Question and answer method and device based on large model, electronic equipment and storage medium

Abstract

The invention provides a question answering method and device based on a large model, electronic equipment and a storage medium, and relates to the technical field of computers, in particular to the technical field of artificial intelligence such as natural language processing, intelligent question answering and large models. According to the specific implementation scheme, based on a first large model, a first result corresponding to the query statement is determined, and the first result comprises a first reply corresponding to the query statement, first reply logic corresponding to the first reply and first background knowledge; based on the query statement, performing retrieval in a database, and determining at least one retrieval result corresponding to the query statement; determining the type of the knowledge conflict between the first result and the retrieval result; and based on the first result, the retrieval result and the type of the knowledge conflict, determining a target reply corresponding to the query statement.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and in particular to the field of artificial intelligence technology such as natural language processing, intelligent question and answer, and large models, and specifically to a question and answer method, device, electronic device, and storage medium based on a large model. Background Art

[0002] In the field of natural language processing, systems based on the Retrieval-Augmented Generation (RAG) model are widely used in tasks such as text generation and question-answering systems. The RAG model generates more accurate answers by combining the parameter knowledge internalized in the Large Language Model (LLM) with the external knowledge retrieved from an external database. However, when dealing with complex and long-tail questions, there may be conflicts between the model's parameter knowledge and external knowledge, which may lead to inaccurate answers. Summary of the invention

[0003] The present invention provides a question-answering method, device, electronic device and storage medium based on a large model.

[0004] According to a first aspect of the present disclosure, a question-answering method based on a large model is provided, comprising:

[0005] Based on the first large model, determine a first result corresponding to the query statement, wherein the first result includes a first answer corresponding to the query statement, a first answer logic corresponding to the first answer, and first background knowledge;

[0006] Based on the query statement, searching in a database to determine at least one search result corresponding to the query statement;

[0007] determining a type of knowledge conflict between the first result and the search result;

[0008] Based on the first result, the search result and the type of the knowledge conflict, a target response corresponding to the query statement is determined.

[0009] According to a second aspect of the present disclosure, a large model-based question-answering device is provided, comprising:

[0010] A first determination module is used to determine a first result corresponding to the query statement based on the first large model, wherein the first result includes a first answer corresponding to the query statement, a first answer logic corresponding to the first answer, and first background knowledge;

[0011] A retrieval module, configured to search a database based on the query statement and determine at least one retrieval result corresponding to the query statement;

[0012] A second determination module, used to determine the type of knowledge conflict between the first result and the search result;

[0013] The third determination module is used to determine a target response corresponding to the query statement based on the first result, the search result and the type of the knowledge conflict.

[0014] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0015] at least one processor; and

[0016] a memory communicatively connected to the at least one processor; wherein,

[0017] The memory stores instructions that can be executed 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 large model-based question-answering method as described in the first aspect.

[0018] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the large model-based question-answering method as described in the first aspect.

[0019] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising computer instructions, which, when executed by a processor, implement the steps of the large model-based question-answering method as described in the first aspect.

[0020] The large model-based question-answering method, device, electronic device, and storage medium provided by the present disclosure have the following beneficial effects:

[0021] In the disclosed embodiment, the first result corresponding to the query statement is first determined based on the first large model, wherein the first result includes the first reply corresponding to the query statement, the first reply logic corresponding to the first reply, and the first background knowledge. Then, based on the query statement, a search is performed in the database to determine at least one retrieval result corresponding to the query statement, and then the type of knowledge conflict between the first result and the retrieval result is determined. Finally, based on the first result, the retrieval result, and the type of knowledge conflict, the target reply corresponding to the query statement is determined. Thus, the first result corresponding to the query statement is determined based on the first large model, and the retrieval result corresponding to the query statement is determined based on the database. Then, according to the type of knowledge conflict between the first result and the retrieval result, the first result and the retrieval result are processed to obtain the target reply, thereby avoiding the influence of the knowledge conflict between the first result and the retrieval result on the target reply, and thus improving the accuracy of the determined target reply.

[0022] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended 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

[0023] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0024] Figure 1 It is a flowchart of a question-answering method based on a large model provided according to an embodiment of the present disclosure;

[0025] Figure 2 is a flowchart of a question-answering method based on a large model provided according to another embodiment of the present disclosure;

[0026] Figure 3 is a flowchart of a question-answering method based on a large model provided according to another embodiment of the present disclosure;

[0027] Figure 4 is a flowchart of a question-answering method based on a large model provided according to another embodiment of the present disclosure;

[0028] Figure 5 is a flowchart of a question-answering method based on a large model provided according to another embodiment of the present disclosure;

[0029] Figure 6 is a structural schematic diagram of a large model-based question-answering device provided according to an embodiment of the present disclosure;

[0030] Figure 7 It is a block diagram of an electronic device used to implement the large model-based question-answering method of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0031] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0032] The present disclosure relates to artificial intelligence technology fields such as deep learning, intelligent question answering, and large models.

[0033] Artificial Intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence.

