Question and answer information processing method and device based on intelligent knowledge base and electronic equipment

Abstract

The embodiment of the invention discloses a question and answer information processing method and device based on an intelligent knowledge base and electronic equipment. A specific embodiment of the method comprises the following steps: in response to received initial question information sent by a target user, identifying each question keyword in the initial question information, and querying a first alternative question fragment sequence and a second alternative question fragment sequence corresponding to the question information from a pre-constructed intelligent knowledge base; in response to determining that the first alternative question fragment sequence has a corresponding question reply document, retrieving a target number of question fragments in a distributed search database according to the question reply document and the question information; and inputting the target number of question fragments into the intelligent question and answer large model to generate question reply information corresponding to the question information. According to the embodiment, whether the question information is a familiar question type can be determined, intelligent question answering is performed through the large model, and the accuracy of the intelligent question is improved.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technologies, and more particularly, to a method, an apparatus, and an electronic device for processing question-and-answer information based on an intelligent knowledge base. Background Art

[0002] Currently, with the advent of the era of artificial intelligence, the application of intelligent question-and-answer is becoming increasingly widespread in people's daily lives. Currently, when processing question-and-answer information raised by users, the commonly adopted method is: the user directly inputs the question into a large language model to let the large language model output the answer content to the question.

[0003] However, when using the above method, the following technical problems often exist: the answers are not given in combination with various historical consultation questions, and the questions input by the user may have semantic ambiguities or the natural language processing ability of the large language model is poor, resulting in the large language model being difficult to accurately answer the questions.

[0004] The above information disclosed in this background art section is only used to enhance the understanding of the background of the inventive concept, and thus, it may include information that does not form the prior art known to ordinary skilled artisans in the art of this country. Summary of the Invention

[0005] The content part of the present disclosure is used to briefly introduce the concepts, which will be described in detail in the subsequent detailed implementation part. The content part of the present disclosure is not intended to identify the key features or essential features of the claimed technical solution, nor is it intended to be used to limit the scope of the claimed technical solution.

[0006] Some embodiments of the present disclosure propose a method, an apparatus, an electronic device, and a computer-readable medium for processing question-and-answer information based on an intelligent knowledge base to solve one or more of the technical problems mentioned in the above background art section.

[0007] In a first aspect, some embodiments of the present disclosure provide a method for processing question-and-answer information based on an intelligent knowledge base, which is applied to an intelligent customer service assistant. The method includes: in response to receiving initial question information sent by a target user, identifying each question keyword in the initial question information as question information, and querying a first alternative question fragment sequence and a second alternative question fragment sequence corresponding to the question information from a pre-constructed intelligent knowledge base, where the intelligent knowledge base includes: a native vector database and a distributed search database, the first alternative question fragment sequence corresponds to the native vector database, and the second alternative question fragment sequence corresponds to the distributed search database; determining whether the question type corresponding to the second alternative question fragment sequence is a preset question type; in response to determining that the question types corresponding to the second alternative question fragment sequence are not preset question types, determining whether there is a question answer document corresponding to the first alternative question fragment sequence that satisfies the target matching condition; in response to determining that there is a corresponding question answer document for the first alternative question fragment sequence, retrieving a target number of third alternative question fragments in the distributed search database according to the question answer document and the question information; performing splicing processing on the target number of third alternative question fragments to obtain spliced question fragment information, and inputting the spliced question fragment information and the question information into a pre-trained intelligent question-and-answer large model to generate question answer information corresponding to the question information, and sending the question answer information to the user terminal corresponding to the target user.

[0008] Second aspect, some embodiments of the present disclosure provide a question and answer information processing device based on an intelligent knowledge base. The device includes: a query unit configured to, in response to receiving initial question information sent by a target user, identify each question keyword in the initial question information as question information, and query a first alternative question fragment sequence and a second alternative question fragment sequence corresponding to the question information from a pre-constructed intelligent knowledge base. The intelligent knowledge base includes a native vector database and a distributed search database. The first alternative question fragment sequence corresponds to the native vector database, and the second alternative question fragment sequence corresponds to the distributed search database; a first determination unit configured to determine whether the question type corresponding to the second alternative question fragment sequence is a preset question type; a second determination unit configured to, in response to determining that none of the question types corresponding to the second alternative question fragment sequence are preset question types, determine whether there is a question answer document corresponding to the first alternative question fragment sequence that meets the target matching condition; a retrieval unit configured to, in response to determining that there is a corresponding question answer document for the first alternative question fragment sequence, retrieve a target number of third alternative question fragments in the distributed search database according to the question answer document and the question information; an input unit configured to splice the target number of third alternative question fragments to obtain spliced question fragment information, and input the spliced question fragment information and the question information into a pre-trained intelligent question and answer large model to generate question answer information corresponding to the question information, and send the question answer information to the user terminal corresponding to the target user.

[0009] Third aspect, some embodiments of the present disclosure provide an electronic device, including: one or more processors; a storage device storing one or more programs thereon, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner of the first aspect.

[0010] Fourth aspect, some embodiments of the present disclosure provide a computer-readable medium storing a computer program thereon, wherein when the program is executed by a processor, the method described in any implementation manner of the first aspect is implemented.

