Intelligent interaction method, device and system based on large model, medium and equipment

Abstract

The invention provides an intelligent interaction method, device and system based on a large model, a medium and equipment, and is applied to the technical field of artificial intelligence. The method comprises the steps of obtaining a query intention from user query content, performing knowledge retrieval to generate an item candidate list, generating a model cue word by using the list, and screening out a target business item. After the user confirms, relevant content is retrieved in the database, a new model prompt word is generated, and the large model is input again to generate query reply information and return to the user. According to the method, a large-scale language model and knowledge retrieval are utilized, so that the service efficiency and quality of the intelligent service assistant are improved, and the user satisfaction is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a large model-based intelligent interaction method, device, system, medium and equipment. Background Art

[0002] In the service industry, including government services, the service assistant plays a vital role as a digital tool. Through automated processes and intelligent interactions, the service assistant can significantly improve service efficiency and user satisfaction. This tool not only reduces labor costs, but also enhances the accessibility and transparency of services. In addition, it provides personalized services and quick responses to meet the growing needs of the public.

[0003] However, traditional service assistants rely on old technical architectures and have difficulty implementing advanced data processing and natural language understanding. Limited technical conditions limit their effectiveness in complex environments, especially when dealing with diverse user needs, thus reducing service efficiency and quality.

[0004] Therefore, how to improve the service efficiency and service quality of intelligent service assistants based on artificial intelligence has become a technical problem that technicians in this field need to solve. Summary of the invention

[0005] In view of the above problems, the present invention provides a large model-based intelligent interaction method, device, system, medium and equipment that overcome the above problems or at least partially solve the above problems. The technical solution is as follows:

[0006] An intelligent interaction method based on a large model, comprising:

[0007] Based on the query content input by the user, obtaining the query intention information of the user;

[0008] Performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, wherein the candidate list of items includes at least one business item associated with the query intention information;

[0009] Generate a first model prompt word using the query intention information and the item candidate list;

[0010] Inputting the first model prompt word into the large model, so that the large model selects at least one target business item from the item candidate list according to the first model prompt word;

[0011] In response to the user's confirmation operation on any of the target business items, searching the database for item contents related to the target business item;

[0012] Generate a second model prompt word using the matter content and the query intention information;

[0013] Inputting the second model prompt word into the large model so that the large model generates query response information according to the second model prompt word;

[0014] The query reply information is returned to the user.

[0015] Optionally, the query intention information includes user query intention and query keywords, and the performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information includes:

[0016] Preprocessing and vectorizing the query intent information to generate a query vector;

[0017] By presetting an inverted index, a document set containing the query keyword is identified, wherein the document set includes document vectors of a plurality of documents, and the documents are associated with business items;

[0018] At least using the similarity between the query vector and each of the document vectors to screen out target documents associated with the query intent information;

[0019] Based on the business items associated with the target document, a candidate list of items of the query intention information is constructed.

[0020] Optionally, before performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, the method further includes:

[0021] Preprocessing and vectorizing documents associated with a plurality of business items to obtain vectorization results, wherein the vectorization results include document vectors of the plurality of documents;

[0022] An inverted index is constructed using the vectorization result so that each word group in the document vector is associated with the corresponding document.

[0023] Optionally, the preprocessing includes cleaning the text, standardizing the format, and removing all non-text elements.

[0024] Optionally, the step of at least using the similarity between the query vector and each of the document vectors to filter out a target document associated with the query intent information includes:

[0025] Utilizing the similarity between the query vector and each of the document vectors and the document metadata of each of the documents, target documents associated with the query intent information are screened out, wherein the document metadata includes timeliness data, authority data and user preference data of the document.

[0026] Optionally, before returning the query reply information to the user, the method further includes:

[0027] Performing a logic check on the query reply information, and after the check passes, executing the step of returning the query reply information to the user.

[0028] An intelligent interactive device based on a large model comprises: a query intention information obtaining unit, an item candidate list obtaining unit, a first model prompt word generating unit, a target business item screening unit, an item content retrieval unit, a second model prompt word generating unit, a query reply information generating unit and a query reply information returning unit.