[0034] Deep learning (DL) is the process of learning the inherent laws and representation levels of sample data. The information obtained in the learning process is very helpful for interpreting data such as text, images, and sounds. The ultimate goal of deep learning is to enable machines to have analytical learning capabilities like humans and to recognize data such as text, images, and sounds.

[0035] Intelligent question answering, the intelligent question answering system aims to design and develop a system that can parse, understand and answer natural language questions raised by users. Its core is to use natural language processing technology to analyze and understand the questions input by users, extract the key information and intention of the questions, and finally generate and return accurate, concise and direct answers by matching or reasoning with the knowledge base.

[0036] The large model can also be called the Foundation Model. The model extracts knowledge through billions of corpora or images, learns and then produces a large model with billions of parameters.

[0037] Parameter knowledge conflict: Since parameter knowledge is internalized in LLM, it can provide higher accuracy for high-frequency questions, but for long-tail questions and uncommon scenarios, parameter knowledge often cannot effectively support them and may even produce wrong answers. Parameter knowledge conflict refers to the conflict phenomenon that occurs when the knowledge internalized in the large language model (LLM) generates text. LLM learns a lot of general knowledge through pre-training, which is encoded into the parameters of the model, so it is called "parameter knowledge."

[0038] External knowledge conflict: When using external knowledge, the model may retrieve content that conflicts with internalized knowledge, resulting in logical contradictions or knowledge errors in the generated results. Especially in black box models, such conflicts are difficult to troubleshoot and correct. External knowledge conflict refers to the conflict phenomenon that occurs when the information retrieved from the external knowledge base conflicts with the parameter knowledge inside the LLM in the RAG model. External knowledge usually comes from real-time retrieval and may include literature, databases, web pages, etc.

[0039] Parameter knowledge conflict and external knowledge conflict are the two main challenges faced by the RAG model in the knowledge fusion process. The former stems from the limitations and inconsistencies of the model's internalized knowledge, while the latter stems from the conflict between dynamic and extensive external knowledge and internalized knowledge. Solving these conflicts is the key to improving the performance of the RAG model on complex and long-tail problems.

[0040] In the technical solution disclosed herein, the acquisition, storage and application of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0041] The following describes the big model-based question-answering method, device, electronic device, and storage medium according to an embodiment of the present disclosure with reference to the accompanying drawings.

[0042] It should be noted that the executor of the big model-based question-answering method of this embodiment is a big model-based question-answering device, which can be implemented by software and / or hardware. The device can be configured in an electronic device, and the electronic device may include but is not limited to a terminal, a server, etc.

[0043] Figure 1 It is a flowchart of a question-answering method based on a large model provided according to an embodiment of the present disclosure.

[0044] like Figure 1 As shown, the large model-based question answering method includes:

[0045] S101, based on the first large model, determine the first result corresponding to the query statement, wherein the first result includes a first reply corresponding to the query statement, a first reply logic corresponding to the first reply, and first background knowledge.

[0046] The first result may be a result obtained by processing the query statement based on the first large model.

[0047] Among them, the first reply can be a reply corresponding to the query statement.

[0048] Among them, the first reply logic can generate a first reply for the first large model, using the reply logic used.

[0049] Among them, the first background knowledge can be the background knowledge based on which the first answer is generated by the first large model.

[0050] The first large model may be a model for answering user questions, and may be a pre-set model of any type and structure. For example, the first large model may be a model composed of a deep neural network and an LLM fine-tuning model, etc., and the present disclosure does not limit this.

[0051] S102: Search in a database based on the query statement to determine at least one search result corresponding to the query statement.

[0052] Among them, the database may include web pages, documents, pre-built knowledge bases, etc.

[0053] In some embodiments, a search may be performed in different databases based on a query statement to obtain multiple search results corresponding to the query statement.

[0054] In some embodiments, a search may also be performed in a database based on a query statement to obtain at least one search result corresponding to the query statement.

[0055] S103: Determine the type of knowledge conflict between the first result and the search result.

[0056] In some embodiments, the types of knowledge conflicts may include semantic errors, factual errors, grammatical errors, etc. The present disclosure is not limited to this.

[0057] For example, the first result is "the color of mung beans is red" and the search result is "the color of mung beans is green", then the type of knowledge conflict between the first result and the search result is factual error.

[0058] For example, the first result is "Xiao Lan's brother is Xiao Hong" and the search result is "Xiao Hong's brother is Xiao Lan", then the type of knowledge conflict between the first result and the search result is semantic error.

[0059] In some embodiments, based on the definition corresponding to each type of knowledge conflict, conflict identification may be performed on the first result and the search result, thereby determining the type of knowledge conflict between the first result and the search result.

[0060] In some embodiments, if there are multiple search results, conflicts are identified between the first result and each search result respectively to obtain the type of knowledge conflict between each search result and the first result, and then the type of knowledge conflict with the largest number is determined as the type of knowledge conflict between the first result and the search result.

[0061] For example, the number of search results is 5, among which the type of knowledge conflict is grammatical error, the number is 3, the number is semantic error, and the number is factual error, then it is determined that the type of knowledge conflict between the first result and the search result is grammatical error.

[0062] S104: Determine a target response corresponding to the query statement based on the first result, the search result, and the type of knowledge conflict.

