Question and answer information processing method and device, electronic equipment and storage medium

Abstract

The invention provides a question and answer information processing method and device, electronic equipment and a storage medium, and can be applied to the technical field of information processing. According to the technical scheme of the embodiment, after the vector database comprising the vector information corresponding to the knowledge information of the preset field is constructed, when the object query information is obtained, retrieval query is performed in the vector database according to the object query information to obtain the corresponding target output information; therefore, the inquired information is closer to the current demand information of the object, the matching degree of the information retrieval result and the current demand information is effectively improved, and the experience effect of the retrieval process is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of information processing, and in particular to a method, device, electronic device, and storage medium for processing question-and-answer information. Background Art

[0002] In the related art, a question-and-answer system can effectively improve the efficiency and accuracy of information retrieval to quickly obtain the information required by an object. However, since the current question-and-answer system is mainly a conversational question-and-answer system for the open domain, which is applicable to the open question-and-answer retrieval process. When an object needs to retrieve information in a specific field, the information retrieved by the current retrieval system may be different from the current demand information of the object, thereby affecting the experience effect of the retrieval process. Summary of the Invention

[0003] The present invention aims to at least solve one of the technical problems existing in the prior art. For this purpose, the present invention provides a method, device, electronic device, and storage medium for processing question-and-answer information, which can effectively improve the matching degree between the information retrieval result and the current demand information.

[0004] On the one hand, an embodiment of the present invention provides a method for processing question-and-answer information, the method comprising the following steps:

[0005] Construct a vector database, the vector database including vector information corresponding to knowledge information in a preset field;

[0006] Obtain object query information;

[0007] Perform a retrieval query in the vector database according to the object query information to obtain target output information.

[0008] According to some embodiments of the present invention, the constructing of the vector database includes:

[0009] Obtain first demand document information in the preset field, the first demand document information including knowledge information in the preset field;

[0010] Preprocess the first demand document information to obtain second demand document information;

[0011] Perform a chunking process on the second demand document information according to a preset chunking strategy to obtain target chunk content information;

[0012] Construct the vector database according to the target chunk content information.

[0013] According to some embodiments of the present invention, the preprocessing of the first demand document information to obtain second demand document information includes:

[0014] Extract the information that meets the first preset requirement in the first requirement document information as the first document information to be processed;

[0015] Extract the information that does not meet the first preset requirement in the first requirement document information as the second document information to be processed;

[0016] Perform a structure conversion on the first document information to be processed to obtain the third document information to be processed;

[0017] Merge the second document information to be processed and the third document information to be processed to obtain the second requirement document information.

[0018] According to some embodiments of the present invention, the block processing of the second requirement document information according to the preset block strategy to obtain the target block content information includes:

[0019] Classify the second requirement document information to obtain a classification result;

[0020] Perform block processing on the second requirement document information according to the classification result and the preset block strategy to obtain the target block content information.

[0021] According to some embodiments of the present invention, the block processing of the second requirement document information according to the classification result and the preset block strategy to obtain the target block content information includes:

[0022] Perform block processing on the document content in the second requirement document information according to the classification result to obtain the first block content information;

[0023] Perform block processing on the first block content information according to the preset block strategy to obtain the target block content information.

[0024] According to some embodiments of the present invention, the block processing of the first block content information according to the preset block strategy to obtain the target block content information includes:

[0025] Select at least one strategy from several preset block strategies as the target block strategy, and the preset block strategies include a fixed-size block strategy, a structure block strategy, a semantic block strategy, or a recursive block strategy;

[0026] Perform block processing on the first block content information according to the target block strategy to obtain the target block content information.

[0027] According to some embodiments of the present invention, the retrieval query in the vector database according to the object query information to obtain the target output information includes:

[0028] Convert the object query information into an object query vector;

[0029] Compare the object query vector with the vectors in the vector database to obtain the target output information.

[0030] According to some embodiments of the present invention, the comparing the object query vector with the vectors in the vector database to obtain the target output information includes:

[0031] Calculate the similarity between the object query vector and the vectors in the vector database;

[0032] Obtain at least one vector to be processed from the vector database according to the similarity;

[0033] Generate the target output information according to the vector to be processed.

[0034] According to some embodiments of the present invention, the generating the target output information according to the vector to be processed includes:

[0035] Obtain a preset prompt template;

[0036] Integrate the object query vector and the vector to be processed through the preset prompt template to obtain prompt content information;

[0037] Generate the target output information according to the prompt content information.

[0038] According to some embodiments of the present invention, the generating the target output information according to the prompt content information includes:

[0039] Obtain object usage feedback information;

[0040] Adjust the parameters of the first preset large language model according to the object usage feedback information to obtain a second preset large language model;

[0041] Input the prompt content information into the second preset large language model to generate the target output information.

[0042] On the other hand, an embodiment of the present invention provides a question and answer information processing device, and the device includes:

[0043] A construction module for constructing a vector database, where the vector database includes vector information corresponding to knowledge information in a preset field;

[0044] An acquisition module for acquiring object query information;

[0045] A retrieval query module for performing retrieval query in the vector database according to the object query information to obtain target output information.

[0046] On the other hand, an embodiment of the present invention provides an electronic device, including at least one control processor and a memory communicatively connected to the at least one control processor; the memory stores instructions executable by the at least one control processor, and when the instructions are executed by the at least one control processor, the at least one control processor is enabled to execute the above-mentioned question-and-answer information processing method.

[0047] On the other hand, an embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions for causing a computer to execute the above-mentioned question-and-answer information processing method.

