Question and answering method and apparatus, computer device, and readable storage medium

Abstract

The present application relates to the field of artificial intelligence, and in particular to a question and answering method, comprising: inputting initial question text of a user into a quantized language model to obtain a slot parameter corresponding to the initial question text, wherein the slot parameter comprises description information related to a target question and answer scenario, and the quantized language model is a language model obtained by compressing a model parameter of a large language model; inputting the slot parameter and the initial question text into a quantized language model to obtain standardized question text, wherein the standardized question text is text that conforms to a predefined text structure and clearly reflects a user intention; inputting the standardized question text into a vector model to obtain a plurality of documents having a highest degree of correlation with the standardized question text; and inputting the standardized question text and the plurality of documents into the quantized language model to obtain response text generated for answering the initial question text. In addition, the present application also relates to the field of natural language processing.

Description

Question-and-answer methods, apparatus, computer equipment, and readable storage media

[0001] Related applications

[0002] This application claims priority to Chinese patent application No. 2024117211465, filed on November 28, 2024, entitled "Question-answering method, apparatus, computer equipment and readable storage medium", the entire contents of which are incorporated herein by reference. Technical Field

[0003] This application relates to the field of natural language processing technology, and in particular to a question-answering method, apparatus, computer device, computer-readable storage medium, and computer program product. Background Technology

[0004] A knowledge-based question-answering system is a system that directly generates accurate answers by mapping user questions to a structured knowledge base. Natural Language Processing (NLP) is the technology that studies how computers understand, generate, and process human language. It is the core technology of a knowledge-based question-answering system, providing the system with capabilities such as semantic parsing, intent recognition, and text generation.

[0005] However, the inventors realized that traditional question-answering systems typically use large language models (LLMs) to reason and answer user questions. Directly deploying these large language models requires a lot of computing resources and storage space, resulting in slow reasoning speed. Summary of the Invention

[0006] According to various embodiments disclosed in this application, a question-answering method, apparatus, computer device, computer-readable storage medium, and computer program product are provided that can improve the reasoning speed of a question-answering system.

[0007] A question-and-answer method, including:

[0008] The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model;

[0009] The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0010] The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and

[0011] The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

[0012] A question-and-answer device, comprising:

[0013] The parameter acquisition module is used to input the user's initial question text into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model;

[0014] The standardization module is used to input the slot parameters and the initial question text into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0015] The document acquisition module is used to input the standardized question text into a vector model to obtain several documents that are most relevant to the standardized question text; and

[0016] The answer module is used to input the standardized question text and the plurality of documents into the quantized language model to obtain the answer text generated in response to the initial question text.

[0017] A computer device includes a memory and one or more processors, the memory storing computer-readable instructions that, when executed by the processors, cause the one or more processors to perform the following steps:

[0018] The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model;

[0019] The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0020] The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and

[0021] The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text. One or more computer-readable storage media storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the following steps: inputting the user's initial question text into the quantized language model to obtain slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model;

[0022] The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0023] The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and

[0024] The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

[0025] A computer program product includes a computer program that, when executed by one or more processors, causes the one or more processors to perform the following steps:

[0026] The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model;

[0027] The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0028] The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and

[0029] The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

[0030] The aforementioned question-answering method, apparatus, computer equipment, computer-readable storage medium, and computer program product input the user's initial question text into a quantized language model to obtain slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model; the slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent; the standardized question text is input into a vector model to obtain several documents with the highest relevance to the standardized question text; the standardized question text and several documents are input into the quantized language model to obtain the answer text generated by answering the initial question text. By compressing the model parameters of a large language model to obtain a quantized language model, the model parameters can be compressed into a representation with fewer bits, significantly reducing the requirements for computing resources and storage space. This allows the question-answering system to maintain answer accuracy while greatly improving inference speed. Furthermore, by extracting slot parameters, user questions are associated with the target question-answering scenario, enabling efficient and accurate identification of user intent. At the same time, the generation of standardized question text allows user questions to be efficiently matched with the semantic retrieval logic in the vector model, thereby finding the document with the highest semantic relevance to the user question. The quantized language model is then used to further analyze the relevant documents and generate answers. The entire process, through modular design and model parameter compression technology, improves the efficiency, accuracy, and inference speed of the question-answering system.

[0031] Details of one or more embodiments of this application are set forth in the following drawings and description. Other features and advantages of this application will become apparent from the specification, drawings, and claims. Attached Figure Description

[0032] To more clearly illustrate the technical solutions in the embodiments of this application or the conventional technology, the drawings used in the description of the embodiments or the conventional technology will be briefly introduced below. Obviously, the drawings described below are only embodiments of this application. For those skilled in the art, other drawings can be obtained based on the disclosed drawings without creative effort.

[0033] Figure 1 is an application environment diagram of a question-answering method according to one or more embodiments;

[0034] Figure 2 is a flowchart illustrating a question-and-answer method according to one or more embodiments;

[0035] Figure 3 is a logic diagram of an intent recognition mechanism according to one or more embodiments;

[0036] Figure 4 is a logic diagram of a text retrieval method in another embodiment;

[0037] Figure 5 is a flowchart illustrating a question-and-answer method in another embodiment;

[0038] Figure 6 is a structural block diagram of a question-and-answer device according to one or more embodiments;

[0039] Figure 7 is an internal structural diagram of a computer device according to one or more embodiments. Detailed Implementation

[0040] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0041] The question-and-answer method provided in this application embodiment can be applied to the application environment shown in Figure 1. The terminal 102 communicates with the server 104 via a network. A data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104 or placed on a cloud or other network server. The terminal 102 inputs the user's initial question text into a quantized language model to obtain slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-and-answer scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model; the terminal 102 inputs the slot parameters and the initial question text into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent; the terminal 102 inputs the standardized question text into a vector model to obtain several documents with the highest relevance to the standardized question text; the terminal 102 inputs the standardized question text and several documents into the quantized language model to obtain the answer text generated in response to the initial question text. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle systems, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0042] In one embodiment, as shown in FIG2, a question-answering method is provided. Taking the application of this method to terminal 102 in FIG1 as an example, the method includes:

[0043] Step S202: Input the user's initial question text into the quantized language model to obtain the slot parameters corresponding to the initial question text.