[0011] The above-mentioned various embodiments of the present disclosure have the following beneficial effects: Through the question-and-answer information processing method based on an intelligent knowledge base according to some embodiments of the present disclosure, the question information is retrieved in fragments through a pre-established intelligent knowledge base; thereby, it can be determined whether the question information is a familiar question type. Then, using the retrieved question fragments, intelligent question answering is performed through a large model, improving the accuracy of intelligent questions. First, in response to receiving the question information sent by the target user, each question keyword in the above-mentioned initial question information is identified as the question information, and the first alternative question fragment sequence and the second alternative question fragment sequence corresponding to the above-mentioned question information are retrieved from the pre-constructed intelligent knowledge base, where the above-mentioned intelligent knowledge base includes: a native vector database and a distributed search database, the first alternative question fragment sequence corresponds to the native vector database, and the second alternative question fragment sequence corresponds to the distributed search database. Thus, different question fragments corresponding to the question information can be determined. Next, it is determined whether the question types corresponding to the first alternative question fragment sequence and the second alternative question fragment sequence are preset question types. Thus, it can be determined whether the question information is a conventional question, facilitating the determination of the question-and-answer method. Then, in response to determining that the question types corresponding to the first alternative question fragment sequence and the second alternative question fragment sequence are not preset question types, it is determined whether there is a question reply document corresponding to the first alternative question fragment sequence that meets the target matching condition. Thus, it can be determined whether there is a question reply document containing the question fragment. Finally, in response to determining that there is a corresponding question reply document for the first alternative question fragment sequence, a target number of question fragments are retrieved from the above-mentioned question reply document and the above-mentioned question information in the above-mentioned distributed search database; the above-mentioned target number of question fragments are input into a pre-trained intelligent question answering large model to generate question reply information corresponding to the above-mentioned question information, and the above-mentioned question reply information is sent to the user terminal corresponding to the above-mentioned target user. Thus, through the pre-established intelligent knowledge base, the question information is retrieved in fragments; thereby, it can be determined whether the question information is a familiar question type. Then, using the retrieved question fragments, intelligent question answering is performed through a large model, improving the accuracy of intelligent questions. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In combination with the accompanying drawings and with reference to the following specific embodiments, the above and other features, advantages and aspects of the various embodiments of the present disclosure will become more obvious. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic and the elements and elements are not necessarily drawn to scale.

[0013] Figure 1 is a flowchart of some embodiments of the question-and-answer information processing method based on an intelligent knowledge base according to the present disclosure;

[0014] Figure 2Schematic structural diagram of some embodiments of a question-answering information processing device based on an intelligent knowledge base according to the present disclosure;

[0015] Figure 3 Schematic structural diagram of an electronic device suitable for implementing some embodiments of the present disclosure;

[0016] Figure 4 Schematic diagram of a page for a customer service assistant to reply to question information in a question-answering information processing method based on an intelligent knowledge base according to the present disclosure. Detailed implementation manners

[0017] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. On the contrary, these embodiments are provided to more thoroughly and completely understand the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not used to limit the protection scope of the present disclosure.

[0018] In addition, it should be noted that for the sake of convenience of description, only parts related to the relevant invention are shown in the drawings. Without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other.

[0019] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependent relationships.

[0020] It should be noted that the modifications of "one" and "multiple" mentioned in the present disclosure are illustrative rather than restrictive. Those skilled in the art should understand that unless otherwise clearly specified in the context, it should be understood as "one or more".

[0021] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only for illustrative purposes and are not used to limit the scope of these messages or information.

[0022] The present disclosure will be described in detail below with reference to the drawings and in combination with the embodiments.

[0023] Figure 1 Flow 100 of some embodiments of a question-answering information processing method based on an intelligent knowledge base according to the present disclosure is shown. The question-answering information processing method based on an intelligent knowledge base, which is applied to an intelligent customer service assistant, includes the following steps:

[0024] Step 101: In response to receiving the initial question information sent by the target user, identify each question keyword in the above initial question information as question information, and query the first alternative question fragment sequence and the second alternative question fragment sequence corresponding to the above question information from the pre-constructed intelligent knowledge base.

[0025] In some embodiments, the execution subject of the question and answer information processing method based on the intelligent knowledge base (e.g., a computing device) can, in response to receiving the initial question information sent by the target user, identify each question keyword in the above initial question information as question information, and query the first alternative question fragment sequence and the second alternative question fragment sequence corresponding to the above question information from the pre-constructed intelligent knowledge base. Among them, the above intelligent knowledge base includes: a native vector database and a distributed search database. The first alternative question fragment sequence corresponds to the native vector database, and the second alternative question fragment sequence corresponds to the distributed search database. The question information can represent the voice consultation information or text consultation information submitted by the target user. The intelligent knowledge base can be a question and answer knowledge base for various questions pre-constructed by combining technologies such as full-text retrieval, vector models, and AIGC large models. For example, the intelligent knowledge base can store questions about securities, insurance, etc. The native vector database can refer to the milvus vector database. The distributed search database can refer to the electric search vector database. For example, first, convert the above question information into word vectors; then, the first N question fragments with the highest similarity can be retrieved from the native vector database as the first alternative question fragment sequence. The first N question fragments with the highest similarity can be retrieved from the distributed search database as the second alternative question fragment sequence. A question fragment can refer to a question fragment after slicing a specific question. Multiple question fragments form a complete question. The initial question information can include the question asked by the target user and tags. Among them, the tags can represent the question type. One question type corresponds to one question document type. The intelligent customer service assistant can refer to the intelligent agent.

[0026] For example, each question keyword in the above initial question information can be identified as question information. For example, the initial question information can be sliced by slicing, and each corresponding question keyword can be found from each phrase obtained from the slicing. For another example, each question keyword in the initial question information can be identified by a pre-trained question keyword recognition model. For example, the question keyword recognition model can be a recurrent neural network (RNN), a convolutional neural network (CNN), a Transformer, a generative adversarial network (GAN), or a long short-term memory network (LSTM).