[0029] The query intention information obtaining unit is used to obtain the query intention information of the user based on the query content input by the user;

[0030] The item candidate list obtaining unit is used to perform knowledge retrieval on the query intention information to obtain an item candidate list of the query intention information, wherein the item candidate list includes at least one business item associated with the query intention information;

[0031] The first model prompt word generating unit is used to generate a first model prompt word by using the query intention information and the item candidate list;

[0032] The target business item screening unit is used to input the first model prompt word into the large model, so that the large model screens at least one target business item from the item candidate list according to the first model prompt word;

[0033] The item content retrieval unit is used to retrieve the item content related to the target business item in the database in response to the user's confirmation operation on any of the target business items;

[0034] The second model prompt word generating unit is used to generate a second model prompt word by using the matter content and the query intention information;

[0035] The query reply information generating unit is used to input the second model prompt word into the large model, so that the large model generates query reply information according to the second model prompt word;

[0036] The query reply information returning unit is used to return the query reply information to the user.

[0037] An intelligent interactive system based on a large model, comprising: a user module, a control module, a knowledge retrieval module, an application algorithm module and a large model module.

[0038] The user module is used to obtain the user's query intention information based on the query content input by the user, and send the query intention information to the control module; request the user to confirm the target business item, and return the user confirmation result to the control module; display the query reply information;

[0039] The control module is used to receive the query intention information sent by the user module and forward the query intention information to the knowledge retrieval module; send the item candidate list and query intention information returned by the knowledge retrieval module to the application algorithm module; send the target business item to the user module; retrieve the item content related to the target business item in the database based on the user confirmation result sent by the user module, and send the item content to the application algorithm module; forward the query reply information to the user module;

[0040] The knowledge retrieval module is used to perform knowledge retrieval on the query intention information sent by the control module, obtain a candidate list of items of the query intention information, and return the candidate list of items to the control module; forward the query reply information to the control module;

[0041] The application algorithm module is used to generate a first model prompt word using the query intent information and the item candidate list, and send the first model prompt word to the large model module; send the target business item returned by the large model module to the control module; generate a second model prompt word using the item content and the query intent information, and send the second model prompt word to the large model module; forward the query reply information to the control module;

[0042] The large model module is used to screen out at least one target business item from the item candidate list based on the first model prompt word, and return the target business item to the application algorithm module; generate query response information based on the second model prompt word, and send the query response information to the application algorithm module.

[0043] A computer-readable storage medium stores a program, which, when executed by a processor, implements any of the above-mentioned large-model-based intelligent interaction methods.

[0044] An electronic device, comprising at least one processor, and at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other via the bus; and the processor is used to call program instructions in the memory to execute any of the above-mentioned large model-based intelligent interaction methods.

[0045] By means of the above technical scheme, the present invention provides a method, device, system, medium and equipment for intelligent interaction based on a large model, the method comprising: obtaining the user's query intention information based on the query content input by the user; performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, wherein the candidate list of items includes at least one business item associated with the query intention information; generating a first model prompt word using the query intention information and the candidate list of items; inputting the first model prompt word into the large model so that the large model can filter out at least one target business item from the candidate list of items based on the first model prompt word; in response to the user's confirmation operation on any target business item, retrieving the item content related to the target business item in the database; generating a second model prompt word using the item content and the query intention information; inputting the second model prompt word into the large model so that the large model can generate query reply information based on the second model prompt word; and returning the query reply information to the user. The present invention uses large-scale language models and knowledge retrieval to provide an efficient and accurate intelligent interaction method, thereby improving the service efficiency and service quality of the intelligent service assistant based on artificial intelligence, and improving user service satisfaction.

[0046] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0048] Figure 1 A flow chart showing an implementation of a large model-based intelligent interaction method provided in an embodiment of the present invention;

[0049] Figure 2 A flow chart showing another implementation of a large model-based intelligent interaction method provided in an embodiment of the present invention is shown;

[0050] Figure 3 A schematic diagram of the structure of an intelligent interactive device based on a large model provided by an embodiment of the present invention is shown;

[0051] Figure 4 A schematic diagram of the structure of an intelligent interactive system based on a large model provided by an embodiment of the present invention is shown;

[0052] Figure 5 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0053] The exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present invention are shown in the accompanying drawings, it should be understood that the present invention can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided in order to enable a more thorough understanding of the present invention and to enable the scope of the present invention to be fully communicated to those skilled in the art.