[0044] Among them, the target question-and-answer scenario can refer to a specific question-and-answer scenario, such as a government question-and-answer scenario, a hospital question-and-answer scenario, a police question-and-answer scenario, etc.

[0045] The slot parameters include descriptive information related to the target question-and-answer scenario. For example, taking the target question-and-answer scenario as a government affairs question-and-answer scenario, based on the question-and-answer characteristics of the government affairs scenario, the slot parameters can include scenario slot parameters, business type slot parameters, and data variable slot parameters, which are based on the scenario, business type, and data variables respectively to achieve accurate parsing and classification of user queries.

[0046] The scenario slot parameter can include a preliminary macro-classification of user questions, such as dividing user consultation needs into different government service areas such as social security cards, pension insurance, housing provident funds, or marriage registration.

[0047] The business type slot parameter can include further detailed consultation content, such as distinguishing whether the user is inquiring about specific business such as provident fund withdrawal, provident fund deposit or provident fund loan in the context of provident fund.

[0048] The data variable slot parameter can include specific information points that the user inquires about, which may include details such as the business process, the time required, the required materials, the processing location, and relevant contact numbers.

[0049] The user's initial question text can be a question posed in response to the target question-and-answer scenario. For example, if the target question-and-answer scenario is a government affairs scenario, the user's initial question text could include "I would like to inquire about how to withdraw my housing provident fund".

[0050] Specifically, a large language model (MLM) is a deep learning-based language model trained on massive datasets. Through unsupervised or semi-supervised training on large corpora, it can capture the semantics, syntax, and contextual relationships of a language. MLM can achieve a wide range of natural language processing (NLP) tasks, such as question-answering, text summarization, translation, and code generation, through a pre-training-fine-tuning approach. Further, taking question-answering as an example, MLM can execute different sub-tasks based on different prompts. These sub-tasks might include generating slot parameters for the initial question text, converting the initial question text into standardized question text, and generating answer text for the initial question text. Prompts can guide the MLM to perform specific tasks or output specific content. Prompts can be a simple question, a task description, a piece of context, or any information that helps the model understand the required task. Optionally, the prompts can be designed by developers and predefined based on the task objectives; they can also be based on fixed templates and dynamically generated by filling in user input or contextual data into the templates; or they can be dynamically generated by language models based on specific tasks, etc.

[0051] The quantized language model is obtained by compressing the model parameters of the Large Language Model (LLM). Optionally, the model parameters of the large language model can be compressed into a representation with fewer bits based on the Greedy Coordinate Descent (GCD) quantization technique to significantly improve the inference speed of the model.

[0052] The terminal can parse the user's question text through a quantized language model. The quantized language model can then combine predefined scenarios and semantic analysis techniques to extract slot parameters related to the target question-and-answer scenario. These slot parameters can more concisely and efficiently reflect the user's core intent.

[0053] Step S204: Input the slot parameters and the initial question text into the quantized language model to obtain the standardized question text.

[0054] The standardized question text is text that conforms to the preset text structure and clearly reflects the user's intent.

[0055] The terminal uses a quantized language model, combined with extracted slot parameters and the initial question text, to intelligently rewrite the query, generating a structured and clearly defined user question statement. This eliminates ambiguity in the user input, ensuring the accuracy and completeness of the question statement while maintaining consistency with the user's original intent. The final output standardized question text is a standardized question statement that clearly expresses the user's query intent and is suitable for subsequent document retrieval and information provision. For example, the initial question text "I want to withdraw my housing provident fund" can be rewritten as the standardized question text "The user applies for housing provident fund withdrawal, specifically in November 2024".

[0056] Step S206: Input the standardized question text into the vector model to obtain several documents that are most relevant to the standardized question text.

[0057] Vector models can be used to transform text (such as user questions and document content) into semantic vectors, and then use the semantic similarity (such as cosine similarity) of these vectors to retrieve documents related to user questions.

[0058] Optionally, the vector model can be a lightweight embedding model that can quickly retrieve the documents most relevant to the user's question from a local knowledge base or online data. For example, the vector model vectorizes the standardized question text to obtain a semantic vector, and calculates its similarity with document vectors in a local or online knowledge base, thereby retrieving several documents from the knowledge base that are most relevant to the standardized question text.

[0059] Step S208: Input the standardized question text and several documents into the quantized language model to obtain the response text generated in response to the initial question text.

[0060] In practice, the terminal constructs appropriate prompts based on a standardized question text and several documents. These prompts then guide the quantized language model to answer the question using the standardized question text and the documents, generating an answer text. For example, if the standardized question text is "What materials are needed to reissue an ID card?", and the documents are 5 local knowledge base documents and 5 online documents, the prompts could be: "Answer the user's question 'What materials are needed to reissue an ID card?' based on the following document content, ensuring the answer is accurate and complete." These prompts clarify the answer task the quantized language model needs to complete and limit the context of the generated answer.