[0048] The embodiments of the present application at least include the following beneficial effects: The present application provides a question-and-answer information processing method, device, electronic device, and storage medium. After constructing a vector database including vector information corresponding to knowledge information in a preset field, when obtaining object query information, the corresponding target output information is retrieved and queried in the vector database according to the object query information, so that the retrieved information is closer to the current demand information of the object, effectively improving the matching degree between the information retrieval result and the current demand information, and further improving the experience effect of the retrieval process. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] Figure 1 is an interactive scenario diagram of the question-and-answer information processing method provided by an embodiment of the present invention;

[0050] Figure 2 is a flowchart of the question-and-answer information processing method provided by an embodiment of the present invention;

[0051] Figure 3 is a flowchart of constructing a vector database provided by an embodiment of the present invention;

[0052] Figure 4 is a flowchart of preprocessing first demand document information to obtain second demand document information provided by an embodiment of the present invention;

[0053] Figure 5 is a flowchart of performing chunking processing on the second demand document information according to a preset chunking strategy to obtain target chunk content information provided by an embodiment of the present invention;

[0054] Figure 6 is a flowchart of performing chunking processing on the second demand document information according to a classification result and a preset chunking strategy to obtain target chunk content information provided by the present invention;

[0055] Figure 7It is a flowchart provided by the present invention for performing chunking processing on the first chunk content information according to a preset chunking strategy to obtain target chunk content information;

[0056] Figure 8 It is a flowchart provided by the present invention for performing retrieval query in a vector database according to object query information to obtain target output information;

[0057] Figure 9 It is a flowchart provided by the present invention for comparing an object query vector with vectors in a vector database to obtain target output information;

[0058] Figure 10 It is a flowchart provided by the present invention for generating target output information according to a vector to be processed;

[0059] Figure 11 It is a flowchart provided by the present invention for generating target output information according to prompt content information;

[0060] Figure 12 It is a schematic diagram of an application architecture of a question - answering information processing method provided by the present invention;

[0061] Figure 13 It is a complete implementation flowchart of a question - answering information processing method provided by the present invention;

[0062] Figure 14 It is a schematic structural diagram of a question - answering information processing device provided by an embodiment of the present invention;

[0063] Figure 15 It is a structural diagram of an electronic device provided by another embodiment of the present invention. Detailed Description of the Embodiment

[0064] The embodiments of the present invention will be described in detail below. The examples of the embodiments are shown in the accompanying drawings, where the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention, and should not be construed as a limitation of the present invention.

[0065] In the description of the present invention, it should be understood that for orientation descriptions, such as up, down, front, back, left, right, etc., the orientation or positional relationship indicated is based on the orientation or positional relationship shown in the accompanying drawings. It is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and thus should not be construed as a limitation of the present invention.

[0066] In the description of the present invention, "several" means one or more, "multiple" means more than two, and understandings such as "greater than", "less than", "exceeding", etc. do not include the corresponding number, while understandings such as "above", "below", "within", etc. include the corresponding number. If "first" and "second" are described, they are only used to distinguish technical features and cannot be understood as indicating or implying relative importance or implicitly indicating the quantity of the indicated technical features or implicitly indicating the sequence relationship of the indicated technical features.

[0067] In the description of the present invention, unless otherwise clearly defined, words such as "set", "install", "connect", etc. should be understood in a broad sense, and those skilled in the art can reasonably determine the specific meanings of the above words in the present invention in combination with the specific content of the technical solution.

[0068] Before elaborating on the embodiments of the present application in detail, some nouns and terms involved in the embodiments of the present application are first explained, and the nouns and terms involved in the embodiments of the present application are applicable to the following explanations:

[0069] QA (Question Answering System) is an advanced form of information retrieval system that can understand the questions raised by users through natural language processing technology and return accurate and concise answers. The question answering system combines multiple technologies such as artificial intelligence, natural language processing, knowledge graph, and information extraction, and can handle various types of questions, including factual queries, advisory queries, and interpretive queries, etc.

[0070] RAG (Retrieval-Augmented Generation) is a model that combines retrieval and generation technologies, and generates answers or content by referring to information from an external knowledge base. The RAG model has strong interpretability and customization capabilities and is applicable to multiple natural language processing tasks such as question answering systems, document generation, and intelligent assistants. The working principle of RAG is to introduce the retrieval results of the external knowledge base during the generation process to enhance the generation ability of the model and improve the accuracy and relevance of the generated text.

[0071] Markdown is a lightweight markup language that allows users to write documents in a plain text format that is easy to read and write, and then convert them into structured HTML (or other formats). The original design intention of Markdown is to simplify the process of document writing and formatting, so that users can focus on the text content without worrying about complex typesetting and formatting issues.

[0072] HTML (Hyper Text Markup Language, Hypertext Markup Language) is an essential tool for web page production. The structure of Hypertext Markup Language (or Hypertext Tag Language) includes a "head" part and a "body" part. The "head" part provides information about the web page, and the "body" part provides the specific content of the web page.

[0073] NLTK (Natural Language Toolkit) is a Python library for natural language processing (NLP). NLTK provides many tools and interfaces for processing and analyzing human language data. The Python library contains a series of functions and classes for text processing, analysis, linguistic research, and machine learning.

[0074] Embedding-based methods are a representation technique used to map high-dimensional data into a low-dimensional space for easier processing and analysis of this data. The representation techniques of Embedding-based methods include distributed representation, manifold representation, low-dimensional vector representation, application domain representation, triple prediction, and relation path reasoning representation, etc.

[0075] LLM (Large Language Model) is a natural language processing tool based on deep learning technology. Through the training of a large corpus, LLM has the ability to understand and generate text and performs well in tasks such as translation, writing, and dialogue. LLM can creatively generate text content that has never appeared before, has the ability to efficiently compress the information of the training text, and has a certain generalization ability. Different from databases or search engines, the main feature of LLM is its ability to generate new text content, rather than just retrieving or storing information.

[0076] It should be noted that QA can answer questions raised by the object in natural language accurately and concisely. Current QAs are all for open-domain conversational question-and-answer systems and are suitable for open-ended question-and-answer retrieval processes. However, with the development of intelligent technologies, the object's demand for information retrieval in specific fields is also increasing, making the information retrieved by the existing QA systems in specific field applications may have a large difference from the information required by the current object, thus affecting the object's usage experience of QA.

[0077] Therefore, the embodiments of the present application provide a method for processing question-and-answer information. After constructing a vector database composed of vector information corresponding to knowledge information in a preset field, when an object performs information retrieval in the preset field, it can perform a retrieval query in the vector database based on the object query information, so that the retrieved target output information is closer to the current demand information of the object, thereby effectively improving the matching degree between the information retrieval result and the current demand information, and further improving the experience effect of the retrieval process.

[0078] In the present invention, a method, device, electronic device, and storage medium for processing question-and-answer information are provided, and will be described in detail one by one in the following embodiments.