[0054] As a digital tool, service assistants play an important role. Through automated processes and intelligent interactions, service assistants have greatly improved service efficiency and user satisfaction. Such tools not only reduce labor costs, but also improve the accessibility and transparency of services. In addition, they provide personalized services and rapid responses to meet the increasing needs of the public. However, traditional service assistants still have some significant technical defects and deficiencies:

[0055] Technical limitations: Traditional service assistants rely on old technical architectures and are unable to achieve advanced data processing and natural language understanding. Limited technical conditions limit their effectiveness in complex environments, especially when dealing with diverse user needs.

[0056] Low degree of automation: Many processes still require a lot of manual intervention and lack the ability to automate complex tasks. This not only increases the workload, but also reduces efficiency and fails to meet the requirements of rapid response.

[0057] Poor interactivity: The user interface design may not be intuitive enough to effectively guide users to interact. This leads to a poor user experience, and users are easily confused or blocked during use, affecting the usability of the service.

[0058] Lack of learning ability: Traditional service assistants lack self-learning and adaptive capabilities and cannot optimize based on time and user feedback. This means that they cannot improve as user needs change, and service quality always stagnates.

[0059] Low user engagement: As traditional service assistants are not user-friendly and complex to operate, user engagement is generally low. Users may feel that the user experience is poor, which reduces the frequency of use and affects the maximization of tool benefits.

[0060] To address these challenges, a number of AI-based smart assistants have emerged in recent years. These innovative tools typically use natural language processing (NLP) and large-scale pre-trained models (such as GPT-3) to answer user questions. However, while these technological advances have brought significant improvements, existing technologies still have several limitations:

[0061] Data quality and availability: The performance of AI assistants is highly dependent on the quality of the data. If the data is inaccurate, incomplete, or biased, the AI’s decisions and responses will be negatively affected. This is a serious problem for government services that require high accuracy.

[0062] Generalization: Although large-scale models perform well on specific tasks, they may not generalize well to new or unseen situations. This limits the effectiveness of AI assistants in dealing with diverse and dynamically changing environments.

[0063] Explanation: Many AI models, especially deep learning models, are considered “black boxes” and it is difficult to provide a clear explanation of the decision-making process. This lack of explanation is particularly prominent in government services that require a high degree of transparency and traceability.

[0064] Complexity of user interaction: Users may ask ambiguous or ambiguous questions, and AI assistants need to be able to accurately understand and provide appropriate answers. This places higher demands on AI's natural language understanding capabilities.

[0065] In summary, although AI-based smart service assistants have made significant progress in improving service efficiency and user experience, problems such as data quality, generalization ability, interpretability, and complex user interactions still need to be addressed in order to achieve higher service levels and user satisfaction.

[0066] Based on this, in an embodiment of the present invention, an intelligent interaction method based on a large model is provided, which obtains the query intent from the user's query content, performs knowledge retrieval to generate a candidate list of items, uses the list to generate model prompt words, and screens out target business items. After the user confirms, the relevant content is retrieved from the database, a new model prompt word is generated, and the large model is input again to generate query reply information and return it to the user. It can be seen that the embodiment of the present invention uses large-scale language models and knowledge retrieval to enhance the retrieval and interaction capabilities of the intelligent service assistant, thereby improving the service efficiency and quality of the intelligent service assistant, and further improving user satisfaction.

[0067] like Figure 1 As shown, a flowchart of an implementation of a large model-based intelligent interaction method provided by an embodiment of the present invention is provided, and the method may include:

[0068] S100: Obtaining user query intention information based on query content input by the user.

[0069] The query content refers to the questions, requests or instructions input or submitted by the user.

[0070] The query intention information refers to the query purpose and demand information expressed by the user through the input content. Optionally, the query intention information includes the user's query intention and query keywords.

[0071] The user can type the query content in the input box or use the voice input function to ask the query content. When the user submits the query content in the input box, the embodiment of the present invention will instantly receive and record the text information of the query content, and the text information will be seamlessly transmitted to the natural language processing (NLP) engine with powerful text parsing capabilities to deeply understand the text content.

[0072] The embodiment of the present invention can start the automatic speech recognition (ASR) technology when the user inputs the query content through voice, and convert the user's voice command into text information in real time. The conversion process includes accurate audio capture, noise reduction, voice-to-text conversion, and possible further optimization of the language model, so as to ensure the accuracy of the conversion result.