[0063] In some embodiments, after determining the type of knowledge conflict, the first result and the search result may be processed based on the type of knowledge conflict to determine a target response corresponding to the query statement.

[0064] In the disclosed embodiment, the first result corresponding to the query statement is first determined based on the first large model, wherein the first result includes the first reply corresponding to the query statement, the first reply logic corresponding to the first reply, and the first background knowledge. Then, based on the query statement, a search is performed in the database to determine at least one retrieval result corresponding to the query statement, and then the type of knowledge conflict between the first result and the retrieval result is determined. Finally, based on the first result, the retrieval result, and the type of knowledge conflict, the target reply corresponding to the query statement is determined. Thus, the first result corresponding to the query statement is determined based on the first large model, and the retrieval result corresponding to the query statement is determined based on the database. Then, according to the type of knowledge conflict between the first result and the retrieval result, the first result and the retrieval result are processed to obtain the target reply, thereby avoiding the influence of the knowledge conflict between the first result and the retrieval result on the target reply, and thus improving the accuracy of the determined target reply.

[0065] Figure 2 is a flowchart of a question-answering method based on a large model provided according to another embodiment of the present disclosure;

[0066] like Figure 2 As shown, the large model-based question answering method includes:

[0067] S201, based on the first large model, determine the first result corresponding to the query statement, wherein the first result includes a first reply corresponding to the query statement, a first reply logic corresponding to the first reply, and first background knowledge.

[0068] S202, based on the query statement, searching in the database to determine at least one search result corresponding to the query statement.

[0069] The specific implementation forms of step S201 to step S202 can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0070] S203, obtaining definitions and reference examples corresponding to factual errors, and definitions and reference examples corresponding to semantic errors.

[0071] In some embodiments of the present disclosure, the type of knowledge conflict includes at least one of the following: factual error, semantic error. Therefore, in the embodiments of the present disclosure, the first results and search results corresponding to various types of knowledge conflict can be processed to obtain a target answer.

[0072] It should be noted that factual errors and semantic errors are problems that are difficult for large models to solve when generating target responses based on external knowledge and parameter knowledge.

[0073] For example, a reference example of factual error may be: There is a factual error between "the color of mung beans is red" and "the color of mung beans is green".

[0074] For example, a semantic error reference example may be: There is a semantic error between "Xiao Lan's brother is Xiao Hong" and "Xiao Hong's brother is Xiao Lan".

[0075] There may be one or more reference examples of factual errors, and the present disclosure does not limit this.

[0076] There may be one or more reference examples of semantic errors, which is not limited in the present disclosure.

[0077] S204, input the first result, the search result, the definition and reference examples corresponding to the factual error, and the definition and reference examples corresponding to the semantic error into the second largest model to determine the type of knowledge conflict.

[0078] The second large model and the first large model may be the same large model or different large models, which is not limited in the present disclosure.

[0079] Among them, the definition and reference examples corresponding to factual errors are used to prompt the second largest model what kind of knowledge conflict is a factual error, so that the second largest model can determine whether there is a factual error between the first result and the search result.

[0080] Among them, the definition and reference examples corresponding to semantic errors are used to prompt the second largest model what kind of knowledge conflict is a semantic error, so that the second largest model can determine whether there is a semantic error between the first result and the search result.

[0081] S205, based on the first result, the search result and the type of knowledge conflict, determine the target response corresponding to the query statement.

[0082] In some embodiments, when the type of knowledge conflict is a semantic error, the first result, the search result, the type of knowledge conflict, and the definition and reference examples corresponding to the semantic error are input into the third model to determine the target answer. Thus, the first result, the search result, and the type of knowledge conflict can be processed based on the third model, so that the target answer corresponding to the query statement can be determined quickly and accurately.

[0083] The third large model and the first large model and the second large model may be the same large model or different large models, which is not limited in the present disclosure.

[0084] Among them, the type of knowledge conflict is input into the third model to prompt the third model that there is a semantic error between the first result and the search result, and then the semantic error between the first result and the search result can be processed to obtain the target answer.

[0085] In the disclosed embodiment, based on the first large model, the first result corresponding to the query statement is determined, based on the query statement, a search is performed in the database to determine at least one search result corresponding to the query statement, the first result, the search result, the definition and reference examples corresponding to the factual error, and the definition and reference examples corresponding to the semantic error are input into the second large model to determine the type of knowledge conflict, and finally based on the first result, the search result, and the type of knowledge conflict, the target response corresponding to the query statement is determined. Thus, the second large model can be prompted to identify whether there is a factual conflict or a semantic conflict between the first result and the search result in combination with the definition and reference examples corresponding to the factual error and the definition and reference examples corresponding to the semantic error, so that the type of knowledge conflict between the first result and the search result can be quickly and accurately determined, and then based on the type of knowledge conflict, the target response can be accurately determined.

[0086] Figure 3 is a flowchart of a question-answering method based on a large model provided according to another embodiment of the present disclosure; Figure 3 As shown, the large model-based question answering method includes:

[0087] S301, based on the first large model, determine the first result corresponding to the query statement, wherein the first result includes the first reply corresponding to the query statement, the first reply logic corresponding to the first reply, and the first background knowledge.