[0027] In practice, the above-mentioned execution entity can query the first alternative question segment sequence and the second alternative question segment sequence corresponding to the above-mentioned question information from a pre-constructed intelligent knowledge base through the following steps:

[0028] First, according to the tags included in the above-mentioned question information, query the initial native alternative document set from the above-mentioned native vector database. For example, various native alternative documents corresponding to the tags can be queried from the above-mentioned native vector database as the initial native alternative document set. Among them, one native alternative document corresponds to one question document type.

[0029] Second, for each initial native alternative document in the above-mentioned initial native alternative document set, perform the following processing steps:

[0030] 1. Query the first initial question segment corresponding to the above-mentioned question information from the above-mentioned initial native alternative document.

[0031] 2. Merge the context segment of the above-mentioned first initial question segment with the first initial question segment to obtain the first alternative question segment. For example, the question segment of the previous paragraph of the first initial question segment, the first initial question segment, and the question segment of the next paragraph of the first initial question segment can be merged to obtain the first alternative question segment.

[0032] Third, determine the obtained first alternative question segments as the first alternative question segment sequence.

[0033] Fourth, according to the tags included in the above-mentioned question information, query the initial distributed alternative document set from the above-mentioned distributed search database. For example, various distributed alternative documents corresponding to the tags can be queried from the above-mentioned distributed search database as the initial distributed alternative document set. Among them, one distributed alternative document corresponds to one question document type.

[0034] Fifth, for each initial distributed alternative document in the above-mentioned initial distributed alternative document set, perform the following processing steps:

[0035] 1. Query the second initial question segment corresponding to the above-mentioned question information from the above-mentioned initial distributed alternative document.

[0036] 2. Merge the context segment of the above-mentioned second initial question segment with the second initial question segment to obtain the second alternative question segment. For example, the question segment of the previous paragraph of the second initial question segment, the second initial question segment, and the question segment of the next paragraph of the second initial question segment can be merged to obtain the second alternative question segment.

[0037] 3. Determine the obtained second alternative question segments as the second alternative question segment sequence.

[0038] Optionally, the intelligent knowledge base can be constructed through the following steps:

[0039] In the first step, obtain a set of historical question and answer documents. The historical question and answer documents can be the documents that have been used historically and include questions and answers.

[0040] In the second step, perform format conversion on each historical question and answer document in the set of historical question and answer documents to generate a converted historical question and answer document, and obtain a set of converted historical question and answer documents. For example, the historical question and answer document can be converted into a fixed format. For example, the fixed format can be a document format (PDF).

[0041] In the third step, perform slicing on each converted historical question and answer document in the set of converted historical question and answer documents to generate a group of question fragments, and obtain a set of question fragment groups. For example, first, after converting the document into text format, slice it according to the line break character, and ignore empty lines by default; second, slice it preferentially according to two or more line break characters, and define it as a paragraph, and check the paragraph length. ① Too long: If it exceeds 1000 characters, slice it again according to a single line break character; ② Too short: If it is less than 5 characters, splice the fragment with the following fragment until it is greater than 200 characters and then stop; then, if it still exceeds 1000 characters after slicing according to a single line break character, slice it according to symbols, and preferentially slice it according to full stops. ① Too long: If it exceeds 1000 characters, slice it again according to semicolons; ② Too short: If it is less than 5 characters, splice the sentence with the next sentence until it is greater than 200 characters and then stop (if the result of slicing by a single symbol still exceeds 1000 characters, slice it according to the next symbol in priority); finally, if it still exceeds 1000 characters after slicing according to all symbols, slice the fragment according to the rule of (number of characters / 1000 + 1).

[0042] In the fourth step, store the set of question fragment groups and the corresponding set of historical question and answer documents into the native vector database and the distributed search database respectively according to a preset format. Here, the preset format can be a vector format, a picture format, or a text format; it can be set according to requirements.

[0043] Step 102, determine whether the question type corresponding to the above second alternative question fragment sequence is a preset question type.

[0044] In some embodiments, the above-mentioned execution entity may determine whether the question type corresponding to the above-mentioned second alternative question fragment sequence is a preset question type. The preset question type may refer to the frequently-asked questions (FAQ) type. Here, the FAQ type may be a pre-set question type. For example, the question types may include: type A, type B, type C, type D, type E. "Type A" and "Type D" may be set as the FAQ types. It should be noted that each question fragment corresponds to a question type. The question types corresponding to each question fragment may be the same or different.

[0045] Step 103, in response to determining that none of the question types corresponding to the second alternative question fragment sequence are preset question types, determine whether there is a question response document corresponding to the above-mentioned first alternative question fragment sequence that meets the target matching condition.

[0046] In some embodiments, the above-mentioned execution entity may, in response to determining that none of the question types corresponding to the second alternative question fragment sequence are preset question types, determine whether there is a question response document corresponding to the above-mentioned first alternative question fragment sequence that meets the target matching condition. The target matching condition may be that there is only one question response document in the native vector database that contains the most first alternative question fragments. For example, the first alternative question fragment sequence includes: question fragment A, question fragment B, question fragment C, question fragment D; among them, there is a question response document A in the native vector database that contains "question fragment A, question fragment B, question fragment C"; there is a question response document B in the native vector database that contains "question fragment A, question fragment B"; there is a question response document C in the native vector database that contains "question fragment A, question fragment D", then the question response document A meets the target matching condition.

[0047] For another example, there is a question response document A in the native vector database that contains "question fragment A, question fragment B, question fragment C"; there is a question response document B in the native vector database that contains "question fragment A, question fragment B, question fragment D"; there is a question response document C in the native vector database that contains "question fragment A, question fragment D", and the number of the question response document A and the question response document B is greater than or equal to two, then there is no question response document that meets the target matching condition for the above-mentioned first alternative question fragment sequence.