[0073] Whether it is text information directly input by the user or text information converted from voice commands, it can be further semantically analyzed and processed by the NLP engine. The NLP engine uses a series of complex algorithms and models such as intent recognition, entity extraction, and sentiment analysis to parse the deep meaning of the input text.

[0074] In the intent recognition stage, the NLP engine identifies the user's overall query intent through pattern matching, machine learning or deep learning methods, such as whether the user is seeking information, making a complaint or making an appointment, etc. Keyword extraction focuses on the core content of the user's query and provides a basis for providing accurate answers or actions.

[0075] Pattern matching refers to a method based on predefined rules and templates to detect specific patterns in user input to identify intent. For example: User input: "I want to book a flight tomorrow." Matching template: I want to book ABC (this template can match a variety of booking requests). Identify intent: Book a flight.

[0076] Machine learning mainly trains models to automatically identify user intent. For example, a user inputs: "How do I change my password?" The model predicts that the intent is "account settings."

[0077] Deep learning mainly uses neural networks to capture complex patterns and semantics in user input and identify user intent. For example, if a user inputs: "What's the weather like today?", the deep learning model predicts that the intent is "check the weather".

[0078] The embodiments of the present invention may combine multiple intent recognition methods to improve the accuracy and robustness of intent recognition. For example, the embodiments of the present invention may first use a pattern matching method for preliminary screening, and then combine a machine learning or deep learning model for further accurate classification.

[0079] S110. Perform knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, wherein the candidate list of items includes at least one business item associated with the query intention information.

[0080] Among them, knowledge retrieval refers to finding business matters related to the query intention information by accessing the stored knowledge base or database.

[0081] Among them, business items refer to specific tasks, activities or events that need to be handled in a specific business or workflow. For example, the content of a business item may include specific task instructions, operation steps, related documents or reference materials, etc.

[0082] Specifically, the embodiment of the present invention can generate a candidate list of items of the query intention information based on one or more business items retrieved by the query intention information.

[0083] Optionally, the number of business items in the item candidate list may be limited to 40.

[0084] S120: Generate a first model prompt word using the query intent information and the item candidate list.

[0085] The model prompt refers to the text snippet provided to the big model. The model prompt can be a question, a command, an incomplete sentence, or anything you want the big model to continue. The main function of the model prompt is to guide the model to generate relevant and coherent text. The big model will understand the context based on the provided prompt and generate logical and semantic continuation text.

[0086] Specifically, the embodiment of the present invention can combine the query intent information and the candidate list of items to generate the first model prompt word. The first model prompt word contains a clear description of the user's needs and guides the large model to filter from the candidate list of items. For example: the first model prompt word: "The user wants to change the account password. Please select the most relevant from the following items: account setting steps, password change guide, security question answering".

[0087] S130. Input the first model prompt word into the large model, so that the large model selects at least one target business item from the item candidate list according to the first model prompt word.

[0088] Among them, big models refer to deep learning models that have been trained with large-scale data and have a large number of parameters in the fields of natural language processing (NLP) and artificial intelligence (AI).

[0089] Specifically, an embodiment of the present invention can pass the generated first model prompt word as input to the big model. The big model parses the first model prompt word to understand the user's intention and candidate items. The model uses its language understanding ability, combined with context and semantic information, to evaluate the relevance of each candidate item. The model evaluates the relevance of candidate items based on its internal algorithm and training data and sorts or filters them, and finally outputs the target business items that best meet the user's query intention. The big model can use attention mechanisms, context understanding, and semantic matching methods for screening. For example: The first model prompt word input to the big model: "The user wants to change the account password, please select the most relevant from the following items: account setup steps, password change guide, security question answer". The big model outputs: "Password Change Guide".

[0090] The embodiment of the present invention can accurately select a target business item that best meets user needs from multiple candidate items by effectively utilizing the understanding and reasoning capabilities of the high-level language model.

[0091] S140 . In response to a user's confirmation operation on any target business item, retrieve item contents related to the target business item from a database.

[0092] Among them, the content of the matter refers to the detailed information, steps, rules or guidance related to the business matter. These contents usually include the specific operating procedures, required materials, precautions, relevant policies, etc. required to perform a certain business. For example: the content of the business matter "account registration" may include the required personal information (such as name, email, phone number), registration steps, verification process, and the next step after successful registration. The content of the business matter "apply for a refund" may include the refund policy, application steps, required materials to be submitted (such as purchase receipts, order numbers, etc.), review process and time, refund arrival time and method. The content of the business matter "service appointment" may include the types of services that can be booked, appointment time periods, appointment methods (such as online or telephone appointments), regulations for canceling or modifying appointments, and confirmation information after successful appointments.