[0088] S302: Search the database based on the query statement to determine at least one search result corresponding to the query statement.

[0089] S303, determining the type of knowledge conflict between the first result and the search result.

[0090] The specific implementation forms of steps S301 to S303 can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0091] S304, determining a first weight corresponding to the first result and a second weight corresponding to the search result.

[0092] In some embodiments, the first weight and the second weight may be preset, which is not limited in the present disclosure.

[0093] In some embodiments of the present disclosure, the first weight corresponding to the first result is determined according to the parameter amount corresponding to the first large model, wherein the parameter amount is proportional to the first weight, and the second weight is determined according to the first weight. Thus, the first weight corresponding to the first result can be determined according to the parameter amount of the first large model, and then the second weight corresponding to the search result can be determined, so that the accuracy of the first result generated by the first large model can be determined based on the parameter amount of the first large model, and then the first weight corresponding to the first result and the second weight corresponding to the search result can be determined more accurately.

[0094] The number of parameters of the first largest model may be the total number of parameters included in the first largest model. It should be noted that the more parameters the first largest model has, the more training data is required, and the more knowledge the first largest model learns. Therefore, the accuracy of the generated first result is higher.

[0095] The sum of the first weight and the second weight may be 1. This disclosure does not limit this.

[0096] In some embodiments, the more parameters the first large model has, the larger the corresponding first weight is.

[0097] S305, when the type of knowledge conflict is a semantic error, the first result, the search result, the type of knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third model to determine the target response.

[0098] Among them, the first weight corresponding to the first result and the second weight corresponding to the search result are input into the third largest model, which can prompt the third largest model of the accuracy corresponding to the first result and the accuracy corresponding to the detection result, so that the third largest model can process the first result and the search result based on the first weight and the second weight, so that the third largest model can output a more accurate target answer.

[0099] In the disclosed embodiment, firstly, based on the first large model, the first result corresponding to the query statement is determined, and based on the query statement, a search is performed in the database to determine at least one search result corresponding to the query statement, and then the type of knowledge conflict between the first result and the search result is determined, and finally the first weight corresponding to the first result and the second weight corresponding to the search result are determined. In the case where the type of knowledge conflict is a semantic error, the first result, the search result, the type of knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third large model to determine the target reply. Thus, in the case where the type of knowledge conflict between the first result and the search result is a semantic error, the third large model can be prompted to process the first result and the search result based on the first weight corresponding to the first result, the second weight corresponding to the search result, and the type of knowledge conflict to obtain the target reply, so that the target reply corresponding to the query statement can be determined more accurately.

[0100] Figure 4 is a flowchart of a question-answering method based on a large model provided according to another embodiment of the present disclosure; Figure 4 As shown, the large model-based question answering method includes:

[0101] S401, input the query statement into the first large model a preset number of times, and obtain a preset number of output results, wherein each output result includes a second reply corresponding to the query statement, a second reply logic corresponding to the second reply, and second background knowledge.

[0102] It should be noted that, since the first large model has learned a lot of parameter knowledge, different results may be output when processing the same query statement. In the disclosed embodiment, the query statement can be output to the first large model multiple times to obtain multiple output results.

[0103] The second reply logic may be a reply logic used by the first large model to determine the second reply.

[0104] Among them, the second background knowledge is the knowledge based on which the first model determines the second answer.

[0105] S402, process a preset number of second replies to determine the first reply.

[0106] In some embodiments, a preset number of second replies are cross-validated to determine the first reply.

[0107] In some embodiments, a preset number of second replies may be input into the fourth largest model so that the fourth largest model summarizes the preset number of second replies to obtain a first reply.

[0108] S403, processing a preset number of second reply logics to determine the first reply logic.

[0109] In some embodiments, a preset number of second reply logics are cross-validated to determine the first reply logic.

[0110] In some embodiments, a preset number of second reply logics may be input into the fourth largest model, so that the fourth largest model summarizes the preset number of second reply logics to obtain the first reply logic.

[0111] S404: Process a preset number of second background knowledge to determine first background knowledge.

[0112] In some embodiments, a preset number of second background knowledge are cross-validated to determine the first background knowledge.

[0113] In some embodiments, a preset number of second background knowledge may be input into the fourth largest model, so that the fourth largest model summarizes the preset number of second background knowledge to obtain the first background knowledge.

[0114] In some embodiments, a preset number of first replies, a preset number of second replies, and a preset number of second background knowledge are input into the fourth model to determine the first reply, the first reply logic, and the first background knowledge. Thus, based on the fourth model, the output results of the third model for the query statement can be quickly and accurately processed to obtain the first reply, the first reply logic, and the first background knowledge.

[0115] S405: Search the database based on the query statement to determine at least one search result corresponding to the query statement.

[0116] S406, determining the type of knowledge conflict between the first result and the search result.

[0117] The specific implementation forms of step S405 to step S406 can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0118] S407, based on the first result, the search result and the type of knowledge conflict, determine the target response corresponding to the query statement.