[0061] The quantized language model can be a large language model with reduced parameters obtained through quantization techniques such as Greedy Coordinate Descent (GCD). This reduces the number of model parameters while retaining high semantic understanding and response capabilities. It can efficiently process prompt words and comprehensively analyze the content of standardized question text and related documents. For example, the response text could be, "To reissue an ID card, you need an ID photo, household registration booklet, and application form. You can bring these materials to your local public security bureau to apply."

[0062] In the above question-answering method, the user's initial question text is input into a quantized language model to obtain slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model; the slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent; the standardized question text is input into a vector model to obtain several documents with the highest relevance to the standardized question text; the standardized question text and several documents are input into the quantized language model to obtain the answer text generated by answering the initial question text. By compressing the model parameters of a large language model to obtain a quantized language model, the model parameters can be compressed into a representation with fewer bits, significantly reducing the requirements for computing resources and storage space. This allows the question-answering system to maintain answer accuracy while greatly improving inference speed. Furthermore, by extracting slot parameters, user questions are associated with the target question-answering scenario, enabling efficient and accurate identification of user intent. At the same time, the generation of standardized question text allows user questions to be efficiently matched with the semantic retrieval logic in the vector model, thereby finding the document with the highest semantic relevance to the user question. The quantized language model is then used to further analyze the relevant documents and generate answers. The entire process, through modular design and model parameter compression technology, improves the efficiency, accuracy, and inference speed of the question-answering system.

[0063] In one embodiment, the user's initial question text is input into a quantized language model to obtain the slot parameters corresponding to the initial question text. This includes: inputting the user's initial question text into the quantized language model; identifying the scenario slot parameters corresponding to the initial question text through the quantized language model; determining the current target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters; identifying the business type slot parameters and data variable slot parameters corresponding to the initial question text through the quantized language model based on the context information of the current target question-and-answer scenario and the initial question text; and determining the slot parameters corresponding to the initial question text based on the scenario slot parameters, business type slot parameters, and data variable slot parameters.

[0064] Taking the target question-and-answer scenario as a government affairs question-and-answer scenario as an example, the scenario slot parameters include a preliminary macro-classification of user questions, such as dividing user consultation needs into different government service areas like social security cards, pension insurance, housing provident funds, or marriage registration. Therefore, each target question-and-answer scenario can include subdivided government affairs question-and-answer scenarios such as social security cards, pension insurance, housing provident funds, and marriage registration.

[0065] Contextual information can be relevant background information and environmental factors that affect the understanding of objects or events in a specific scenario. The contextual information of the current target question-answering scenario is used to help the quantized language model understand the current target question-answering scenario.

[0066] For example, if a user's initial question text is "How do I withdraw my housing provident fund", the quantized language model identifies the corresponding scenario slot parameter as "housing provident fund" and determines the current target question-and-answer scenario as the housing provident fund scenario. The context information of the housing provident fund scenario is used to indicate that the current target application scenario is the housing provident fund scenario, and background knowledge and information related to the housing provident fund scenario are provided.

[0067] For example, the quantized language model can combine the contextual information of the current target question-and-answer scenario with the initial question text to identify the business type slot parameters and data variable slot parameters corresponding to the initial question text. For instance, if the user's initial question text is "I want to withdraw my housing provident fund tomorrow", the scenario slot parameter is "housing provident fund", the business type slot parameter is "withdrawal", and the data variable slot parameters are "housing provident fund processing time", "housing provident fund withdrawal process", "materials required for housing provident fund withdrawal", etc.

[0068] As an example, the terminal determines the slot parameters corresponding to the initial question text based on the scenario slot parameters, business type slot parameters, and data variable slot parameters. This can include: when the terminal detects that all slot parameters in the scenario slot parameters, business type slot parameters, and data variable slot parameters are complete, the terminal uses the scenario slot parameters, business type slot parameters, and data variable slot parameters together as the slot parameters corresponding to the initial question text.

[0069] The technical solution in this embodiment uses a quantized language model to analyze user questions step by step, extracting slot parameters such as scenarios, business types, and data variables. The generated slot parameters ensure the accurate capture and description of user intent, improving the efficiency and accuracy of the subsequent question-and-answer system.

[0070] In one embodiment, determining the slot parameters corresponding to the initial question text based on the scenario slot parameters, business type slot parameters, and data variable slot parameters includes: if at least one of the scenario slot parameters, business type slot parameters, and data variable slot parameters is detected to be incomplete, outputting guidance information through a quantized language model; the guidance information is used to guide the user to supplement the missing slot parameters; if the missing slot parameters are obtained, determining the slot parameters corresponding to the initial question text through the quantized language model based on the missing slot parameters, scenario slot parameters, business type slot parameters, and data variable slot parameters.

[0071] For example, incomplete slot parameters can mean that at least one of the scenario slot parameters, business type slot parameters, and data variable slot parameters is completely missing, or it can mean that the details of at least one of the scenario slot parameters, business type slot parameters, and data variable slot parameters are missing. For instance, if the initial question text is "I am inquiring about housing provident fund matters," the terminal may consider that the initial question text is missing the business type slot parameter. Therefore, the guidance information output by the quantized language model could be "Please provide which housing provident fund business you need to handle." The user can supplement the missing business type slot parameter, such as "I want to withdraw my housing provident fund." The terminal can then generate complete business type slot parameters based on the missing slot parameters supplemented by the user through the quantized language model, and generate the slot parameters corresponding to the initial question text based on the complete data variable slot parameters, scenario slot parameters, and business type slot parameters.

[0072] The technical solution in this embodiment analyzes user questions step by step using a quantized language model, extracting slot parameters such as scenarios, business types, and data variables to ensure accurate capture and description of user intent. At the same time, it uses an intelligent guidance mechanism to improve incomplete parameters, thereby enhancing the intelligence and efficiency of the question-and-answer system.