[0093] In the embodiment of the present invention, when a target business item is screened out by the large model, it can be assumed that the user has confirmed the target business item, and the detailed item content can be directly retrieved from the database according to the item name of the confirmed target business item.

[0094] In the case where a large model screens out multiple target business items, the embodiment of the present invention can display these target business items to the user through a pre-designed interactive process, and request the user to confirm the specific target business item, so as to further clarify and refine the user's needs. When the user confirms any target business item, the detailed item content will be retrieved from the database according to the item name of the confirmed target business item.

[0095] S150: Generate a second model prompt word using the matter content and query intent information.

[0096] Specifically, the embodiments of the present invention can construct a structured prompt word based on the query intention information, and combine it with the matter content to convert the user's query intention into a specific question or task. For example: if the user asks about the details of a service, the prompt word may include the name of the service and the specific aspects that the user is concerned about (such as cost, time, steps, etc.). The matter content and the extracted prompt words are combined according to a predetermined template or format to generate a second model prompt word that can clearly and accurately express the user's needs. For example: The second model prompt word may be: "Detailed information about {business name}, with special attention to {specific aspects that the user is concerned about}".

[0097] S160: Input the second model prompt word into the large model, so that the large model generates query response information according to the second model prompt word.

[0098] Specifically, the embodiment of the present invention can input the generated second model prompt words into the big model. After receiving the second model prompt words, the big model will use its built-in knowledge and understanding ability to analyze the content of the second model prompt words, extract relevant information from its huge pre-training data, and combine the context in the second model prompt words to generate coherent and meaningful query response information. The query response information is used to comprehensively and accurately answer the user's query and provide specific business details or guidance. For example: the big model may generate a detailed text to explain various aspects of the business matters that the user is concerned about.

[0099] For ease of understanding, an example is given here: when the user intends to apply for a refund, and the content of the matter is the refund policy, application steps, required materials and review process, the generated second model prompt words can be "Detailed information about applying for a refund, with special attention to the refund policy, application steps and required materials". After the second model prompt words are input into the large model, the query response information generated by the large model can be "The steps to apply for a refund are as follows: 1. Log in to your account and visit the order history page. 2. Select the order to be refunded and click the "Apply for Refund" button. 3. Submit the required materials (such as purchase receipt, order number, etc.). 4. Wait for review, the review process usually takes 3-5 working days. 5. After the review is passed, the refund will be returned to your payment account within 7 working days." Through this process, users can get clear and detailed answers to meet their query needs.

[0100] S170: Return the query reply information to the user.

[0101] Specifically, the embodiments of the present invention can display the query reply information to the user through the user interface, and set a feedback mechanism to enable the user to share the use experience and suggestions related to the query reply information.

[0102] The present invention provides an intelligent interaction method based on a large model, the method comprising: obtaining the user's query intention information based on the query content input by the user; performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, wherein the candidate list of items includes at least one business item associated with the query intention information; generating a first model prompt word using the query intention information and the candidate list of items; inputting the first model prompt word into the large model so that the large model can filter out at least one target business item from the candidate list of items based on the first model prompt word; in response to the user's confirmation operation on any target business item, retrieving the item content related to the target business item in the database; generating a second model prompt word using the item content and the query intention information; inputting the second model prompt word into the large model so that the large model can generate query reply information based on the second model prompt word; and returning the query reply information to the user. The present invention uses a large-scale language model and knowledge retrieval to provide an efficient and accurate intelligent interaction method, thereby improving the service efficiency and service quality of the intelligent service assistant based on artificial intelligence, and improving user service satisfaction.

[0103] Optional, based on Figure 1 The method shown, such as Figure 2 As shown, a flowchart of another implementation of the large model-based intelligent interaction method provided in an embodiment of the present invention is shown, and step S110 may include:

[0104] S200: Preprocess and vectorize the query intent information to generate a query vector.

[0105] Among them, preprocessing is the process of cleaning and normalizing the query intent information for subsequent analysis and processing.