[0079] Refer to Figure 1 , Figure 1 FIG. Figure 1 As shown, after determining the preset field, the server 120 can construct vector information corresponding to the knowledge information in the preset field, and then save these vector information to the specified vector database 130, so that all the information in the vector database 130 is information in the preset field; after completing the construction of the vector database 130, the object can input the query information corresponding to the current demand information as the object query information at the request end 110 and send the object query information to the server 120; after receiving the object query information, the server 120 performs an information retrieval query operation in the specified vector database based on the query information to obtain the target output information in the preset field in the vector database, so that the target output information is close to the current demand information of the object, thereby effectively improving the matching degree between the information retrieval result and the current demand information; then the target output information is sent to the request end for display, so that the object can browse at the request end and thus understand the knowledge information corresponding to the current demand information.

[0080] It should be noted that the execution subject of the question-and-answer information processing method in this embodiment can be Figure 1 the system architecture shown, including a request end, a server, and a database. The request end can be a human-computer interaction device such as a personal computer or a terminal, and the server can be a common server. It can also be implemented through an integrated data analysis device. For example, an artificial intelligence (AI) analysis module and a memory are directly integrated in the data analysis device, and the intelligent analysis and construction process of vector information corresponding to the knowledge information in the preset field is realized through the AI analysis module, and the available historical results are stored in the memory. Without special instructions, the form of the execution subject of the question-and-answer information processing method in this embodiment is not limited.

[0081] Refer to Figure 2 ,Figure 2 The figure is a flowchart of a question-and-answer information processing method provided by an embodiment of the present invention. The question-and-answer information processing method includes, but is not limited to, steps S210 to S230:

[0082] Step S210: Construct a vector database, where the vector database includes vector information corresponding to knowledge information in a preset domain.

[0083] Specifically, the preset domain may be a scenario domain corresponding to specific information retrieval questions and answers within a preset time period. Exemplarily, when market analysis of a certain game needs to be conducted within the next 10 years, developers can construct a vector database based on all the information involved in the game. Based on this example, the preset domain may be the game scenario domain; the knowledge information includes all the information involved in the game, specifically including but not limited to game development knowledge information, player experience information corresponding to adjacent games, or operation information of adjacent games, etc.; the vector information in the vector database may be all the vectors obtained after vector transformation of the knowledge information in the preset domain. It can be understood that a certain element in the vector information represents a certain information point in the corresponding knowledge information.

[0084] It can be understood that the construction process of the vector database includes, but is not limited to, the construction of vector information in the database, the setting of database parameters and configuration files, and the construction of index information in the database. Among them, the vector information in the database is constructed based on the knowledge information in the preset domain and is used to represent the knowledge information in the preset domain. The setting process of database parameters and configuration files can be set based on the demand information of the current query retrieval and is used to guide information such as the storage directory form, index type, and vector dimension of the vector knowledge in the vector database. The construction process of the index information in the database is constructed based on the current query retrieval function information in combination with the corresponding interface or command-line tool and is used to provide the index method and index type of the vector information in the vector database.

[0085] In the embodiments of the present application, the constructed vector database can be used as an external knowledge source for the question system to provide all necessary information within a preset field. Exemplarily, when the preset field is the medical scenario field, the information in the external knowledge source includes, but is not limited to, several disease profile information, corresponding treatment method information, and drug information involved in the treatment method, etc. It can be understood that the external knowledge source of this embodiment can be directly obtained from public websites through preset tools. Since the information disclosed in multiple public websites is relatively scattered and the information introduction for an application scenario field in each website is lacking, this embodiment can obtain all the information about the same application scenario field from multiple public websites, and then clean, integrate, classify in detail, and perform vector conversion on this information to obtain the vector database for this application scenario field. Furthermore, when the retrieval system performs information query and retrieval, it can provide the external knowledge information for this application scenario field through this vector database, so as to improve the accuracy of information query and retrieval in the preset field.

[0086] Step S220: Obtain object query information;

[0087] Specifically, the object query information in this embodiment can be an information instruction triggered based on the current demand information, which is used to guide the subsequent processing process of the information processing end. Among them, the current demand information can be input through a human-computer interaction interface. It can be understood that when an object needs to perform information query and retrieval in a preset field, relevant description content of the current demand information is input into a preset window in the human-computer interaction interface. After the processing end corresponding to the human-computer interaction interface receives the relevant description content of the current demand information, it generates object query information that conforms to a preset format, so that the subsequent steps can perform information query and retrieval based on the object query information, effectively improving the query and retrieval efficiency.

[0088] In the embodiments of the present application, the process of obtaining object query information can be through a web interface or through an application programming interface provided by a third party. Exemplarily, taking the retrieval system as an example, when the retrieval system is deployed in the local area network of the processing terminal to which the object query information belongs, since the object query information is triggered within this local area network and belongs to local triggering, the object query information can be obtained through a preset web interface within this local area network. When the retrieval system is deployed on the server side, since the server side needs to interact with processing terminals in multiple local area networks, in order to ensure the security of data transmission, the server side of this embodiment can obtain the object query information from the processing terminal through a preset application programming interface. It can be understood that the preset application programming interface can be an interface that has passed security authentication in the server side in advance, thereby improving the security during the information transmission process.

[0089] Step S230: Perform a retrieval query in the vector database based on the object query information to obtain the target output information.

[0090] Specifically, the target output information can be at least one piece of information corresponding to the object query information in the vector database. It can be understood that in this embodiment, after receiving the object query information, a retrieval query is performed in the corresponding vector database to obtain at least one piece of knowledge information in the corresponding scenario field, and then all the knowledge information obtained through the query retrieval is integrated to obtain the target output information.

[0091] In practical applications, the object query information may include a query instruction for one scenario field or multiple query instructions for multiple scenario fields. Exemplarily, when the object query information only includes a query instruction for one scenario field, after determining the target scenario field based on this query instruction in this embodiment, a database that conforms to the target scenario field is determined from several vector databases as the target database, and then after analyzing the target problem corresponding to this query instruction, a query retrieval is performed in the target database to obtain the answer analysis information corresponding to the current problem as the target output information. When the object query information includes two or more query instructions for scenario fields, after determining the corresponding several target scenario fields based on this query instruction in this embodiment, databases that conform to each target scenario field are determined from several vector databases as the target databases, and then after analyzing the target problems of this query instruction in the corresponding target scenario fields, a query retrieval is performed in each target database based on each target problem to obtain the answer analysis information corresponding to each target problem, and then all the answer analysis information is integrated and processed to obtain the target output information.