[0073] In traditional question-and-answer systems targeting a broad user base, non-specialist users may struggle to provide precise and clear descriptions when asking questions. This makes it difficult for vector-based models to accurately interpret user query intent, thus affecting the accuracy and relevance of retrieved documents. To address this issue, this embodiment introduces an interactive intent recognition mechanism. This mechanism employs an iterative guidance strategy, using a series of carefully designed interactive steps to help users gradually clarify and refine their query intent. This results in more accurate document retrieval and high-quality question-and-answer services, enhancing the intelligence of the question-and-answer system and the user experience.

[0074] In one embodiment, determining the current target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters includes: determining the actual target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters using a quantized language model, and updating the actual target question-and-answer scenario to the current target question-and-answer scenario if the actual target question-and-answer scenario does not match the current target question-and-answer scenario in which the user is located.

[0075] In practice, the terminal uses a quantized language model to determine the actual target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters extracted from the initial question text. For example, if the initial question text is "I want to check my housing provident fund balance," the extracted scenario slot parameter is "housing provident fund," and the determined actual target question-and-answer scenario is the housing provident fund scenario. The terminal compares the determined actual target question-and-answer scenario with the user's current target question-and-answer scenario. If they match, the user's question continues to be processed in the current scenario; if they do not match, it means the user's question involves a new scenario and needs to be updated. When a mismatch is detected, the system updates the actual target question-and-answer scenario to the current target question-and-answer scenario, allowing for timely adaptation to the user's new question scenario. For example, if the user previously asked a question in the "social security inquiry" scenario and now asks a "housing provident fund inquiry" question, the system updates the current scenario from "social security inquiry" to "housing provident fund inquiry."

[0076] The technical solution in this embodiment can dynamically adjust the question-and-answer scenario, ensure accurate understanding and response to user questions, and improve the flexibility and intelligence of the question-and-answer system.

[0077] To facilitate understanding by those skilled in the art, Figure 3 provides an exemplary logical diagram of an intent recognition mechanism. In its specific implementation, the user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text. This may include the following steps: (1) Scene analysis: The quantized language model analyzes the user's initial question text to obtain the actual target question-and-answer scene, and determines whether the actual target question-and-answer scene matches the user's current target question-and-answer scene; (2) Scene recognition and updating: If the actual target question-and-answer scene does not match the current target question-and-answer scene, the quantized language model updates the actual target question-and-answer scene to the current target question-and-answer scene, and obtains the updated scene slot parameters based on the updated current target question-and-answer scene; (3) Business type and data variable extraction: If the actual target question-and-answer scene matches the current target question-and-answer scene, the quantized language model will combine the current target question-and-answer scene... (3) Context information and initial question text, extract business type slot parameters and data variable slot parameters to further refine the slot parameters; (4) Slot completeness check: evaluate the completeness of the three slot parameters: scenario slot parameters, business type slot parameters and data variable slot parameters; (5) Intent output: if all slot parameters are complete, all slot parameters can be output. These slot parameters can accurately reflect the user's intent for subsequent processing; (6) Slot parameter supplementation guidance: if the slot parameters are incomplete, the quantized language model will guide the user to supplement the missing business type and data variable information until all slot parameters are accurately filled; (7) Final intent confirmation: after the user has supplemented all the necessary slot parameters, confirm again and output the slot parameters that reflect the user's complete intent.

[0078] In one embodiment, the standardized question text is input into a vector model to obtain a number of documents that are most relevant to the standardized question text. This includes: inputting the standardized question text into a vector model, vectorizing the standardized question text using the vector model to obtain vectorized text; retrieving a number of local documents that are most relevant to the vectorized text from a local knowledge base using the vector model, and retrieving a number of online documents that are most relevant to the vectorized text from a website platform.

[0079] In practice, the terminal can input the standardized question text into a vector model, converting it into a high-dimensional vector representation. Using text embedding technology, the semantic information of the text is encoded into numerical vectors, facilitating subsequent similarity calculations. The vector model then calculates the similarity between the vector representation of the standardized question and the vectors of documents already stored in the local knowledge base. Based on the similarity score, it retrieves the local documents with the highest relevance to the standardized question text. Similarly, the vector representation of the standardized question is used to retrieve documents on a web platform. By calculating the similarity between the vector representation of the web documents and the web documents, it obtains the web documents with the highest relevance to the standardized question text.

[0080] To facilitate understanding by those skilled in the art, Figure 4 provides an exemplary logical diagram of text retrieval. Specifically, the terminal can perform local knowledge base retrieval, using a trained vector model to vectorize the standardized question, and then using vector retrieval to obtain the 5 most relevant documents to the user's question in the local knowledge base. Simultaneously, the terminal can perform online searches, firstly pre-setting several commonly used government websites as data sources for the query, then combining search engine and web crawler technologies to extract the most relevant documents from each website, and finally using a re-ranking model to sort the real-time online documents and obtain the 5 most relevant documents.

[0081] In traditional technologies, the construction of vector databases is typically based on a predefined set of documents, resulting in fixed database content and timestamps after construction. This static nature limits the database's responsiveness to real-time data, meaning that when users ask questions related to immediate events, the database cannot provide the latest information, thus affecting the accuracy of query results. This embodiment addresses this problem by proposing a networked search assistance mechanism. By retrieving and evaluating relevant documents online, it ensures that the question-answering system obtains the most relevant input corpus, thereby ensuring that the system can provide timely and accurate information feedback. Through the vector database's real-time reflection of data, it ensures that users receive accurate and timely information feedback.