[0119] In some embodiments, when the output result includes the confidence level corresponding to the second reply, the first weight corresponding to the first result can also be determined based on a preset number of confidence levels, and then the second weight corresponding to the search result is obtained. Finally, when the type of knowledge conflict is a semantic error, the search result, the first result, the type of knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third model to determine the target reply. Thus, the first weight of the first result can be dynamically determined based on the confidence level of each output result of the first model, so that the determined first weight is more accurate, and thus the determined target reply is more accurate.

[0120] The confidence level may be the credibility of the second reply to the query statement output by the first large model.

[0121] In some embodiments, an average value, variance, etc. of a preset number of confidence levels may be determined as the first weight.

[0122] In some embodiments, the second weight may be fixed, or may change according to the first weight. For example, if the first weight increases, the second weight decreases.

[0123] In the disclosed embodiment, the query statement can be input into the first large model for a preset number of times to obtain a preset number of output results, wherein each output result includes the second reply corresponding to the query statement, the second reply logic corresponding to the second reply, and the second background knowledge, and then the preset number of second replies are processed to determine the first reply, the preset number of second reply logics are processed to determine the first reply logic, the preset number of second background knowledge are processed to determine the first background knowledge, and then based on the query statement, a search is performed in the database to determine at least one search result corresponding to the query statement, and the type of knowledge conflict between the first result and the search result is determined, and finally based on the first result, the search result and the type of knowledge conflict, the target reply corresponding to the query statement is determined. Thus, the second reply, the second reply logic and the second background knowledge corresponding to the query statement can be obtained multiple times through the first large model, and then the preset number of second replies are cross-validated to determine the first reply, the preset number of second reply logics are cross-validated to determine the first reply logic, and the preset number of second background knowledge are processed to determine the first background knowledge, so that the parameter knowledge conflict of the first large model can be resolved, so that the determined first reply, the first reply logic and the first background knowledge are more accurate, and the accuracy of the determined target reply is further improved.

[0124] In the present disclosure, Figure 5 As shown, the large model-based question answering method includes:

[0125] S501, based on the first large model, determine the first result corresponding to the query statement, wherein the first result includes a first reply corresponding to the query statement, a first reply logic corresponding to the first reply, and first background knowledge.

[0126] S502: Search in a database based on the query statement to determine at least one search result corresponding to the query statement.

[0127] S503: Determine the type of knowledge conflict between the first result and the search result.

[0128] The specific implementation forms of steps S501 to S503 can refer to the detailed descriptions in other embodiments of the present disclosure, and will not be described in detail here.

[0129] S504: When the type of knowledge conflict is factual error, a third result corresponding to the query statement is determined again based on the first large model.

[0130] The third result includes a third reply corresponding to the query statement, a third reply logic corresponding to the third reply, and third background knowledge;

[0131] In some embodiments, the query statement may be input into the first large model again to obtain a third result.

[0132] In some embodiments, the query statement is input into the first large model again a preset number of times to obtain a preset number of output results, wherein each output result includes a fourth reply corresponding to the query statement, a fourth reply logic corresponding to the fourth reply, and fourth background knowledge; the preset number of fourth replies are processed to determine a third reply; the preset number of fourth reply logics are processed to determine the third reply logic; the preset number of fourth background knowledge are processed to determine the third background knowledge.

[0133] S505: Search the database again based on the query statement to determine at least one new search result corresponding to the query statement.

[0134] The database used for the second search may be the same as or different from the database used for the first search, and the present disclosure does not limit this.

[0135] The new search result may be the same as or different from the search result obtained by the first search, and the present disclosure does not limit this.

[0136] S506, determining the type of knowledge conflict between the third result and the new search result until the type of knowledge conflict is not a factual error.

[0137] Among them, the specific implementation method of determining the type of knowledge conflict between the third result and the new search result can refer to the description of determining the type of knowledge conflict between the first result and the search result in other embodiments, and will not be repeated here.

[0138] It should be noted that the type of knowledge conflict is factual error, which means that there is a factual error in the first result or in the search result, and it is difficult to determine whether there is a factual error in the first result or in the search result. Therefore, in the embodiment of the present disclosure, the third result corresponding to the query statement can be obtained again based on the first large model, and the query statement can be searched again in the database to obtain a new search result, until the type of knowledge conflict between the third result obtained again and the new search result is not a factual error. In this way, the target answer corresponding to the query statement can be accurately determined.

[0139] In the disclosed embodiment, based on the first large model, the first result corresponding to the query statement is determined, wherein the first result includes the first reply corresponding to the query statement, the first reply logic corresponding to the first reply, and the first background knowledge, and then based on the query statement, a search is performed in the database to determine at least one search result corresponding to the query statement, and the type of knowledge conflict between the first result and the search result is determined, and finally, in the case where the type of knowledge conflict is a factual error, the third result corresponding to the query statement is determined again based on the first large model, and based on the query statement, a search is performed in the database again to determine at least one new search result corresponding to the query statement, and finally the type of knowledge conflict between the third result and the new search result is determined, until the type of knowledge conflict is not a factual error. Thus, in the case where the type of knowledge conflict between the first result and the search result is a factual error, the third result and the new search result corresponding to the query statement can be obtained again, until the type of knowledge conflict between the third result and the new search result corresponding to the query statement obtained again is not a factual error, thereby avoiding factual errors in the output target reply, and further improving the accuracy of the determined target reply.