[0092] It can be understood that the query retrieval processes for multiple target problems in this embodiment can be performed synchronously or asynchronously. When the query retrieval processes for multiple target problems are performed synchronously, the query retrieval processes are synchronously performed for knowledge query retrieval in the target database to quickly query and retrieve the answer analysis information corresponding to each target problem, thereby effectively improving the query retrieval efficiency. When the query retrieval processes for multiple target problems are performed asynchronously, in this embodiment, a query retrieval can be performed based on the first target problem, then a query retrieval for the second target problem can be performed based on the answer analysis information corresponding to the first target problem, and then a query retrieval for the third target problem can be performed based on the answer analysis information corresponding to the second target problem, and so on, until the query retrieval process for the last target problem is completed, and then all the answer analysis information obtained by querying and retrieving the corresponding target problems is integrated to obtain the target output information, thereby effectively improving the accuracy of the target output information.

[0093] In summary, in the embodiments of the present application, after a vector database for a preset field is pre-constructed, when object query information is obtained, knowledge information can be queried and retrieved in the vector database in the corresponding field based on the object query information, and then target output information of the current demand information of the object can be obtained, reducing the difference between the query and retrieval information and the current demand information of the object, and effectively improving the matching degree between the information retrieval result and the current demand information.

[0094] In some embodiments of the present application, as Figure 3 shown, the process of constructing the vector database includes but is not limited to steps S310 to S340:

[0095] Step S310, obtain first demand document information of a preset field, where the first demand document information includes knowledge information of the preset field;

[0096] Step S320, preprocess the first demand document information to obtain second demand document information;

[0097] Step S330, perform chunking processing on the second demand document information according to a preset chunking strategy to obtain target chunk content information;

[0098] Step S340, construct a vector database according to the target chunk content information.

[0099] Specifically, since the knowledge information of the preset field is relatively scattered, resulting in a large amount of information processing during the retrieval query process, in this embodiment, the scattered knowledge information can be organized into a document to obtain the first demand document information. It can be understood that in some embodiments, when the knowledge information in the first document demand information is obtained from the same website, in this embodiment, after classifying all the knowledge information, the knowledge information of the same type can be organized into the same document as a first document demand information. Exemplarily, taking the medical scenario field as an example, when multiple disease knowledge information is obtained from the same website, the knowledge information can be classified based on the disease name to obtain the first document demand information corresponding to each disease name.

[0100] In other embodiments, when the knowledge information in the first document demand information is obtained from different websites, in this embodiment, the knowledge information can be organized and summarized based on the identification information of each website to obtain the first demand document information corresponding to each website. Exemplarily, taking the financial scenario as an example, when the knowledge information in the first document demand information is obtained from stock website A, securities website B, and savings website C respectively, in this embodiment, the identification information of the corresponding website can be obtained synchronously when obtaining the knowledge information, and then the knowledge information can be organized and summarized according to the identification information to obtain the first demand document information corresponding to stock website A, securities website B, and savings website C respectively.

[0101] It can be understood that after obtaining all the first document requirement information in this embodiment, the first document requirement information will also be preprocessed to obtain the second document requirement information. Specifically, the preprocessing process of this embodiment includes, but is not limited to, processes such as information cleaning and information integration. Exemplarily, taking information cleaning as an example, in this embodiment, by analyzing the first document requirement information, it is determined that there is information such as missing information and abnormal information. For the document requirement information with missing information or abnormal information, context analysis is performed. If the actual information of the missing information or the correct information of the abnormal information can be obtained based on the context meaning, then these information are complemented based on the context information to obtain the second requirement document information; if the actual information of the missing information or the correct information of the abnormal information cannot be obtained based on the context meaning, then the document requirement information corresponding to these information is deleted, and the remaining first document requirement information is used as the second requirement document information. Taking information integration as an example, in this embodiment, by analyzing the information sources of the first requirement document information, the same type of information corresponding to different data sources can be integrated together to obtain the second requirement document information.

[0102] In the embodiment of the present application, since the second requirement document information contains a large amount of knowledge information, the accuracy of query and retrieval is not high. Therefore, in this embodiment, the second requirement document information is block-processed to obtain the target block content information, and then the vector information in the vector database is constructed based on the target block content information. It can be understood that this embodiment can block-process the second requirement document information based on a preset block strategy. Among them, the preset block strategy can be determined based on the type corresponding to the second requirement document information. Exemplarily, if the structure of the second requirement document information is relatively clear, the structure block method can be used to block-process the second requirement document information; if the semantics of the second requirement document information is relatively clear, the semantic block method can be used to block-process the second requirement document information.

[0103] From the above content, it can be seen that in the embodiment of the present application, after obtaining the first requirement document information in the preset field, the first requirement document information is preprocessed, so that the content in the preprocessed second requirement document information is clearer and more accurate; then, after the second requirement document information is block-processed by the preset block strategy, the vector database is constructed based on the target block content information obtained by the block, so that the information meaning represented by each vector information in the vector database is clearer, and further the accuracy of the subsequent information query and retrieval results can be improved.

[0104] In some embodiments of the present application, as Figure 4 shown, the process of preprocessing the first requirement document information to obtain the second requirement document information includes, but is not limited to, steps S321 to S324:

[0105] Step S321: Extract the information that meets the first preset requirement in the first requirement document information as the first document information to be processed;

[0106] Step S322: Extract the information that does not meet the first preset requirement in the first requirement document information as the second document information to be processed;

[0107] Step S323: Perform a structural conversion on the first document information to be processed to obtain the third document information to be processed;

[0108] Step S324: Merge the second document information to be processed and the third document information to be processed to obtain the second requirement document information.

[0109] Specifically, the first preset requirement may be the information format that needs to be processed currently. The information format that needs to be processed currently may be a non-structured format such as a picture or a table. The format of the third document information to be processed may refer to the corresponding structured format for query and retrieval. The information format for query and retrieval may be a text format. In some embodiments, taking the first preset requirement as a picture format as an example, in this embodiment, all picture information is extracted from the first requirement document information as the first document information to be processed, and the remaining information in the first requirement document information is used as the second document information to be processed. Then, after converting the picture information into a format type corresponding to the format of the second requirement document information, the third document information to be processed is obtained. Finally, the second document information to be processed and the third document information to be processed are merged to obtain the second requirement document information.