[0082] In one embodiment, before inputting the user's initial question text into the quantized language model to obtain the slot parameters corresponding to the initial question text, the method further includes: acquiring an input data matrix; the input data matrix contains feature vectors corresponding to training question samples; determining target coordinates and quantized values ​​corresponding to the target coordinates from the original weight matrix of the language model to be quantized; the original weight values ​​corresponding to the target coordinates maximize the reduction of the loss function when updated to quantized values; the loss function is determined based on the difference between the original weight matrix and the quantized weight matrix of the language model to be quantized and the input data matrix; determining the quantized weight matrix based on the target coordinates and the quantized values; and determining the quantized language model based on the quantized weight matrix.

[0083] The input data matrix contains feature vectors corresponding to the training question samples. Optionally, the training question sample text can be vectorized to obtain the corresponding feature vectors. The input data matrix consists of these feature vectors, with each row representing a feature vector of a training question sample and each column corresponding to a feature dimension. Optionally, the training question samples can include various questions that users might input in actual question-and-answer scenarios within the question-and-answer system.

[0084] The target coordinate is a coordinate in the original weight matrix. The original weight value at this coordinate has the greatest impact on the model weight matrix. Quantizing it can significantly reduce the value of the loss function.

[0085] The quantized value is the original weight value corresponding to the target coordinates that is replaced by the quantized value. Optionally, it can be a value from a set of discrete values ​​(such as a low-precision integer or a fixed-point number).

[0086] The terminal can continuously select target coordinates and corresponding quantization values ​​from all coordinates. The original weight value corresponding to the target coordinates will have the largest reduction in loss function after being updated to the corresponding quantization value.

[0087] The loss function measures the difference between the original weight matrix and the quantized weight matrix, and, in conjunction with the input data matrix, reflects the model's performance on the training problem samples. This loss function guides the optimization process, ensuring that the quantized weight matrix is ​​compressed while preserving model performance.

[0088] In practice, the terminal selects new target coordinates and quantization values ​​in each round, updates the corresponding original weight values ​​in the weight matrix, and generates the quantized weight matrix by gradually optimizing the weight matrix. By optimizing the weight matrix one coordinate at a time, model compression is completed while minimizing the loss.

[0089] In traditional technologies, question-answering systems typically use large language models to reason about and answer customer questions. However, directly deploying these large language models requires significant computing resources and storage space, resulting in slow inference speeds and complex deployment processes. To address this issue, a large model quantization technique, such as greedy coordinate descent quantization, can be introduced to compress model parameters into a representation with fewer bits. This significantly reduces computational requirements and storage overhead, improves inference speed, and makes the model easier to deploy and operate on various devices and platforms, thereby enhancing the usability of the question-answering system.

[0090] In one embodiment, quantizing a large model can be achieved by compressing the weight matrix parameters of the large model into a smaller number of bits. Assume d in and d outThese represent the input and output dimensions of any computational layer of the language model to be quantized, respectively. This is the original weight matrix of this computational layer. It is an input data matrix containing n data points.

[0091] During the quantization process, the goal is to represent the original weight matrix W as Q×diag(a)+1b T In the form of, These represent the scaling and bias parameters, respectively. `diag(a)` is the diagonal matrix of `a`, and `b`... T Let b be the transpose of the matrix. This is the quantized weight matrix, where c represents the bit precision used during quantization. Post-training Quantization (PTQ) can address the problem of layer-by-layer optimization of the language model to be quantized: Optionally, the problem can be decomposed into d in the output dimension. out This is an independent problem; therefore, we can focus on any one of the following quantizations, d. in The problem of dimensional vectors: Where w can be a column of the original weight matrix W, i.e., a d in d-dimensional vector; q can be a quantized vector of w, or a d-dimensional vector. in A dimensional vector is a column in the quantized weight matrix. Assuming a and b are fixed, this is an integer linear regression problem. Assuming suitable a and b are already determined, we can focus on optimizing q. During the quantization process, in each round, we can find the coordinates that maximize the reduction of the loss function, i.e., the target coordinates, and decrease along these coordinates.

[0092] Let the loss function We can try to find a coordinate i and a quantization value r such that updating the i-th coordinate to r results in the maximum reduction in the loss function: min i,r L(q+(rq i )e i )-L(q);

[0093] Among them, e i It is a standard basis vector, with 1 at the i-th position and 0 at the other positions.

[0094] Let H = X T X, then we have: L(q+(rq) i )e i )-L(q)=(rq i ) 2 H i,i +(rq i )gi ;

[0095] Where H and g are the Hessian matrix and gradient of L(q) evaluated at q, respectively.

[0096] In one embodiment, as shown in FIG5, a question-and-answer method is provided. Taking the application of this method to terminal 102 in FIG1 as an example, the method includes the following steps:

[0097] S502, the user's initial question text is input into the quantized language model, and the scene slot parameters corresponding to the initial question text are identified through the quantized language model.

[0098] In one embodiment, before inputting the user's initial question text into the quantized language model and identifying the scene slot parameters corresponding to the initial question text through the quantized language model, the method further includes: obtaining an input data matrix; the input data matrix contains feature vectors corresponding to training question samples; determining target coordinates and corresponding quantized values ​​from the original weight matrix of the language model to be quantized; the original weight values ​​corresponding to the target coordinates maximize the reduction of the loss function when updated to quantized values; the loss function is determined based on the difference between the original weight matrix and the quantized weight matrix of the language model to be quantized and the input data matrix; determining the quantized weight matrix based on the target coordinates and quantized values; and determining the quantized language model based on the quantized weight matrix.