[0140] In some embodiments, the big model-based question-answering method in the present disclosure can be applied to at least one of the following scenarios:

[0141] (1) Intelligent question-answering system: In intelligent question-answering systems, users’ questions usually involve information sources from multiple fields. Traditional models are prone to conflicts when processing knowledge from these different sources. The large model-based question-answering method disclosed in this disclosure can effectively handle the conflicts between parameter knowledge and external knowledge, ensuring that the system provides more accurate and consistent answers when answering user questions. This will greatly improve the reliability of the question-answering system and user satisfaction.

[0142] (2) Medical diagnosis support system: Medical diagnosis support system usually needs to combine the patient's historical medical records (internal knowledge) and the latest medical research (external knowledge) to provide diagnostic suggestions. Knowledge from different sources may be inconsistent or conflicting, affecting the accuracy of diagnosis. Through the large model-based question-answering method disclosed in this disclosure, it is possible to accurately integrate internal medical records and external research data in the presence of knowledge conflicts, provide more accurate diagnostic suggestions, and improve the reliability of medical decision support.

[0143] (3) Legal consulting system: The legal consulting system needs to process regulations, cases and expert opinions from different legal fields. There may be conflicts between these information sources, resulting in inconsistent legal opinions. Through the big model-based question-answering method disclosed in this disclosure, the system can better integrate legal knowledge from different sources and provide more unified and logically consistent legal advice, which helps reduce legal decision-making errors caused by knowledge conflicts.

[0144] (4) Personalized recommendation system. In the field of e-commerce or content recommendation, the recommendation system needs to combine the user's historical behavior (internal knowledge) and current popular trends (external knowledge) to make recommendations. The conflict between the two may lead to poor recommendation results. The large model-based question-answering method disclosed in this disclosure can effectively solve the conflict between the user's historical behavior and the current popular trends, provide more accurate and personalized recommendation results, and thus improve the user experience and the conversion rate of the platform.

[0145] (5) Enterprise knowledge management system: The knowledge management system within an enterprise needs to integrate the internal experience of the enterprise (parameter knowledge) and the latest developments outside the industry (external knowledge). The conflict between the two may lead to wrong decisions or chaotic knowledge management. Through the big model-based question-answering method disclosed in this disclosure, enterprises can better manage and integrate internal and external knowledge resources, ensure the consistency and reliability of knowledge, and support the strategic decision-making of enterprises.

[0146] Figure 6 is a structural diagram of a question-answering device based on a large model provided according to an embodiment of the present disclosure; Figure 6 As shown, the large model-based question-answering device 600 includes:

[0147] A first determination module 601 is used to determine a first result corresponding to the query statement based on the first large model, wherein the first result includes a first reply corresponding to the query statement, a first reply logic corresponding to the first reply, and first background knowledge;

[0148] A search module 602 is used to search the database based on the query statement and determine at least one search result corresponding to the query statement;

[0149] A second determination module 603, used to determine the type of knowledge conflict between the first result and the search result;

[0150] The third determination module 604 is used to determine a target response corresponding to the query statement based on the first result, the search result and the type of knowledge conflict.

[0151] In some embodiments, the type of knowledge conflict includes at least one of the following:

[0152] Factual errors and semantic errors.

[0153] In some embodiments, the second determining module 603 is used to:

[0154] Obtain definitions and reference examples corresponding to factual errors and definitions and reference examples corresponding to semantic errors;

[0155] The first result, the search result, the definition and reference examples corresponding to the factual error, and the definition and reference examples corresponding to the semantic error are input into the second largest model to determine the type of knowledge conflict.

[0156] In some embodiments, the third determining module 604 is configured to:

[0157] Determine a first weight corresponding to the first result and a second weight corresponding to the search result;

[0158] When the type of knowledge conflict is semantic error, the first result, search result, the type of knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third model to determine the target response.

[0159] In some embodiments, the third determining module 604 is configured to:

[0160] Determine a first weight corresponding to the first result according to the parameter amount corresponding to the first large model, wherein the parameter amount is proportional to the first weight;

[0161] According to the first weight, a second weight is determined.

[0162] In some embodiments, the third determining module 604 is configured to:

[0163] In the case where the type of knowledge conflict is a semantic error, the first result, the search result, the type of knowledge conflict, and the definition and reference examples corresponding to the semantic error are input into the third model to determine the target response.

[0164] In some embodiments, the first determining module 601 is used to:

[0165] Input the query statement into the first large model a preset number of times, and obtain a preset number of output results, wherein each output result includes a second reply corresponding to the query statement, a second reply logic corresponding to the second reply, and second background knowledge;

[0166] Processing a preset number of second replies to determine the first reply;

[0167] Processing a preset number of second reply logics to determine the first reply logic;

[0168] A preset number of second background knowledge are processed to determine the first background knowledge.