[0110] In other embodiments, taking the first preset requirement as a table format as an example, in this embodiment, all table information is extracted from the first requirement document information as the first document information to be processed, and the remaining information in the first requirement document information is used as the second document information to be processed. Then, after converting the table information into a format type corresponding to the format of the second requirement document information, the third document information to be processed is obtained. Finally, the second document information to be processed and the third document information to be processed are merged to obtain the second requirement document information.

[0111] In some other embodiments, taking the first preset requirement including picture format and table format as an example, in this embodiment, all picture information and table information are respectively extracted from the first requirement document information as the first document information to be processed, and the remaining information in the first requirement document information is used as the second document information to be processed. Then, after converting both the drawing information and the table information into a format type corresponding to the format of the second requirement document information, for example, after converting both the drawing information and the table information into a text format corresponding to the second requirement document information, the third document information to be processed is obtained. Then, the second document information to be processed and the third document information to be processed are merged to obtain the second requirement document information.

[0112] In the embodiments of the present application, this embodiment can also use a large language model to perform structure conversion on the first document information to be processed to obtain the third document information to be processed. Among them, the large language model of this embodiment can be a pre-trained model, and the training process of this model can be training for the structure conversion of document information. Exemplarily, when it is necessary to convert the picture format into a document format, in this embodiment, after collecting a number of picture data and the text data corresponding to the picture data in advance, the large language model is trained and tested for structure conversion based on the picture data and the text data; when it is necessary to convert the table format into a document format, in this embodiment, after collecting a number of table data and the text data corresponding to the picture data in advance, the large language model is trained and tested for structure conversion based on the table data and the text data. After this embodiment completes the training and testing process of the large language model, the structure conversion is performed on the first document information to be processed based on the trained large language model, thereby effectively improving the accuracy of the structure conversion result.

[0113] From the above content, it can be seen that in the embodiments of the present application, after analyzing the format type of the first requirement document information, all the information formats in the first requirement document information are converted into information formats corresponding to query retrieval, thereby improving the accuracy of the vector information construction in the subsequent vector database.

[0114] In some embodiments of the present application, as Figure 5 shown, the process of performing chunking processing on the second requirement document information according to the preset chunking strategy to obtain the target chunk content information includes, but is not limited to, steps S331 to S332:

[0115] Step S331: Classify the second requirement document information to obtain a classification result;

[0116] Step S332: Perform chunking processing on the second requirement document information according to the classification result and the preset chunking strategy to obtain the target chunk content information.

[0117] Specifically, the ways of information classification include, but are not limited to, classification methods based on nature, classification methods based on carrier, and classification methods based on application fields. Among them, the classification method based on nature can be to classify each piece of information in the second requirement document information based on the nature analysis result after analyzing the nature of each piece of information in the second requirement document information. It can be understood that the information in the second requirement document information can be factual information, directive information, entertainment information, etc. The classification method based on carrier can be to classify each piece of information in the second requirement document information based on the analysis result of the dissemination carrier after analyzing the dissemination carrier of each piece of information in the second requirement document information. Specifically, the dissemination carrier of the information in the second requirement document information can be a written symbol transmission carrier, an oral language dissemination carrier, or a visual element dissemination carrier. The classification method based on application fields is to classify each piece of information in the second requirement document information based on the analysis result of the application field after analyzing the application field corresponding to each piece of information in the second requirement document information. Specifically, the application fields of the information in the second requirement document information can be the commercial field, the industrial field, or the political field.

[0118] It can be understood that in this embodiment, after classifying each piece of information in the second requirement document information to obtain a classification result, the second requirement document information is then segmented by combining a preset segmentation strategy, so that the types of the target segmented content information obtained by segmentation are clearer and the segmentation result is more accurate.

[0119] In some embodiments of the present application, as Figure 6 shown, the process of segmenting the second requirement document information according to the classification result and the preset segmentation strategy to obtain the target segmented content information includes, but is not limited to, steps S610 to S620:

[0120] Step S610: Segment the document content in the second requirement document information according to the classification result to obtain the first segmented content information;

[0121] Step S620: Segment the first segmented content information according to the preset segmentation strategy to obtain the target segmented content information.

[0122] Specifically, in some embodiments, the classification result may include, but is not limited to, the type identifier of each document content in the second requirement document information. The type identifier is used to represent the type to which the corresponding document content belongs. Exemplarily, taking the game field as an example, when the second requirement document information includes content such as game character introductions and skills corresponding to the game characters, an introduction type identifier information may be set in the information content of the game character introductions, and a function type identifier information may be set in the information content of the skills corresponding to the game characters. Then, all the document contents in the second requirement document information are split based on these identifier information, and then the split document contents with the same identifier information are combined to obtain the first chunk content information.

[0123] In other embodiments, the classification result may include, but is not limited to, the content theme information in the second requirement document information. The content theme information is used to briefly summarize the main information of the corresponding text or paragraph. Exemplarily, taking the legal scenario field as an example, since each legal document will contain multiple legal theme contents and the content information corresponding to each legal theme content is different, in this embodiment, after the second requirement document information is divided based on the content theme information, the same or similar contents are combined based on the division result to obtain the first chunk content information.

[0124] It can be understood that in this embodiment, by first performing chunking processing on the second requirement document information based on the classification result, the division process of the second requirement document information in multiple information dimensions can be realized, that is, the content can be refined, and the semantics in each first chunk content information can be complete. Furthermore, the accuracy of the subsequent preset chunking strategy for chunking the first chunk content information can be improved.

[0125] In some embodiments of the present application, as Figure 7 shown, the process of obtaining the target chunk content information by performing chunking processing on the first chunk content information according to the preset chunking strategy includes, but is not limited to, steps S621 to S622:

[0126] Step S621: Select at least one strategy from several preset chunking strategies as the target chunking strategy;

[0127] Step S622: Perform chunking processing on the first chunk content information according to the target chunking strategy to obtain the target chunk content information.