[0048] Step 104, in response to determining that there is a corresponding question response document for the above-mentioned first alternative question fragment sequence, retrieve a target number of third alternative question fragments from the above-mentioned distributed search database according to the above-mentioned question response document and the above-mentioned question information.

[0049] In some embodiments, the above-mentioned execution entity may, in response to determining that there is a corresponding question answer document for the above-mentioned first alternative question segment sequence, retrieve a target number of third alternative question segments in the above-mentioned distributed search database according to the above-mentioned question answer document and the above-mentioned question information. For example, a target number of third alternative question segments associated with the question answer document and the question information may be retrieved through the distributed search database. Here, the target number of third alternative question segments corresponds to both the question answer document and the question information. For example, a target number of third alternative question segments with the greatest simultaneous similarity to the question answer document and the question information may be retrieved from the above-mentioned distributed search database.

[0050] In practice, the above-mentioned execution entity may retrieve a target number of third alternative question segments in the above-mentioned distributed search database through the following steps:

[0051] First step, retrieve a target number of third initial question segments in the above-mentioned question answer document through the above-mentioned distributed search database.

[0052] Second step, for each of the above-mentioned target number of third initial question segments, merge the context segment of the above-mentioned third initial question segment with the third initial question segment to obtain a third alternative question segment.

[0053] Step 105, perform splicing processing on the above-mentioned target number of third alternative question segments to obtain spliced question segment information, and input the above-mentioned spliced question segment information and the above-mentioned question information into a pre-trained intelligent question-answering large model to generate question answer information corresponding to the above-mentioned question information, and send the above-mentioned question answer information to the user terminal corresponding to the above-mentioned target user.

[0054] In some embodiments, the above-mentioned execution entity may perform splicing processing on the above-mentioned target number of third alternative question segments to obtain spliced question segment information, and input the above-mentioned spliced question segment information and the above-mentioned question information into a pre-trained intelligent question-answering large model to generate question answer information corresponding to the above-mentioned question information, and send the above-mentioned question answer information to the user terminal corresponding to the above-mentioned target user. The intelligent question-answering large model may be a pre-trained large language model that takes question information as input and question answer information as output. For example, the intelligent question-answering large model may be a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), or a generative adversarial network (GAN). For example, the question information may be input into a pre-trained intelligent question-answering large model to obtain question answer information corresponding to the above-mentioned question information. The user terminal may refer to the mobile terminal / computing terminal of the target user.

[0055] For example, a target number of third alternative question segments can be sequentially spliced into spliced question segment information. Then, the above-mentioned spliced question segment information and the above-mentioned question information are input into a pre-trained large intelligent question-answering model to generate question reply information corresponding to the above-mentioned question information.

[0056] In practice, the above-mentioned execution entity can generate question reply information corresponding to the above-mentioned question information through the following steps:

[0057] In the first step, through the first question-answering network model group included in the large intelligent question-answering model, question-related information corresponding to the above-mentioned question information is generated. Among them, each first question-answering network model in the first question-answering network model group can be a large language model with different model structures, question types that the model is good at, etc. The question-related information can be information related to the semantic content of the question corresponding to the question information. For example, the question-related information can include: questions similar to the question information, the question reply content of the similar questions, and question domain inquiry information related to the question information. For example, first, the above-mentioned execution entity can generate associated prompt information corresponding to the above-mentioned question information. Then, the associated prompt information is input into each first question-answering network model in the above-mentioned first question-answering network model group to obtain each output information. Finally, each information in each output information is summarized to obtain the question-related information. Among them, the associated prompt information can be a prompt word (Prompt) representing the generation of question information related to the question information. The first question-answering network model can be a trained large language model.

[0058] In the second step, according to the above-mentioned question information and the above-mentioned question-related information, using the above-mentioned first question-answering network model group, question prompt information corresponding to the above-mentioned question information is generated. Among them, the question prompt information can be a prompt word representing the generation of the question reply content corresponding to the question information. For example, first, prompt generation information for generating corresponding question information according to the above-mentioned question information and the above-mentioned question-related information is generated. Then, a first question-answering network model is randomly selected from the first question-answering network model group as the target question-answering network model. Finally, the prompt generation information is input into the target question-answering network model to obtain the question prompt information. Among them, the prompt generation information can be prompt information for generating a prompt word for answering the question information based on the question-related information.

[0059] Among them, the above-mentioned second step can include:

[0060] The first sub-step is to determine the problem intention information corresponding to the above problem information. Among them, the problem intention information can be the content intention of the main inquiry content corresponding to the problem information. For example, first, the above execution entity can perform text preprocessing on the problem information to obtain a preprocessing result. Among them, the text preprocessing can include: word segmentation processing, stop word removal processing, and stemming processing. Then, the preprocessing result is input into a word vector conversion model (Word2Vec model, BERT model) to obtain a text vector. Finally, the above text vector is input into an intention classification model to obtain the problem intention information. For example, the intention classification model can be: Support Vector Machine (SVM), random forest model.

[0061] The second sub-step is to, in response to determining that the confidence level corresponding to the above problem intention information is greater than the preset confidence level, use the above first Q&A network model group to generate problem prompt information corresponding to the above problem information according to the above problem information, the above problem intention information, and the above problem association information. Among them, the confidence level corresponding to the problem intention information can represent the generation accuracy of the problem intention information. That is, the larger the confidence level value, the more accurate the problem intention information is. The confidence level can be generated by the intention classification model together with the problem intention information. The problem prompt information corresponding to the problem information can be a prompt word representing the reply to the problem information. For example, first, the above execution entity can generate prompt information for generating prompt words based on the above problem information, the above problem intention information, and the above problem association information. Then, the prompt information is input into any one of the first Q&A network models in the first Q&A network model group to obtain the problem prompt information corresponding to the problem information.

