.

[0162] It should be understood that various forms of processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions provided in this disclosure can be achieved, and no limitations are imposed herein.

[0163] The above specific embodiments do not constitute a limitation on the protection scope of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure shall be included within the protection scope of this disclosure.

Claims

1. A knowledge Q&A method, comprising: Obtaining an inquiry statement, historical associated statements of the inquiry statement, and a knowledge base; wherein, the knowledge base stores Q&A data generated based on reference text content; the knowledge base includes a first knowledge base and a second knowledge base; the first knowledge base stores Q&A pairs included in the reference text content; the second knowledge base stores knowledge generation data of the reference text content; the knowledge generation data is data generated by performing knowledge understanding and in-depth mining on the reference text content; the reference text content is determined by extracting text content from rich media data, and the rich media data carries information of at least one media form of pictures, videos, and files; the inquiry statement and the historical associated statements of the inquiry statement are inquiry statements input by the same account or the same inquirer; Based on a large language model, supplementing the omitted content and referring content in the inquiry statement according to the historical associated statements to generate an inquiry derivative statement; wherein, the inquiry derivative statement is an inquiry statement without ambiguity and without missing content; Querying the Q&A data matching the inquiry derivative statement in the first knowledge base; If the matching fails in the first knowledge base, querying the Q&A data matching the inquiry derivative statement in the second knowledge base; Generating target answer data for the inquiry statement according to the matching result, including: processing the matching result to make the matching result conform to the context situation of the inquiry statement, and using the processing result as the target answer data for the inquiry statement; wherein, the Q&A data with a confidence level exceeding a preset confidence threshold between the inquiry derivative statement and the Q&A data is used as the matching result; Wherein, the knowledge generation data is generated in the following manner: Based on the large language model, performing knowledge generation on the reference text content to obtain knowledge generation data including Q&A data; wherein, the large language model is learned based on training text data under different natural language tasks; the natural language tasks include text generation tasks, knowledge Q&A tasks, text parsing tasks, dialogue tasks, and text continuation tasks; Wherein, if the Q&A data is the matching result in the second knowledge base, then generating the target answer data for the inquiry statement according to the matching result further includes: Generating answer template data according to the matching result and the inquiry derivative statement; wherein, the answer template data is a standard template determined according to the syntactic category of the inquiry derivative statement, and the answer template data is used to restrict the syntactic structure of the target answer data to be generated subsequently; Based on the large language model, generating the target answer data for the inquiry statement according to the answer template data.

2. The method according to claim 1, wherein, The Q&A pairs are identified in the following manner: Identifying the Q&A pairs in the reference text content according to at least one of the document structure template used in the reference content document, the document type of the reference content document, and the usage situation of preset delimiters in the reference content document; Among them, the reference text content is carried in the reference content document.

3. The method according to claim 1, wherein, The method further includes: If the matching fails in the second knowledge base, obtain the standard answer data of the inquiry-derived statement input manually, and use the obtained result as the target answer data; Use the inquiry-derived statement and the standard answer data as a new question-and-answer pair and store them in the first knowledge base.

4. A knowledge Q&A device, including: A data acquisition module, configured to acquire an inquiry statement, historical associated statements of the inquiry statement, and a knowledge base; wherein, the knowledge base stores question-and-answer data generated based on reference text content; the knowledge base includes a first knowledge base and a second knowledge base; the first knowledge base stores question-and-answer pairs included in the reference text content; the second knowledge base stores knowledge generation data of the reference text content; the knowledge generation data is data generated by performing knowledge understanding and in-depth mining on the reference text content; the reference text content is determined by extracting text content from rich media data, and the rich media data carries information of at least one media form of pictures, videos, and files; the inquiry statement and the historical associated statements of the inquiry statement are inquiry statements input by the same account or the same inquirer; An inquiry-derived statement generation module, configured to supplement the omitted content and referential content in the inquiry statement based on a large language model according to the historical associated statements, and generate an inquiry-derived statement; wherein, the inquiry-derived statement is an inquiry statement without ambiguity and content omission; A query matching module, including: a first query matching unit and a second query matching unit; The first query matching unit is configured to query question-and-answer data matching the inquiry-derived statement in the first knowledge base; The second query matching unit is configured to, if the matching fails in the first knowledge base, query question-and-answer data matching the inquiry-derived statement in the second knowledge base; A target answer data generation module, configured to generate target answer data of the inquiry statement according to the matching result; The target answer data generation module is specifically configured to process the matching result so that the matching result can conform to the context of the inquiry statement, and use the processed result as the target answer data of the inquiry statement; wherein, the question-and-answer data with a confidence level exceeding a preset confidence threshold between the inquiry-derived statement and the question-and-answer data is used as the matching result; A knowledge generation data generation module, specifically configured to perform knowledge generation on the reference text content based on the large language model to obtain knowledge generation data including question-and-answer data; wherein, the large language model is learned based on training text data under different natural language tasks; the natural language tasks include text generation tasks, knowledge Q&A tasks, text parsing tasks, dialogue tasks, and text continuation tasks; Among them, if the question-and-answer data is the matching result in the second knowledge base, the target answer data generation module further includes: An answer template data generation unit, configured to generate answer template data according to the matching result and the inquiry-derived statement; A target answer data generation unit, configured to generate target answer data for the inquiry statement based on the large language model and according to the answer template data.

5. The device according to claim 4, wherein The device further includes a Q&A pair recognition module, specifically configured to: Identify Q&A pairs in the reference text content according to at least one of the document structure template used in the reference content document, the document type of the reference content document, and the usage of preset delimiters in the reference content document; Wherein, the reference text content is carried in the reference content document.

6. The apparatus according to claim 4, wherein, The device further includes: An artificial intervention module, configured to, if the matching fails in the second knowledge base, obtain the standard answer data of the inquiry-derived statement input manually, and use the obtained result as the target answer data; A new Q&A pair storage module, configured to store the inquiry-derived statement and the standard answer data as a new Q&A pair in the first knowledge base.

7. An electronic device, comprising: At least one processor; And A memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can execute the knowledge Q&A method according to any one of claims 1-3.

8. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause a computer to execute the knowledge Q&A method according to any one of claims 1-3.

9. A computer program product, comprising computer programs / instructions, and when the computer programs / instructions are executed by a processor, the steps of the knowledge Q&A method according to any one of claims 1-3 are implemented.

Images