[0128] Specifically, the preset chunking strategy includes a fixed-size chunking strategy, a structural chunking strategy, a semantic chunking strategy, or a recursive chunking strategy. It can be understood that the fixed-size chunking strategy is a process of determining the chunking operation by setting the size of the chunks and whether there is overlap. The structural chunking strategy refers to performing a chunking operation on the document content based on structural perception. The semantic chunking strategy can use punctuation marks, natural paragraphs, or a chunking tool to perform a semantic chunking operation on the document content, or use an Embedding-based method to perform a semantic chunking operation on the document content. The recursive chunking strategy is to divide the input document text into smaller chunks in a hierarchical and iterative manner by using a set of delimiters. During the execution of the recursive chunking strategy, if the initially segmented text does not produce chunks of the required size or structure, the recursive segmentation continues until the conditions of the preset chunk size are met.

[0129] In the embodiment of the present application, this embodiment can select a target chunking strategy based on the application scenario of the question-and-answer system. Exemplarily, when the application scenario of the question-and-answer system is a shopping mall platform, since the question-and-answer information received by the question-and-answer system in the shopping mall platform is relatively concise, therefore, a fixed-size chunking strategy can be selected to perform chunking processing on the first chunk content information, thereby improving the efficiency of chunking processing and reducing the workload of chunking processing. When the application scenario of the question-and-answer system is the legal consulting field, since the information such as punctuation marks and natural paragraphs in the chunked content involved in the legal consulting field is relatively accurate and clear, therefore, a structural chunking strategy or a semantic chunking strategy can be adopted to perform chunking processing on the first chunk content information, thereby effectively improving the semantic accuracy of the target chunked content information. When the application scenario of the question-and-answer system is the financial investment scenario field, this embodiment can adopt a recursive chunking strategy to perform chunking processing on the first chunk content information, thereby making the financial information contained in the target chunked content information after chunking more accurate.

[0130] It can be understood that in this embodiment, different chunking strategies are adopted for the first chunk content information in the same field for chunking processing. Exemplarily, in some embodiments, for the first chunk content information A, the structural chunking strategy and the semantic chunking strategy can be respectively adopted for chunking processing, and then the semantic similarity after chunking processing by the two chunking strategies is analyzed. If the semantic similarity after the two chunking processes is relatively high, it indicates that both chunking strategies can perform chunking processing on the first chunk content information A with relatively high accuracy; if the semantic similarity after the two chunking processes is relatively low, it indicates that at least one of the chunking results corresponding to the two chunking strategies has low accuracy. Therefore, in this embodiment, a third chunking strategy can be adopted to perform chunking processing on the first chunk content information A, and then the accuracy between the chunking results corresponding to the structural chunking strategy and the semantic chunking strategy is calculated respectively, so as to determine the final target chunk content information. In other embodiments, for the first chunk content information B, the fixed-size chunking strategy and the recursive chunking strategy can be respectively adopted to perform chunking processing on the specific content information in the first chunk content information B, and then the chunking results after chunking processing by the two chunking strategies are combined to obtain the target chunk content information.

[0131] As can be seen from the above, in this embodiment, at least one strategy is selected from multiple preset chunking strategies to perform chunking processing on the first chunk content information, so as to effectively improve the accuracy of the chunking result corresponding to the first chunk content information, and further improve the accuracy of subsequent query and retrieval.

[0132] In some embodiments of the present application, as Figure 8 shown, the process of retrieving and querying the target output information in the vector database according to the object query information includes, but is not limited to, steps S231 to S232:

[0133] Step S231: Convert the object query information into an object query vector;

[0134] Step S232: Compare the object query vector with the vectors in the vector database to obtain the target output information.

[0135] Specifically, the object query information may be text information input by a human-computer interaction interface. After obtaining the object query information in this embodiment, the object query information may be converted into an object query vector through a preset conversion method. The preset conversion method includes, but is not limited to, the bag-of-words model, the TF-IDF model (term frequency-inverse document frequency model), or the word2vec model. Exemplarily, in some embodiments, taking the bag-of-words model as an example, after inputting the object query information into the bag-of-words model in this embodiment, the number of times each word in the object query information appears is determined within the bag-of-words model, and this number is used as a feature value. The object query vector corresponding to the object query information is obtained through conversion according to the feature values of all words. In other embodiments, taking the word2vec model as an example, after performing word segmentation on the object query information in this embodiment, the corresponding words are mapped into a vector space to obtain the vectors corresponding to each word, and then the average value of all word vectors is solved as the object query vector.

[0136] It can be understood that after obtaining the object query vector in this embodiment, the object query vector is compared with the vectors in the vector database to determine the target output information. Exemplarily, the comparison process in this embodiment may be to compare each element in the object query vector with each element in each vector in the vector database to determine the number of equal elements. Specifically, the more equal elements there are, the closer the current vector in the vector database is to the object query vector. The vector in the vector database with the largest number of equal elements is selected as the target vector of the object query vector, and the content information corresponding to the target vector is obtained as the target output information, thereby effectively improving the matching degree between the target output information and the current demand information.

[0137] In some embodiments of the present application, as Figure 9 shown, the process of comparing the object query vector with the vectors in the vector database to obtain the target output information includes, but is not limited to, steps S910 to S930:

[0138] Step S910: Calculate the similarity between the object query vector and the vectors in the vector database;

[0139] Step S920: Obtain at least one vector to be processed from the vector database according to the similarity;

[0140] Step S930: Generate the target output information according to the vector to be processed.

[0141] Specifically, in some embodiments, the similarity in this embodiment can be obtained by respectively calculating the cosine similarity of the object query vector and each vector in the vector database. Among them, the larger the cosine similarity, the more similar the two vectors are. After obtaining the cosine similarity of the object query vector and the vectors in the vector database, this embodiment selects at least one vector with a larger cosine similarity from the vector database as the vector to be processed, and then composes the target output information according to the content information corresponding to the vector to be processed.

[0142] In other embodiments, the present embodiment may also determine the final similarity by the cosine distance and the Euclidean distance between two vectors. Specifically, the present embodiment first calculates the cosine distance between the object query vector and each vector in the vector database, then synchronously calculates the Euclidean distance between the object query vector and each vector in the vector database, and then combines the weighted calculations of the cosine distance and the Euclidean distance to obtain a weighted value as the similarity of the present embodiment. It is understandable that after the present embodiment obtains the weighted similarity, at least one vector is obtained from the vector database as a vector to be processed based on the weighted similarity, and then the target output information is composed according to the content information corresponding to the vector to be processed.