[0099] S504: After quantization, the language model determines the actual target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters. It then determines whether the actual target question-and-answer scenario matches the current target question-and-answer scenario in which the user is located. If they match, proceed to step S508; otherwise, proceed to step S506.

[0100] S506, update the actual target question-and-answer scenario to the current target question-and-answer scenario, and execute step S508.

[0101] S508 uses a quantized language model to identify the business type slot parameters and data variable slot parameters corresponding to the initial question text based on the context information of the current target question-answering scenario and the initial question text.

[0102] S510. Determine whether the scenario slot parameters, business type slot parameters, and data variable slot parameters are all complete. If yes, proceed to step S512; otherwise, proceed to step S514.

[0103] S512, use the scenario slot parameters, business type slot parameters, and data variable slot parameters as the slot parameters corresponding to the initial question text, and then execute step S516.

[0104] S514, the quantized language model outputs guidance information. If the missing slot parameters are obtained, the quantized language model determines the slot parameters corresponding to the initial question text based on the missing slot parameters, scenario slot parameters, business type slot parameters, and data variable slot parameters, and then executes step S516.

[0105] S516: Input the slot parameters and the initial question text into the quantized language model to obtain the standardized question text.

[0106] S518: Input the standardized question text into the vector model, and vectorize the standardized question text through the vector model to obtain the vectorized text.

[0107] S520 retrieves several local documents with the highest relevance to the vectorized text from the local knowledge base using a vector model, and retrieves several online documents with the highest relevance to the vectorized text from the website platform.

[0108] The technical solution of this embodiment innovatively integrates an interactive intent recognition mechanism and large-scale model quantization technology into the construction process of a question-answering system. It also employs a network search-assisted mechanism to ensure the question-answering system obtains the most relevant input corpus by retrieving and evaluating relevant documents online, thus providing timely and accurate information feedback. This deeply integrates and optimizes traditional document retrieval question-answering systems. It not only enables real-time access to internet resources for dynamic updates and expansion of the knowledge base, overcoming the limitations of static data, but also innovatively proposes an advanced interactive intent recognition mechanism. This mechanism uses an iterative guidance strategy to help users clarify and refine their query intent, enabling... Through intelligent interactive intent recognition algorithms, the system accurately captures and parses user query intent, eliminating ambiguity and thus improving the intelligence of the system and the user interaction experience. Quantization techniques such as Greedy Coordinate Descent (GCD) are introduced, compressing model parameters into a representation with fewer bits. The quantized large model is easier to deploy and its inference speed is significantly increased, significantly improving inference speed and system deployment convenience. This further enhances the usability and efficiency of the question-answering system, while also strengthening its ability to handle complex queries, improving the accuracy and timeliness of information retrieval and question answering, and greatly improving the usability and user satisfaction of the question-answering system.

[0109] It should be noted that the specific limitations of the above steps can be found in the specific limitations of a question-and-answer method described above.

[0110] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0111] Based on the same inventive concept, this application also provides a question-answering device for implementing the question-answering method described above. The solution provided by this device is similar to the solution described in the above method. Therefore, the specific limitations of one or more question-answering device embodiments provided below can be found in the limitations of the question-answering method above, and will not be repeated here.

[0112] In one embodiment, as shown in FIG6, a question-answering device is provided, comprising:

[0113] The parameter acquisition module 610 is used to input the user's initial question text into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model.

[0114] The standardization module 620 is used to input the slot parameters and the initial question text into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0115] Document acquisition module 630 is used to input the standardized question text into a vector model to obtain a plurality of documents with the highest relevance to the standardized question text; and

[0116] The answer module 640 is used to input the standardized question text and the plurality of documents into the quantized language model to obtain the answer text generated in response to the initial question text.

[0117] In one embodiment, the parameter acquisition module 610 is specifically configured to input the user's initial question text into the quantized language model, identify the scene slot parameters corresponding to the initial question text through the quantized language model, and determine the current target question-and-answer scenario corresponding to the initial question text based on the scene slot parameters; identify the business type slot parameters and data variable slot parameters corresponding to the initial question text through the quantized language model based on the context information of the current target question-and-answer scenario and the initial question text; and determine the slot parameters corresponding to the initial question text based on the scene slot parameters, the business type slot parameters, and the data variable slot parameters.

[0118] In one embodiment, the parameter acquisition module 610 is specifically configured to, when detecting that at least one of the scene slot parameters, the business type slot parameters, and the data variable slot parameters is incomplete, output guidance information through the quantized language model; the guidance information is used to guide the user to supplement the missing slot parameters; and when the missing slot parameters are obtained, the quantized language model determines the slot parameters corresponding to the initial question text based on the missing slot parameters, the scene slot parameters, the business type slot parameters, and the data variable slot parameters.

[0119] In one embodiment, the parameter acquisition module 610 is specifically used to determine the actual target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters through the quantized language model, and to update the actual target question-and-answer scenario to the current target question-and-answer scenario when the actual target question-and-answer scenario does not match the current target question-and-answer scenario in which the user is located.

[0120] In one embodiment, the document acquisition module 630 is specifically used to input the standardized question text into the vector model, vectorize the standardized question text through the vector model to obtain vectorized text, and retrieve a number of local documents with the highest relevance to the vectorized text from the local knowledge base and a number of online documents with the highest relevance to the vectorized text from the website platform through the vector model.

[0121] In one embodiment, the question-answering device further includes a model building module; the model building module is specifically used to acquire an input data matrix; the input data matrix contains feature vectors corresponding to training question samples; determine target coordinates and quantized values ​​corresponding to the target coordinates from the original weight matrix of the language model to be quantized; the original weight values ​​corresponding to the target coordinates maximize the reduction of the loss function when updated to the quantized values; the loss function is determined based on the difference between the original weight matrix and the quantized weight matrix of the language model to be quantized and the input data matrix; and determine the quantized weight matrix based on the target coordinates and the quantized values; and determine the quantized language model based on the quantized weight matrix.