[0062] The third sub-step is to, in response to determining that the confidence level corresponding to the above problem intention information is less than or equal to the preset confidence level, use the above first Q&A network model group to determine the intention description information of the problem reply corresponding to the above problem intention information. For example, first, the above execution entity can randomly select a first Q&A network model from the first Q&A network model group. Then, use the above first Q&A network model to generate intention supplementary information representing the description supplement of the problem intention information. Next, display the intention supplementary information on the interaction interface corresponding to the above first Q&A network model to allow the target user to supplement the intention for the intention supplementary information to obtain the intention description information.

[0063] The fourth sub-step is to adjust the above problem intention information according to the above intention description information to generate adjusted problem intention information. Among them, the adjusted problem intention information can be a problem intention with higher accuracy and richness than the intention content corresponding to the problem intention information. For example, the above execution entity can fuse the content of the intention description information and the problem intention information to generate the adjusted problem intention information.

[0064] The fifth sub-step, in response to determining that the confidence level corresponding to the above-mentioned adjusted problem intention information is greater than the preset confidence level, according to the above-mentioned problem information, the above-mentioned problem intention information, and the above-mentioned problem association information, and using the above-mentioned first question-and-answer network model group, generate problem prompt information corresponding to the above-mentioned problem information

[0065] The third step is to generate problem response information corresponding to the above-mentioned problem prompt information through the second question-and-answer network model group included in the intelligent question-and-answer large model. Among them, each second question-and-answer network model in the second question-and-answer network model group can be a large language model with different model structures and different types of problems that the model is good at. For example, first, randomly select a second question-and-answer network model from the second question-and-answer network model group. Then, input the above-mentioned problem prompt information into the above-mentioned second question-and-answer network model to obtain problem response information. The second initial question-and-answer network model can be a large language model with different trained model structures and different types of problems that the model is good at

[0066] Thus, by using the first question-and-answer network model group, to generate supplementary problem-related information (i.e., problem association information) for the problem information and generate high-quality problem prompt information, thereby accurately identifying the problem semantics corresponding to the problem information to accurately generate problem response information

[0067] Optionally, in response to determining that there is no corresponding problem response document for the above-mentioned first alternative problem segment sequence, generate problem response information corresponding to the above-mentioned problem information according to the pre-trained intelligent question-and-answer large model, and send the above-mentioned problem response information to the user terminal corresponding to the above-mentioned target user

[0068] In some embodiments, the above-mentioned execution subject may, in response to determining that there is no corresponding problem response document for the above-mentioned first alternative problem segment sequence, generate problem response information corresponding to the above-mentioned problem information according to the pre-trained intelligent question-and-answer large model, and send the above-mentioned problem response information to the user terminal corresponding to the above-mentioned target user

[0069] In practice, the above-mentioned execution subject may generate problem response information corresponding to the above-mentioned problem information through the following steps

[0070] The first step is to perform reconstruction processing on the above-mentioned problem information to obtain reconstructed problem information. For example, a preset prompt template can be concatenated and combined with the above-mentioned problem information to obtain reconstructed problem information. The prompt template can be a preset text template used to guide the model to generate the output of its problem answer

[0071] In the second step, perform enhancement processing on the above-mentioned reconstructed problem information in the corresponding target field to obtain enhanced problem information. For example, the above-mentioned execution entity can search for text information matching the above-mentioned reconstructed problem information from the intelligent knowledge base. Here, the matching can be that the text similarity is greater than or equal to a preset threshold. Then, each piece of text information found can be concatenated with the above-mentioned reconstructed problem information to obtain enhanced problem information. The intelligent knowledge base can also store question-and-answer text pairs. The question-and-answer text pairs can include question texts and answer texts. The above-mentioned text information can include question-and-answer text pairs or answer texts.

[0072] In the third step, input the above-mentioned enhanced problem information into the intelligent question-answering large model to obtain question reply information corresponding to the above-mentioned problem information.

[0073] Among them, performing reconstruction processing on the above-mentioned problem information to obtain reconstructed problem information includes:

[0074] 1. Determine the problem type corresponding to the above-mentioned problem information.

[0075] 2. Generate reconstructed problem information according to the pre-stored hint template corresponding to the above-mentioned problem type. For example, the problem information can be concatenated with the hint template to obtain reconstructed problem information.

[0076] Among them, performing enhancement processing on the above-mentioned reconstructed problem information in the corresponding target field to obtain enhanced problem information includes:

[0077] 1. Input the above-mentioned reconstructed problem information into a pre-trained question retrieval enhancement model to obtain a question retrieval result. Among them, the above-mentioned question retrieval enhancement model corresponds to the knowledge base of the above-mentioned target field. The above-mentioned question retrieval enhancement model can be a question retrieval enhancement model that takes question information as input and generates a question retrieval result as output. The question retrieval enhancement model can be a hybrid model that combines retrieval and generation. The retrieval scope of the question retrieval enhancement model can be the above-mentioned intelligent knowledge base. The question retrieval result generated by the question retrieval enhancement model can include: the answer text corresponding to the question information, the background knowledge text corresponding to the question information. For example, the question retrieval enhancement model can be a Retrieval-Augmented Generation (RAG) model. Retrieval-Augmented Generation (RAG) addresses these limitations by integrating a retrieval mechanism, allowing the LLM to dynamically access and integrate external data sources. RAG improves the accuracy, relevance, and timeliness of generated responses, making the LLM more powerful and applicable to a wider range of application scenarios.