[0169] In some embodiments, the first determining module 601 is used to:

[0170] A preset number of first answers, a preset number of second answers, and a preset number of second background knowledge are input into the fourth model to determine the first answer, the first answer logic, and the first background knowledge.

[0171] In some embodiments, the third determining module 604 is configured to:

[0172] Determining a first weight corresponding to the first result according to a preset number of confidence levels;

[0173] Obtaining a second weight corresponding to the search result;

[0174] When the type of knowledge conflict is semantic error, the search result, the first result, the type of knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third model to determine the target response.

[0175] In some embodiments, the apparatus further comprises a processing module for:

[0176] When the type of knowledge conflict is factual error, a third result corresponding to the query statement is determined again based on the first model;

[0177] Based on the query statement, searching the database again to determine at least one new search result corresponding to the query statement;

[0178] The type of knowledge conflict between the third result and the new search result is determined until the type of knowledge conflict is not a factual error.

[0179] It should be noted that the aforementioned explanation of the large model-based question-answering method is also applicable to the large model-based question-answering device of this embodiment, and will not be repeated here.

[0180] In the disclosed embodiment, the first result corresponding to the query statement is first determined based on the first large model, wherein the first result includes the first reply corresponding to the query statement, the first reply logic corresponding to the first reply, and the first background knowledge. Then, based on the query statement, a search is performed in the database to determine at least one retrieval result corresponding to the query statement, and then the type of knowledge conflict between the first result and the retrieval result is determined. Finally, based on the first result, the retrieval result, and the type of knowledge conflict, the target reply corresponding to the query statement is determined. Thus, the first result corresponding to the query statement is determined based on the first large model, and the retrieval result corresponding to the query statement is determined based on the database. Then, according to the type of knowledge conflict between the first result and the retrieval result, the first result and the retrieval result are processed to obtain the target reply, thereby avoiding the influence of the knowledge conflict between the first result and the retrieval result on the target reply, and thus improving the accuracy of the determined target reply.

[0181] 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.

[0182] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0183] like Figure 7 As shown, the device 700 includes a computing unit 701, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 702 or a computer program loaded from a storage unit 708 into a random access memory (RAM) 703. In the RAM 703, various programs and data required for the operation of the device 700 can also be stored. The computing unit 701, the ROM 702, and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0184] A number of components in the device 700 are connected to the I / O interface 705, including: an input unit 706, such as a keyboard, a mouse, etc.; an output unit 707, such as various types of displays, speakers, etc.; a storage unit 708, such as a disk, an optical disk, etc.; and a communication unit 709, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 allows the device 700 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0185] The computing unit 701 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 701 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, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 701 performs the various methods and processes described above, such as large model-based question and answer. For example, in some embodiments, the large model-based question and answer may be implemented as a computer software program, which is tangibly contained in a machine-readable medium, such as a storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the large model-based question and answer described above may be performed. Alternatively, in other embodiments, the computing unit 701 may be configured to perform large model-based question and answer in any other appropriate manner (e.g., by means of firmware).

[0186] Various implementations of the systems and techniques described above herein 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 chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations 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 that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0187] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0188] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, 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 disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0189] 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 pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types 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, voice input, or tactile input).

[0190] The systems and techniques described herein can be implemented in a computing system that includes backend 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 frontend components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend 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), the Internet, and a blockchain network.

[0191] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0192] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0193] In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present disclosure, the meaning of "multiple" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined. In the description of the present disclosure, the words "if" and "if" used can be interpreted as "at the time of" or "when" or "in response to determination" or "under the circumstances of".

[0194] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A question-answering method based on a large model, wherein: include: Based on the first large model, determine a first result corresponding to the query statement, wherein the first result includes a first answer corresponding to the query statement, a first answer logic corresponding to the first answer, and first background knowledge; Based on the query statement, searching in a database to determine at least one search result corresponding to the query statement; determining a type of knowledge conflict between the first result and the search result; Based on the first result, the search result and the type of the knowledge conflict, a target response corresponding to the query statement is determined.

2. The method according to claim 1, wherein: The types of knowledge conflicts include at least one of the following: Factual errors and semantic errors.

3. The method according to claim 2, wherein: The determining the type of knowledge conflict between the first result and the search result includes: Obtaining definitions and reference examples corresponding to the factual errors and definitions and reference examples corresponding to the semantic errors; The first result, the search result, the definition and reference examples corresponding to the factual error, and the definition and reference examples corresponding to the semantic error are input into the second largest model to determine the type of the knowledge conflict.

4. The method according to claim 1, wherein: The determining, based on the first result, the search result, and the type of the knowledge conflict, a target response corresponding to the query statement includes: Determine a first weight corresponding to the first result and a second weight corresponding to the search result; In the case where the type of knowledge conflict is a semantic error, the first result, the search result, the type of knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third model to determine the target response.

5. The method according to claim 4, wherein: The determining a first weight corresponding to the first result and a second weight corresponding to the search result includes: Determining the first weight corresponding to the first result according to the parameter amount corresponding to the first large model, wherein the parameter amount is proportional to the first weight; The second weight is determined according to the first weight.