[0143] From the above content, it can be seen that the similarity between the vectors is used to determine the final target output information, thereby effectively improving the matching degree between the target output information and the current demand information.

[0144] In some embodiments of the present application, Figure 10 As shown, the process of generating target output information according to the vector to be processed includes but is not limited to steps S931 to S933:

[0145] Step S931, obtaining a preset prompt template;

[0146] Step S932: Integrate the object query vector and the vector to be processed by using a preset prompt template to obtain prompt content information;

[0147] Step S933: Generate target output information according to the prompt content information.

[0148] Specifically, the preset prompt template may be a prompt framework or prompt structure pre-built according to a preset domain. The preset prompt template includes at least one variable, which can be replaced in the context corresponding to the preset domain, thereby generating a corresponding information prompt. Exemplarily, the preset prompt template of this embodiment may be represented by the following pseudo code:

[0149]

[0150]

[0151] As can be seen from the pseudocode of the above preset prompt template, in the second step of query retrieval in this embodiment, the top k to-be-processed vectors with relatively high similarity to the object query vector can be obtained. At the same time, after embedding the additionally retrieved context information (context_str) into the preset position of the prompt template, the corresponding target output information can be generated.

[0152] In summary, the embodiment of the present application retrieves and enhances the generation (RAG) of the target output information through a preset prompt template, so that the target output information can be determined by combining the context information in the query retrieval process, effectively improving the coherence and accuracy of the target output information.

[0153] In some embodiments of the present application, as Figure 11 shown, the process of generating the target output information according to the prompt content information includes, but is not limited to, steps S1101 to S1103:

[0154] Step S1101: Obtain the object usage feedback information;

[0155] Step S1102: Adjust the parameters of the first preset large language model according to the object usage feedback information to obtain the second preset large language model;

[0156] Step S1103: Input the prompt content information into the second preset large language model to generate the target output information.

[0157] Specifically, in the query retrieval process of this embodiment, the usage feedback information of the object can be continuously received. Among them, the usage feedback information may include the satisfaction of the object with the current retrieval query result and the new query information input based on the current query result. It can be understood that when the object feedback information is received, this embodiment can adjust the parameters of the first preset large language model based on the feedback information, so that the information processing process of the second preset large language model after parameter adjustment is more accurate.

[0158] In the embodiment of the present application, the first preset large language model includes, but is not limited to, a neural network model or a knowledge graph model. Specifically, when the first preset large language model adopts a neural network model, the neural network model can adopt the GPT model. In this embodiment, the GPT model is trained in a large-scale corpus and used for the information processing in the query retrieval process of the present application. When the first preset large language model adopts a model structure combining the GPT model and the knowledge graph model, after training the GPT model and the knowledge graph model, text information query retrieval is performed based on the GPT model, and auxiliary query retrieval is performed through the knowledge graph model.

[0159] As can be seen from the above, in the embodiments of the present application, by synchronously obtaining object feedback information during the query and retrieval process and adjusting the parameters of the preset large language model according to the object feedback information, the accuracy of the target output information can be effectively improved.

[0160] In some embodiments of the present application, when the method of the embodiments of the present application is applied to Figure 12 the system architecture shown, as Figure 13 shown, this embodiment includes but is not limited to steps S1301 to S1306:

[0161] Step S1301: Obtain the first requirement document information from the document knowledge base of the preset domain;

[0162] Step S1302: Perform document chunking processing on the first requirement document information to obtain the target chunk content information;

[0163] Step S1303: Generate vector information according to the target chunk content information, and save the vector information to the vector database;

[0164] Step S1304: Obtain the object query information generated based on the object question;

[0165] Step S1305: Perform context query and retrieval in the vector database according to the object query information, and perform retrieval enhancement processing in combination with the prompt template;

[0166] Step S1306: Request the large language model to output the target output information based on the retrieval enhancement processing result of the context.

[0167] In summary, in the embodiments of the present application, by constructing a vector database based on the requirement document information of the preset domain, it is possible to dynamically adapt to new information when facing the constantly changing requirement documents. Furthermore, when the method of the present application is applied to a retrieval system (QA assistant), the target output information can be quickly queried and retrieved, effectively improving the efficiency and accuracy of query and retrieval.

[0168] Corresponding to the above method embodiments, the present invention also provides embodiments of a question and answer information processing device, Figure 14 showing a schematic structural diagram of a question and answer information processing device according to an embodiment of the present invention. As Figure 14 shown, the question and answer information processing device includes:

[0169] A construction module 1410, configured to construct a vector database, where the vector database includes vector information corresponding to the knowledge information of the preset domain;

[0170] An acquisition module 1420, configured to acquire object query information;

[0171] A retrieval query module 1430 is configured to perform retrieval queries in a vector database based on object query information to obtain target output information.

[0172] The above is a schematic solution of a question-and-answer information processing device according to this embodiment. It should be noted that the technical solution of this question-and-answer information processing device and the technical solution of the above question-and-answer information processing method applied to the question-and-answer information processing device belong to the same concept. For the details not described in the technical solution of the question-and-answer information processing device, reference can be made to the description of the technical solution of the above question-and-answer information processing method.

[0173] As Figure 15 shown, Figure 15 FIG. shows a block diagram of the structure of an electronic device 1500 according to an embodiment of the present invention. The components of the electronic device 1500 include, but are not limited to, a memory 1510 and a processor 1520. The processor 1520 is connected to the memory 1510 through a bus 1530, and a database 1550 is used to store data.

[0174] The electronic device 1500 further includes an access device 1540, which enables the electronic device 1500 to communicate via one or more networks 1560. Examples of these networks include the Public Switched Telephone Network (PSTN), Local Area Network (LAN), Wide Area Network (WAN), Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 1540 may include one or more of any type of wired or wireless network interfaces (e.g., Network Interface Card (NIC)), such as an IEEE802.11 Wireless Local Area Network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, a Near Field Communication (NFC) interface, and so on.

[0175] In an embodiment of the present invention, the above components of the electronic device 1500 and Figure 8 other components not shown in Figure 8 may also be connected to each other, for example, through a bus. It should be understood that the block diagram of the electronic device structure shown in

[0176] The electronic device 1500 can be any type of stationary or mobile electronic device, including mobile computers or mobile electronic devices (e.g., tablet computers, personal digital assistants, laptop computers, notebook computers, netbooks, etc.), mobile phones (e.g., smart phones), wearable electronic devices (e.g., smart watches, smart glasses, etc.) or other types of mobile devices, or stationary electronic devices such as desktop computers or PCs. The electronic device 1500 can also be a mobile or stationary server.