[0078] 2. Perform concatenation processing on the above-mentioned question retrieval result and the above-mentioned reconstructed problem information to obtain enhanced problem information. For example, the above-mentioned execution entity can perform concatenation processing on the above-mentioned question retrieval result and the above-mentioned reconstructed problem information to obtain enhanced problem information.

[0079] In this way, the adaptability of the large model adjusted based on the target domain in the target domain can be improved, and the matching between the generated answers and the user's questions can be improved.

[0080] Optionally, the intelligent question answering model can be trained by the following steps:

[0081] The first step is to obtain the user question dataset corresponding to the target domain. The target domain can represent securities, e-commerce, insurance and other fields.

[0082] In the second step, according to the above-mentioned user question data set, the first initial intelligent question and answer large model for question completion is trained to obtain a second initial intelligent question and answer large model. The first initial intelligent question and answer large model may be a large language model (LLM) that has not yet been trained. The second initial intelligent question and answer large model may be a large language model that has been trained on the user question data set and has mastered more domain knowledge in the target field. The first initial intelligent question and answer large model may include: a first initial question and answer network model group and a second initial question and answer network model group. Among them, the first initial question and answer network model may be an untrained large language model. For example, the first initial question and answer network model may be a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), or a generative adversarial network (GAN), which is used to generate question prompt information corresponding to the question information. The second initial question and answer network model may be a large language model with an untrained model structure and a model that is good at different types of questions. For example, the second initial question and answer network model may be a recurrent neural network (RNN), a long short-term memory network (LSTM), a gated recurrent unit (GRU), or a generative adversarial network (GAN), which is used to answer questions.

[0083] For example, the above-mentioned execution entity can use a model training method (for example, a back propagation method) to perform model training on the first initial intelligent question and answer model for question completion based on the user question data set for the target field, and obtain a second initial intelligent question and answer model.

[0084] The third step is to train the second initial intelligent question-answering model according to the preset question enhancement mark sample set to obtain the third initial intelligent question-answering model. The question enhancement mark samples include: each question completion feature information corresponding to the user question. The question completion feature information may be the feature content corresponding to the question completion feature. The question completion feature may be a feature related to the retrieval question completion operation. For example, the question completion feature may be: question time, question prefix, question completion list corresponding to the question prefix, and question completion association list corresponding to the question prefix. The question enhancement mark samples may be samples in the form of prompt information for the model to enhance its question completion capability.

[0085] In the fourth step, in response to determining that the model training effect corresponding to the third initial intelligent Q&A large model reaches the target training effect, the third initial intelligent Q&A large model is determined as the trained intelligent Q&A large model. The target training effect can be the expected value preset to represent the automatic completion accuracy and completion efficiency of the third initial intelligent Q&A large model. For example, the target training effect can be that the automatic completion accuracy of the third initial intelligent Q&A large model is higher than 70%. The model training effect can be the result of testing the model prediction effect of the third initial intelligent Q&A large model.

[0086] Thus, through multiple model trainings, the model can learn more knowledge corpora in the target domain, combine more question completion features to predict more accurate question answers, and through multiple model tests and adaptive trainings, conduct multiple sample learnings on the model, and an intelligent Q&A large model with more accurate output intelligent Q&A can be obtained.

[0087] Here, the reason for using two vector databases is to increase fault tolerance. Their functions are similar, and both can "find the most matching document according to the score given by aaa -> use bbb to find N most matching segments in this document". Select the milvus vector database to find the matching document, and then use the electric search vector database to find the matching segments in the document, which can optimize the question information to ensure the accuracy of intelligent Q&A.

[0088] Optionally, in response to determining that the question types corresponding to the first alternative question segment sequence and the second alternative question segment sequence are both preset question types, generate question reply information corresponding to the above question information.

[0089] In some embodiments, the above execution subject can generate question reply information corresponding to the above question information in response to determining that the question types corresponding to the first alternative question segment sequence and the second alternative question segment sequence are both preset question types. For example, the same question reply document corresponding to the first alternative question segment sequence and the second alternative question segment sequence can be used as the question reply information corresponding to the above question information. For another example, the question reply document that simultaneously contains the first alternative question segment and the second alternative question segment and has the most segment data can be retrieved from the intelligent knowledge base as the question reply information.

[0090] Such as Figure 4The example shows an intelligent reply page of the intelligent customer service assistant for the question information input by the target user. For example, the question information can be "What is the time for the upgrade plan of the serious illness butler service?"; the question reply information can be "Hello, based on the information you provided, I have sorted out the following for you: 1. Notice of the upgrade plan of the marketing channel big data serious illness butler service; 2. Specific content of the upgrade plan of the serious illness butler service, including: service time, product scope, service standards and content..."; In addition, the intelligent reply page can also include "Push Answer" control, and answer feedback controls of "Useful" and "Useless".

[0091] Further reference Figure 2 As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a question and answer information processing device based on an intelligent knowledge base. These embodiments of the question and answer information processing device based on the intelligent knowledge base correspond to Figure 1 those method embodiments shown, and the question and answer information processing device based on the intelligent knowledge base can be specifically applied to various electronic devices.