[0106] Optionally, preprocessing includes cleaning text, standardizing the format, and removing all non-text elements. Among them, cleaning text is to remove various noises and irrelevant information in the text, with the aim of making the text cleaner and easier to process. These noises and irrelevant information can include: punctuation marks, special characters, web page tags, and extra spaces. Standardizing the format is to convert the text into a unified format to ensure the consistency of each part. For example: converting uppercase and lowercase letters and unifying the encoding format. Removing all non-text elements means deleting parts of the text that do not belong to natural language so that subsequent processing is more focused and accurate. These non-text elements include: numbers and emoticons.

[0107] Among them, vectorization is the process of converting text data into numerical vectors so that computers can understand and process this data. In this process, each text unit (such as a word, phrase, or entire sentence) is mapped to a point in a high-dimensional space. This point is represented by a set of numerical values ​​(vectors) that capture the semantic and contextual information of the text unit.

[0108] The embodiment of the present invention may use the BERT (Bidirectional Encoder Representations from Transformers) word embedding method to vectorize the pre-processed query intent information, thereby generating a query vector.

[0109] The embodiment of the present invention generates a query vector by preprocessing and vectorizing the query intention information, so that the system can understand the user's query content more accurately.

[0110] S210. Identify a document set containing the query keyword through a preset inverted index, wherein the document set includes document vectors of a plurality of documents, and the documents are associated with business items.

[0111] Among them, the inverted index is an efficient data structure for text retrieval and information retrieval. It associates each unique keyword (or phrase) with a list of documents containing the keyword, making query operations faster and more efficient.

[0112] The basic structure of the preset inverted index includes a keyword dictionary and a document list, wherein the keyword dictionary is used to store all unique keywords. The document list is a list of all documents containing each keyword.

[0113] The embodiment of the present invention can use a preset inverted index to quickly identify all documents containing the query keyword, and these documents will be extracted to form a document set.

[0114] S220: Filter out target documents associated with the query intent information by at least using the similarity between the query vector and each document vector.

[0115] In order to evaluate the relevance between the query vector and the document, the embodiment of the present invention may calculate the similarity between the query vector and each document vector. Specifically, the embodiment of the present invention may use cosine similarity to measure the similarity between the query vector and each document vector.

[0116] Specifically, the embodiment of the present invention can sort the documents according to the similarity scores, and select the document with the highest similarity as the target document associated with the query intention information.

[0117] S230: Based on the business items associated with the target document, a candidate list of items of query intent information is constructed.

[0118] The embodiment of the present invention can further improve the relevance and practicality of user search results by constructing a candidate list of matters related to the target document based on the business matters related to the target document, and help to find business matters related to the target document faster and more accurately.

[0119] The embodiment of the present invention preprocesses and vectorizes the query intention information, generates a query vector, identifies a set of related documents through a preset inverted index, calculates the similarity to filter the target documents, and then constructs a candidate list of matters, which can significantly improve the accuracy and efficiency of the query, and facilitates the subsequent rapid generation of model prompt words, speeds up the interaction process, and thus improves the user experience.

[0120] Optional, in the above Figure 2 On the basis of one or more corresponding embodiments, in another optional embodiment provided by the embodiment of the present invention, before performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, the method may further include:

[0121] Documents associated with multiple business items are preprocessed and vectorized to obtain vectorization results, wherein the vectorization results include document vectors of multiple documents; an inverted index is constructed using the vectorization results to associate each phrase in the document vector with a corresponding document.

[0122] Specifically, the embodiment of the present invention can pre-select and extract keywords from each document, establish a mapping relationship from keywords to documents, and associate documents with relevant business items so that relevant business items can be quickly found during retrieval.

[0123] The embodiment of the present invention can significantly improve query efficiency and accuracy by preprocessing and vectorizing documents associated with multiple business items and constructing an inverted index, support complex query operations, improve storage and retrieval efficiency, and associate business items, thereby greatly improving user experience.

[0124] Optional, in the above Figure 2 On the basis of one or more corresponding embodiments, in another optional embodiment provided by the embodiment of the present invention, at least the similarity between the query vector and each document vector is used to filter out the target document associated with the query intent information, which may specifically include:

[0125] The target documents associated with the query intent information are screened out by using the similarity between the query vector and each document vector and the document metadata of each document, wherein the document metadata includes the timeliness data, authority data and user preference data of the document.