6. The method according to claim 1, wherein: The determining, based on the first result, the search result, and the type of the knowledge conflict, a target response corresponding to the query statement includes: In the case where the type of the knowledge conflict is a semantic error, the first result, the search result, the type of the knowledge conflict, and the definition and reference examples corresponding to the semantic error are input into the third model to determine the target response.

7. The method according to claim 1, wherein: The determining, based on the first large model, a first result corresponding to the query statement includes: Input the query statement into the first large model a preset number of times to obtain a preset number of output results, wherein each of the output results includes a second reply corresponding to the query statement, a second reply logic corresponding to the second reply, and second background knowledge; Processing the preset number of second replies to determine the first reply; Processing the preset number of second reply logics to determine the first reply logic; The preset number of second background knowledge are processed to determine the first background knowledge.

8. The method according to claim 7, wherein: After inputting the query statement into the first large model for a preset number of times and obtaining a preset number of output results, the method further includes: The preset number of first replies, the preset number of second replies, and the preset number of second background knowledge are input into the fourth model to determine the first reply, the first reply logic, and the first background knowledge.

9. The method according to claim 7, wherein: The output result includes the confidence level corresponding to the second reply, and the determining the target reply corresponding to the query statement based on the first result, the search result and the type of the knowledge conflict includes: Determining a first weight corresponding to the first result according to the preset number of confidences; Obtaining a second weight corresponding to the search result; In the case where the type of the knowledge conflict is the semantic error, the search result, the first result, the type of the knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third model to determine the target response.

10. The method according to claim 1, wherein: The method further comprises: In the case where the type of the knowledge conflict is a factual error, determining a third result corresponding to the query statement based on the first large model again; Based on the query statement again, searching the database to determine at least one new search result corresponding to the query statement; The type of knowledge conflict between the third result and the new search result is determined until the type of knowledge conflict is not a factual error.

11. A question-answering device based on a large model, wherein: include: A first determination module is used to determine a first result corresponding to the query statement based on the first large model, wherein the first result includes a first answer corresponding to the query statement, a first answer logic corresponding to the first answer, and first background knowledge; A retrieval module, configured to search a database based on the query statement and determine at least one retrieval result corresponding to the query statement; A second determination module, used to determine the type of knowledge conflict between the first result and the search result; The third determination module is used to determine a target response corresponding to the query statement based on the first result, the search result and the type of the knowledge conflict.

12. The device according to claim 11, wherein The types of knowledge conflicts include at least one of the following: Factual errors and semantic errors.

13. The device according to claim 12, wherein: The second determining module is used to: Obtaining definitions and reference examples corresponding to the factual errors and definitions and reference examples corresponding to the semantic errors; The first result, the search result, the definition and reference examples corresponding to the factual error, and the definition and reference examples corresponding to the semantic error are input into the second largest model to determine the type of the knowledge conflict.

14. The device according to claim 11, wherein: The third determining module is used to: Determine a first weight corresponding to the first result and a second weight corresponding to the search result; In the case where the type of knowledge conflict is a semantic error, the first result, the search result, the type of knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third model to determine the target response.

15. The device according to claim 14, wherein: The third determining module is used to: Determining the first weight corresponding to the first result according to the parameter amount corresponding to the first large model, wherein the parameter amount is proportional to the first weight; The second weight is determined according to the first weight.

16. The device according to claim 11, wherein The third determining module is used to: In the case where the type of the knowledge conflict is a semantic error, the first result, the search result, the type of the knowledge conflict, and the definition and reference examples corresponding to the semantic error are input into the third model to determine the target response.

17. The device according to claim 11, wherein: The first determining module is used to: Input the query statement into the first large model a preset number of times to obtain a preset number of output results, wherein each of the output results includes a second reply corresponding to the query statement, a second reply logic corresponding to the second reply, and second background knowledge; Processing the preset number of second replies to determine the first reply; Processing the preset number of second reply logics to determine the first reply logic; The preset number of second background knowledge are processed to determine the first background knowledge.

18. The device according to claim 17, wherein: The first determining module is used to: The preset number of first replies, the preset number of second replies, and the preset number of second background knowledge are input into the fourth model to determine the first reply, the first reply logic, and the first background knowledge.

19. The device according to claim 17, wherein: The third determining module is used to: Determining a first weight corresponding to the first result according to the preset number of confidences; Obtaining a second weight corresponding to the search result; In the case where the type of the knowledge conflict is the semantic error, the search result, the first result, the type of the knowledge conflict, the definition and reference example corresponding to the semantic error, the first weight corresponding to the first result, and the second weight corresponding to the search result are input into the third model to determine the target response.

20. The device according to claim 11, wherein The device also includes a processing module, configured to: In the case where the type of the knowledge conflict is a factual error, determining a third result corresponding to the query statement based on the first large model again; Based on the query statement again, searching the database to determine at least one new search result corresponding to the query statement; The type of knowledge conflict between the third result and the new search result is determined until the type of knowledge conflict is not a factual error.

21. An electronic device, comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 10.

22. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1 to 10.

23. A computer program product, comprising computer instructions, which, when executed by a processor, implement the steps of the method according to any one of claims 1 to 10.

Images