[0122] Each module in the aforementioned question-and-answer device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0123] In one embodiment, a computer device is provided, which may be a terminal, and its internal structure diagram is shown in Figure 7. The computer device includes a processor, memory, input / output interface, communication interface, display unit, and input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are also connected to the system bus via the input / output interface. The processor of the computer device provides computing and control capabilities. The memory of the computer device includes non-volatile or volatile storage media and internal memory. The non-volatile or volatile storage media stores an operating system and computer-readable instructions. The internal memory provides an environment for the operation of the operating system and computer-readable instructions in the non-volatile or volatile storage media. The input / output interface of the computer device is used for exchanging information between the processor and external devices. The communication interface of the computer device is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, mobile cellular networks, Near Field Communication (NFC), or other technologies. When the computer-readable instructions are executed by the processor, a question-and-answer method is implemented. The display unit of this computer device is used to form a visually visible image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be an LCD screen or an e-ink screen. The input device of this computer device can be a touch layer covering the display screen, or buttons, a trackball, or a touchpad set on the casing of the computer device, or an external keyboard, touchpad, or mouse, etc.

[0124] Those skilled in the art will understand that the structure shown in Figure 7 is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0125] A computer device includes a memory and one or more processors, the memory storing computer-readable instructions that, when executed by the processors, cause the one or more processors to perform the following steps:

[0126] The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model;

[0127] The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0128] The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and

[0129] The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

[0130] One or more computer-readable storage media storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the following steps:

[0131] The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model;

[0132] The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0133] The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and

[0134] The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

[0135] A computer program product includes a computer program that, when executed by one or more processors, causes the one or more processors to perform the following steps:

[0136] The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model;

[0137] The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent;

[0138] The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and

[0139] The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

[0140] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0141] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a non-volatile computer-readable storage medium. When executed, these computer-readable instructions can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0142] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0143] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A question-and-answer method, wherein, The method includes: The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model; The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent; The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

2. The method according to claim 1, wherein, The step of inputting the user's initial question text into the quantized language model to obtain the slot parameters corresponding to the initial question text includes: The user's initial question text is input into the quantized language model. The quantized language model identifies the scene slot parameters corresponding to the initial question text and determines the current target question-and-answer scene corresponding to the initial question text based on the scene slot parameters. Based on the contextual information of the current target question-answering scenario and the initial question text, the quantized language model identifies the business type slot parameters and data variable slot parameters corresponding to the initial question text; and Based on the scenario slot parameters, the business type slot parameters, and the data variable slot parameters, determine the slot parameters corresponding to the initial question text.

3. The method according to claim 2, wherein, The step of determining the slot parameters corresponding to the initial question text based on the scenario slot parameters, the business type slot parameters, and the data variable slot parameters includes: If at least one of the scenario slot parameters, the business type slot parameters, and the data variable slot parameters is detected to be incomplete, guidance information is output through the quantized language model; the guidance information is used to guide the user to supplement the missing slot parameters; and If the missing slot parameters are obtained, the quantized language model determines the slot parameters corresponding to the initial question text based on the missing slot parameters, the scenario slot parameters, the business type slot parameters, and the data variable slot parameters.

4. The method according to claim 2, wherein, Determining the current target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters includes: The quantized language model determines the actual target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters, and updates the actual target question-and-answer scenario to the current target question-and-answer scenario if the actual target question-and-answer scenario does not match the current target question-and-answer scenario in which the user is located.

5. The method according to claim 1, wherein, The process involves inputting the standardized question text into a vector model to obtain several documents that are most relevant to the standardized question text, including: The standardized question text is input into the vector model, and the standardized question text is vectorized by the vector model to obtain vectorized text. The vector model is used to retrieve several local documents from the local knowledge base that are most relevant to the vectorized text, and to retrieve several online documents from the website platform that are most relevant to the vectorized text.

6. The method according to claim 1, wherein, Before inputting the user's initial question text into the quantized language model to obtain the slot parameters corresponding to the initial question text, the method further includes: Obtain the input data matrix; the input data matrix contains the feature vectors corresponding to the training problem samples; The target coordinates and the corresponding quantized values ​​are determined from the original weight matrix of the language model to be quantized; the original weight values ​​corresponding to the target coordinates maximize the reduction of the loss function when updated to the quantized values; the loss function is determined based on the difference between the original weight matrix and the quantized weight matrix of the language model to be quantized and the input data matrix. Based on the target coordinates and the quantized values, determine the quantized weight matrix; and The quantized language model is determined based on the quantized weight matrix.

7. A question-and-answer device, comprising: The parameter acquisition module is used to input the user's initial question text into the quantized language model to obtain the slot parameters corresponding to the initial question text. The slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model. The standardization module is used to input the slot parameters and the initial question text into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent; The document acquisition module is used to input the standardized question text into the vector model to obtain several documents that are most relevant to the standardized question text. and The answer module is used to input the standardized question text and the plurality of documents into the quantized language model to obtain the answer text generated in response to the initial question text.