[0092] As Figure 2 shown, some embodiments of the question and answer information processing device 200 based on the intelligent knowledge base include: a query unit 201, a first determination unit 202, a second determination unit 203, and a generation unit 204. Among them, the query unit 201 is configured to, in response to receiving the question information sent by the target user, query from the pre-constructed intelligent knowledge base the first alternative question fragment sequence and the second alternative question fragment sequence corresponding to the above question information. Among them, the above intelligent knowledge base includes: a native vector database and a distributed search database. The first alternative question fragment sequence corresponds to the native vector database, and the second alternative question fragment sequence corresponds to the distributed search database; the first determination unit 202 is configured to determine whether the question types corresponding to the first alternative question fragment sequence and the second alternative question fragment sequence are preset question types; the second determination unit 203 is configured to, in response to determining that the question types corresponding to the first alternative question fragment sequence and the second alternative question fragment sequence are not preset question types, determine whether there is a question reply document corresponding to the first alternative question fragment sequence that satisfies the target matching condition; the generation unit 204 is configured to, in response to determining that there is no corresponding question reply document for the first alternative question fragment sequence, generate the question reply information corresponding to the above question information according to the pre-trained intelligent question and answer large model, and send the above question reply information to the user terminal corresponding to the above target user.

[0093] It can be understood that the various units recorded in the question and answer information processing device 200 based on the intelligent knowledge base and reference Figure 1corresponds to each step in the described method. Thus, the operations, features, and beneficial effects described above for the method also apply to the question and answer information processing apparatus 200 based on the intelligent knowledge base and the units included therein, and will not be elaborated herein.

[0094] Reference is now made to Figure 3 , which shows a schematic structural diagram of an electronic device 300 (e.g., a computing device) suitable for use in implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is merely an example and should not impose any limitation on the functions and usage scope of the embodiments of the present disclosure.

[0095] As Figure 3 shown, the electronic device 300 may include a processing device 301 (e.g., a central processing unit, a graphics processing unit, etc.), which may perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0096] Generally, the following devices may be connected to the I / O interface 305: an input device 306 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 308 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 309. The communication device 309 may allow the electronic device 300 to communicate with other devices wirelessly or wirelesly to exchange data. Although Figure 3 shows the electronic device 300 having various devices, it should be understood that it is not required to implement or include all the shown devices. Instead, more or fewer devices may be implemented or included. Figure 3 Each block shown in

[0097] In particular, according to some embodiments of the present disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of the present disclosure include a computer program product that includes a computer program carried on a computer-readable medium, and the computer program includes program code for performing the methods shown in the flowcharts. In such some embodiments, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above functions defined in the methods of some embodiments of the present disclosure are performed.

[0098] It should be noted that the computer-readable medium described in some embodiments of the present disclosure can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program, and the program can be used by or in combination with an instruction execution system, apparatus, or device. In some embodiments of the present disclosure, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, and the computer-readable signal medium can send, propagate, or transmit a program for use by or in combination with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted by any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0099] In some embodiments, the client and the server can communicate using any currently known or future-developed network protocol such as HTTP (HyperText Transfer Protocol), and can be interconnected with digital data communication in any form or medium (e.g., a communication network). Examples of communication networks include local area networks ("LANs"), wide area networks ("WANs"), the Internet (e.g., the Internet), and end-to-end networks (e.g., ad hoc end-to-end networks), as well as any currently known or future-developed networks.

[0100] The above computer-readable medium can be included in the above electronic device; or can exist separately without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by the electronic device, the electronic device is caused to: in response to receiving question information sent by a target user, query from a pre-constructed intelligent knowledge base a first alternative question fragment sequence and a second alternative question fragment sequence corresponding to the above question information, wherein the above intelligent knowledge base includes: a native vector database and a distributed search database, the first alternative question fragment sequence corresponds to the native vector database, and the second alternative question fragment sequence corresponds to the distributed search database; determine whether the question types corresponding to the above first alternative question fragment sequence and the second alternative question fragment sequence are preset question types; in response to determining that neither the question types corresponding to the first alternative question fragment sequence nor the second alternative question fragment sequence are preset question types, determine whether there is a question response document corresponding to the above first alternative question fragment sequence that meets the target matching condition; in response to determining that there is no corresponding question response document for the above first alternative question fragment sequence, generate question response information corresponding to the above question information according to a pre-trained intelligent question-answering large model, and send the above question response information to the user terminal corresponding to the above target user.

[0101] Computer program code for performing the operations of some embodiments of the present disclosure can be written in one or more programming languages or combinations thereof. The above programming languages include object-oriented programming languages such as Java, Smalltalk, C++, and also include conventional procedural programming languages such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, executed as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer can be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computer (e.g., by using an Internet service provider to connect through the Internet).

[0102] The flowcharts and block diagrams in the accompanying drawings illustrate the possible architectures, functions, and operations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagram may represent a module, a segment of a program, or a portion of code that contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions noted in the blocks may occur in a different order than noted in the accompanying drawings. For example, two consecutive blocks shown may actually be executed substantially in parallel, or they may sometimes be executed in the reverse order, depending on the functions involved. It should also be noted that each block in the block diagram and / or flowchart, and combinations of blocks in the block diagram and / or flowchart, can be implemented by a dedicated hardware-based system that performs the specified functions or operations, or can be implemented by a combination of dedicated hardware and computer instructions.

[0103] The units described in some embodiments of the present disclosure can be implemented in software or in hardware. The described units can also be provided in a processor. For example, it can be described as: a processor includes: a query unit, a first determination unit, a second determination unit, and a generation unit. Among them, the names of these units do not constitute a limitation on the unit itself in some cases. For example, the generation unit can also be described as "a unit that, in response to determining that there is no corresponding question answer document for the above-mentioned first alternative question segment sequence, generates question answer information corresponding to the above-mentioned question information according to a pre-trained intelligent Q&A large model, and sends the above-mentioned question answer information to the user terminal corresponding to the above-mentioned target user".

[0104] The functions described above herein can be performed, at least in part, by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that can be used include: Field Programmable Gate Arrays (FPGAs), Application Specific Integrated Circuits (ASICs), Application Specific Standard Products (ASSPs), Systems on Chip (SOCs), Complex Programmable Logic Devices (CPLDs), and the like.