[0177] Among them, the processor 1520 is used to execute the computer-executable instructions of the above question-and-answer information processing method.

[0178] The above is a schematic solution of an electronic device in this embodiment. It should be noted that the technical solution of this electronic device and the technical solution of the above question-and-answer information processing method belong to the same concept. For the details not described in the technical solution of the electronic device, reference can be made to the description of the technical solution of the above question-and-answer information processing method.

[0179] The embodiment of the present invention also provides a storage medium, which is a computer-readable storage medium. The storage medium stores a computer program, and when the computer program is executed by a processor, the above question-and-answer information processing method is implemented.

[0180] As a non-transitory computer-readable storage medium, the memory can be used to store non-transitory software programs and non-transitory computer-executable programs. In addition, the memory can include high-speed random access memory, and can also include non-transitory memory, such as at least one magnetic disk storage device, a flash memory device, or other non-transitory solid-state storage devices. In some embodiments, the memory can optionally include memories remotely provided relative to the processor, and these remote memories can be connected to the processor through a network. Examples of the above networks include but are not limited to the Internet, enterprise intranets, local area networks, mobile communication networks, and combinations thereof. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and may be located in one place, or may be distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0181] Those of ordinary skill in the art will understand that all or some of the steps and systems disclosed above can be implemented as software, firmware, hardware, and their appropriate combinations. Some or all of the physical components can be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or can be implemented as hardware, or can be implemented as an integrated circuit, such as an application-specific integrated circuit. Such software can be distributed on a computer-readable medium, which can include a computer storage medium (or non-transitory medium) and a communication medium (or transitory medium). As is well known to those of ordinary skill in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information, such as computer-readable instructions, data structures, program modules, or other data. Computer storage media includes, but is not limited to, RAM, ROM, EEPROM, flash memory or other memory technologies, CD-ROM, digital versatile disk (DVD) or other optical disk storage, magnetic cassettes, tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, as is well known to those of ordinary skill in the art, communication media typically includes computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transmission mechanism, and can include any information delivery medium.

[0182] The above is a specific description of the preferred embodiments of the present invention, but the present invention is not limited to the above embodiments. Those skilled in the art can make various equivalent deformations or substitutions without departing from the spirit of the present invention, and these equivalent deformations or substitutions are all included within the scope defined by the claims of the present invention.

Claims

1. A question-answer information processing method, characterized in that: The method comprises the following steps: Constructing a vector database, wherein the vector database includes vector information corresponding to knowledge information in a preset field; Get object query information; A search query is performed in the vector database according to the object query information to obtain target output information.

2. The method according to claim 1, characterized in that The constructing of the vector database comprises: Acquire first requirement document information of the preset field, wherein the first requirement document information includes knowledge information of the preset field; Preprocessing the first requirement document information to obtain second requirement document information; The second requirement document information is processed into blocks according to a preset block strategy to obtain target block content information; The vector database is constructed according to the target block content information.

3. The method according to claim 2, characterized in that The preprocessing of the first requirement document information to obtain the second requirement document information includes: Extracting information that meets the first preset requirement from the first demand document information as first document information to be processed; Extracting information that does not meet the first preset requirement from the first demand document information as second document information to be processed; Performing structural conversion on the first document information to be processed to obtain third document information to be processed; The second to-be-processed document information and the third to-be-processed document information are merged to obtain the second required document information.

4. The method according to claim 2, characterized in that: The step of performing block processing on the second requirement document information according to a preset block strategy to obtain target block content information includes: Classifying the second requirement document information to obtain a classification result; The second demand document information is segmented according to the classification result and a preset segmentation strategy to obtain the target segmentation content information.

5. The method according to claim 4, characterized in that The step of performing block processing on the second requirement document information according to the classification result and the preset block strategy to obtain the target block content information includes: According to the classification result, the document content in the second demand document information is processed into blocks to obtain first block content information; The first block content information is processed by block processing according to the preset block strategy to obtain the target block content information.

6. The method according to claim 5, characterized in that The step of performing block processing on the first block content information according to the preset block strategy to obtain the target block content information includes: Selecting at least one strategy from a plurality of preset block strategies as a target block strategy, wherein the preset block strategy includes a fixed size block strategy, a structural block strategy, a semantic block strategy or a recursive block strategy; The first block content information is processed by block processing according to the target block strategy to obtain the target block content information.

7. The method according to claim 1, characterized in that The step of performing a search query in the vector database according to the object query information to obtain target output information includes: Converting the object query information into an object query vector; The object query vector is compared with the vectors in the vector database to obtain the target output information.

8. The method according to claim 7, characterized in that The step of comparing the object query vector with the vectors in the vector database to obtain the target output information includes: Calculating the similarity between the object query vector and the vectors in the vector database; Acquire at least one vector to be processed from the vector database according to the similarity; The target output information is generated according to the vector to be processed.

9. The method according to claim 8, characterized in that The generating the target output information according to the vector to be processed includes: Get the preset prompt template; Integrating the object query vector and the vector to be processed by using the preset prompt template to obtain prompt content information; The target output information is generated according to the prompt content information.

10. The method according to claim 9, characterized in that The generating the target output information according to the prompt content information includes: Get object usage feedback information; Adjusting parameters of the first preset large language model according to the object usage feedback information to obtain a second preset large language model; The prompt content information is input into the second preset large language model to generate the target output information.

11. A question and answer information processing device, characterized in that: The device comprises: A construction module, used to construct a vector database, wherein the vector database includes vector information corresponding to knowledge information in a preset field; An acquisition module is used to obtain object query information; The retrieval query module is used to perform a retrieval query in the vector database according to the object query information to obtain target output information.

12. An electronic device, characterized in that: It includes at least one control processor and a memory for communicating with the at least one control processor; the memory stores instructions that can be executed by the at least one control processor, and the instructions are executed by the at least one control processor to enable the at least one control processor to execute the question and answer information processing method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, and the computer-executable instructions are used to enable a computer to execute the question and answer information processing method according to any one of claims 1 to 10.

Images