8. A computer device comprising a memory and one or more processors, the memory storing computer-readable instructions that, when executed by the processors, cause the one or more processors to perform the following steps: The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model; The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent; The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

9. The computer device according to claim 8, wherein, When the processor executes the computer-readable instructions, it also performs the following steps: The user's initial question text is input into the quantized language model. The quantized language model identifies the scene slot parameters corresponding to the initial question text and determines the current target question-and-answer scene corresponding to the initial question text based on the scene slot parameters. Based on the context information of the current target question-answering scenario and the initial question text, the quantized language model identifies the business type slot parameters and data variable slot parameters corresponding to the initial question text. and Based on the scenario slot parameters, the business type slot parameters, and the data variable slot parameters, determine the slot parameters corresponding to the initial question text.

10. The computer device according to claim 9, wherein, When the processor executes the computer-readable instructions, it also performs the following steps: If at least one of the scenario slot parameters, the business type slot parameters, and the data variable slot parameters is found to be incomplete, guidance information is output through the quantized language model. The guidance information is used to guide users to fill in the missing slot parameters; and If the missing slot parameters are obtained, the quantized language model determines the slot parameters corresponding to the initial question text based on the missing slot parameters, the scenario slot parameters, the business type slot parameters, and the data variable slot parameters.

11. The computer device according to claim 9, wherein, When the processor executes the computer-readable instructions, it also performs the following steps: The quantized language model determines the actual target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters, and updates the actual target question-and-answer scenario to the current target question-and-answer scenario if the actual target question-and-answer scenario does not match the current target question-and-answer scenario in which the user is located.

12. The computer device according to claim 8, wherein, When the processor executes the computer-readable instructions, it also performs the following steps: The standardized question text is input into the vector model, and the standardized question text is vectorized by the vector model to obtain vectorized text. The vector model is used to retrieve several local documents from the local knowledge base that are most relevant to the vectorized text, and to retrieve several online documents from the website platform that are most relevant to the vectorized text.

13. The computer device according to claim 8, wherein, When the processor executes the computer-readable instructions, it also performs the following steps: Obtain the input data matrix; the input data matrix contains the feature vectors corresponding to the training problem samples; The target coordinates and the corresponding quantized values ​​are determined from the original weight matrix of the language model to be quantized; the original weight values ​​corresponding to the target coordinates maximize the reduction of the loss function when updated to the quantized values. The loss function is determined based on the difference between the original weight matrix and the quantized weight matrix of the language model to be quantized, as well as the input data matrix. Based on the target coordinates and the quantized values, determine the quantized weight matrix; and The quantized language model is determined based on the quantized weight matrix.

14. One or more computer-readable storage media storing computer-readable instructions, which, when executed by one or more processors, cause the one or more processors to perform the following steps: The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model; The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent; The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

15. The storage medium according to claim 14, wherein, When the computer-readable instructions are executed by the processor, the following steps are also performed: The user's initial question text is input into the quantized language model. The quantized language model identifies the scene slot parameters corresponding to the initial question text and determines the current target question-and-answer scene corresponding to the initial question text based on the scene slot parameters. Based on the context information of the current target question-answering scenario and the initial question text, the quantized language model identifies the business type slot parameters and data variable slot parameters corresponding to the initial question text. and Based on the scenario slot parameters, the business type slot parameters, and the data variable slot parameters, determine the slot parameters corresponding to the initial question text.

16. The storage medium according to claim 15, wherein, When the computer-readable instructions are executed by the processor, the following steps are also performed: If at least one of the scenario slot parameters, the business type slot parameters, and the data variable slot parameters is found to be incomplete, guidance information is output through the quantized language model. The guidance information is used to guide users to fill in the missing slot parameters; and If the missing slot parameters are obtained, the quantized language model determines the slot parameters corresponding to the initial question text based on the missing slot parameters, the scenario slot parameters, the business type slot parameters, and the data variable slot parameters.

17. The storage medium according to claim 15, wherein, When the computer-readable instructions are executed by the processor, the following steps are also performed: The quantized language model determines the actual target question-and-answer scenario corresponding to the initial question text based on the scenario slot parameters, and updates the actual target question-and-answer scenario to the current target question-and-answer scenario if the actual target question-and-answer scenario does not match the current target question-and-answer scenario in which the user is located.

18. The storage medium according to claim 14, wherein, When the computer-readable instructions are executed by the processor, the following steps are also performed: The standardized question text is input into the vector model, and the standardized question text is vectorized by the vector model to obtain vectorized text. The vector model is used to retrieve several local documents from the local knowledge base that are most relevant to the vectorized text, and to retrieve several online documents from the website platform that are most relevant to the vectorized text.

19. The storage medium according to claim 14, wherein, When the computer-readable instructions are executed by the processor, the following steps are also performed: Obtain the input data matrix; the input data matrix contains the feature vectors corresponding to the training problem samples; The target coordinates and the corresponding quantized values ​​are determined from the original weight matrix of the language model to be quantized; the original weight values ​​corresponding to the target coordinates maximize the reduction of the loss function when updated to the quantized values. The loss function is determined based on the difference between the original weight matrix and the quantized weight matrix of the language model to be quantized, as well as the input data matrix. Based on the target coordinates and the quantized values, determine the quantized weight matrix; and The quantized language model is determined based on the quantized weight matrix.

20. A computer program product comprising a computer program that, when executed by one or more processors, causes the one or more processors to perform the following steps: The user's initial question text is input into the quantized language model to obtain the slot parameters corresponding to the initial question text; the slot parameters include descriptive information related to the target question-answering scenario; the quantized language model is a language model obtained by compressing the model parameters of a large language model; The slot parameters and the initial question text are input into the quantized language model to obtain standardized question text; the standardized question text is text that conforms to a preset text structure and clearly reflects the user's intent; The standardized question text is input into a vector model to obtain several documents that are most relevant to the standardized question text; and The standardized question text and the plurality of documents are input into the quantized language model to obtain the response text generated in response to the initial question text.

Images