[0105] Some embodiments of the present disclosure also provide a computer program product, including a computer program that, when executed by a processor, implements any one of the above-mentioned question and answer information processing methods based on an intelligent knowledge base.

[0106] The above description is only some preferred embodiments of the present disclosure and an explanation of the applied technical principles. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by the specific combination of the above technical features, but should also cover other technical solutions formed by any combination of the above technical features or their equivalent features without departing from the above inventive concept. For example, the technical solutions formed by mutually replacing the above features with the technical features (but not limited to) disclosed in the embodiments of the present disclosure that have similar functions.

Claims

1. A question-answer information processing method based on an intelligent knowledge base, applied to an intelligent customer service assistant, comprising: In response to receiving the initial question information sent by the target user, identifying each question keyword in the initial question information as question information, and querying a first candidate question fragment sequence and a second candidate question fragment sequence corresponding to the question information from a pre-built intelligent knowledge base, wherein the intelligent knowledge base includes: a native vector database and a distributed search database, the first candidate question fragment sequence corresponds to the native vector database, and the second candidate question fragment sequence corresponds to the distributed search database; Determining whether the question type corresponding to the second candidate question segment sequence is a preset question type; In response to determining that none of the question types corresponding to the second candidate question segment sequence are preset question types, determining whether there is a question answer document corresponding to the first candidate question segment sequence that meets a target matching condition; In response to determining that there is a corresponding question answer document for the first candidate question segment sequence, retrieving a target number of third candidate question segments in the distributed search database according to the question answer document and the question information; The target number of third alternative question fragments are spliced ​​to obtain spliced ​​question fragment information, and the spliced ​​question fragment information and the question information are input into a pre-trained intelligent question and answer model to generate question answer information corresponding to the question information, and the question answer information is sent to the user terminal corresponding to the target user.

2. The method according to claim 1, wherein: The step of retrieving a target number of third candidate question segments from the distributed search database according to the question answer document and the question information includes: Retrieving a target number of third initial question segments from the question answer document through the distributed search database; For each third initial question segment of the target number of third initial question segments, a context segment of the third initial question segment is merged with the third initial question segment to obtain a third candidate question segment.

3. The method according to claim 1, wherein: The method further comprises: In response to determining that there is no corresponding question answer document for the first candidate question fragment sequence, question answer information corresponding to the question information is generated according to a pre-trained intelligent question and answer model, and the question answer information is sent to a user terminal corresponding to the target user.

4. The method according to claim 1, wherein: The method further comprises: In response to determining that the question types corresponding to the first candidate question segment sequence and the second candidate question segment sequence are both preset question types, question answer information corresponding to the question information is generated.

5. The method according to claim 1, wherein: Before generating the question answer information corresponding to the question information according to the pre-trained intelligent question answering model, the method further includes: Obtain a user question dataset corresponding to the target domain; According to the user question data set, model training is performed on the first initial intelligent question-answering large model for question completion to obtain a second initial intelligent question-answering large model; According to a preset question enhancement mark sample set, the second initial intelligent question answering large model is trained to obtain a third initial intelligent question answering large model, wherein the question enhancement mark sample includes: each question completion feature information corresponding to the user question; In response to determining that the model training effect corresponding to the third initial intelligent question and answer big model reaches the target training effect, the third initial intelligent question and answer big model is determined as the trained intelligent question and answer big model.

6. The method according to claim 5, wherein: The step of generating question answer information corresponding to the question information according to the pre-trained intelligent question answering model includes: Reconstructing the problem information to obtain reconstructed problem information; Performing enhancement processing on the reconstructed problem information corresponding to the target field to obtain enhanced problem information; Inputting the enhanced question information into the intelligent question-answering model to obtain question answer information corresponding to the question information; The problem information is reconstructed to obtain the reconstructed problem information, including: Determine the type of question corresponding to the question information; Generate reconstruction question information according to a pre-stored prompt template corresponding to the question type; The reconstruction problem information is enhanced in the corresponding target field to obtain enhanced problem information, including: Inputting the reconstructed question information into a pre-trained question retrieval enhancement model to obtain a question retrieval result, wherein the question retrieval enhancement model corresponds to a knowledge base of the target domain; The question retrieval result and the reconstructed question information are spliced ​​to obtain enhanced question information.

7. A question-answer information processing device based on an intelligent knowledge base, applied to an intelligent customer service assistant, comprising: A query unit is configured to, in response to receiving the initial question information sent by the target user, identify each question keyword in the initial question information as question information, and query a first candidate question fragment sequence and a second candidate question fragment sequence corresponding to the question information from a pre-built intelligent knowledge base, wherein the intelligent knowledge base includes: a native vector database and a distributed search database, the first candidate question fragment sequence corresponds to the native vector database, and the second candidate question fragment sequence corresponds to the distributed search database; A first determining unit is configured to determine whether the question type corresponding to the second candidate question segment sequence is a preset question type; A second determination unit is configured to, in response to determining that none of the question types corresponding to the second candidate question segment sequence are preset question types, determine whether there is a corresponding question answer document that meets the target matching condition for the first candidate question segment sequence; a retrieval unit configured to retrieve a target number of third candidate question segments in the distributed search database according to the question answer document and the question information in response to determining that the first candidate question segment sequence has a corresponding question answer document; The input unit is configured to splice the target number of third alternative question fragments to obtain spliced ​​question fragment information, and input the spliced ​​question fragment information and the question information into a pre-trained intelligent question and answer model to generate question answer information corresponding to the question information, and send the question answer information to the user terminal corresponding to the target user.

8. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer readable medium having a computer program stored thereon, wherein: When the computer program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images