[0126] Among them, document metadata refers to the structured information about the document, which describes the attributes and characteristics of the document.

[0127] Specifically, the embodiment of the present invention can extract the document metadata of each document from the document library through the document ID of the document, and then standardize or normalize all the document metadata to the same scale. For example, the Min-Max normalization method can be used to scale all data to between 0 and 1. The embodiment of the present invention can pre-set weights for similarity scores and each document metadata, so as to calculate a comprehensive score by weighted calculation. Among them, the weight can be adjusted according to actual needs and experience to ensure that the final score can reasonably reflect the importance of the document. Finally, the documents are sorted according to the comprehensive score, and the document with the highest score is selected as the target document associated with the query intent information.

[0128] The embodiment of the present invention can significantly improve the effectiveness of the screened target documents by comprehensively considering similarity, timeliness, authority and user preferences, thereby facilitating the subsequent provision of accurate intelligent interaction.

[0129] Optional, in the above Figure 1 On the basis of one or more corresponding embodiments, in another optional embodiment provided by the embodiment of the present invention, before returning the query reply information to the user, the method may further include:

[0130] Perform a logical check on the query reply information, and after the check passes, execute the step of returning the query reply information to the user.

[0131] The logic check may include at least one of content consistency check, logic coherence check, timeliness check, authority check, user preference check, and form and grammar check.

[0132] Among them, the content consistency check may include the check contents of keyword matching, semantic matching and context consistency. Specifically, the embodiment of the present invention can detect whether the query reply information is logically consistent with the query content, and check the semantic and grammatical correctness of the query reply information. Furthermore, the embodiment of the present invention can check whether the query reply information contains keywords related to the query content, and use natural language processing tools to ensure that the semantics of the query reply information is consistent with the query meaning. Figure 1 Then check whether there is any contradiction in the context of the query reply information.

[0133] Among them, the logical coherence check includes logical reasoning check, causal relationship analysis and paragraph connectivity. Specifically, the embodiment of the present invention can detect whether the logical structure of the query reply information is clear, and whether the viewpoints and conclusions are obtained based on a reasonable reasoning process. Furthermore, the embodiment of the present invention can use logical reasoning tools or algorithms to check whether the reasoning process in the query reply information is reasonable, and check whether the causal relationship in the query reply information is clear and reasonable, and then check the logical connection between paragraphs to ensure that the content of the query reply information is coherent and avoid jumps or breaks.

[0134] Among them, the timeliness check includes timestamp check and timeliness comparison. Specifically, the embodiment of the present invention can perform timeliness check on time-sensitive data or events in the query reply information. Furthermore, the embodiment of the present invention can verify the timestamps of the data and events referenced in the query reply information to ensure that these data and events are the latest, and compare the query reply information with the latest relevant information to ensure timeliness.

[0135] Among them, the authority verification may include source verification and citation quality verification. Specifically, the embodiment of the present invention may perform source verification and application quality verification for questions related to professional fields in the query reply information. Furthermore, the embodiment of the present invention may verify whether the source of the content related to professional fields in the query reply information is published by an authoritative organization or expert, and evaluate the quality and credibility of the cited documents, articles or reports.

[0136] Among them, user preference verification includes user historical behavior analysis and personalized recommendation. Specifically, the embodiment of the present invention can verify whether the query reply information meets the personalized needs and preferences of the user. Furthermore, the embodiment of the present invention can adjust the content of the query reply information to match the user's preferences based on the user's historical query and click behavior, and use a recommendation algorithm to collect the user's interests and preferences to provide personalized query reply information.

[0137] Among them, the form and grammar check may include grammar check, spell check and format check. Specifically, the embodiment of the present invention can use a grammar check tool to check whether the grammar of the reply content of the query reply information is correct, and use a spell check tool to check whether there are spelling errors in the reply content of the query reply information, and check whether the format of the query reply information is clear and the paragraphs are distinct, so that the user can read it.

[0138] Optionally, in the case where the logic check fails, the embodiment of the present invention can regenerate query reply information for the specific check method that fails the check.

[0139] The embodiment of the present invention can effectively improve the quality of query reply information through logical verification, and ensure its accuracy, relevance and user satisfaction.

[0140] Although operations are depicted in a particular order, this should not be understood as requiring that the operations be performed in the particular order shown or in a sequential order.Multitasking and parallel processing may be advantageous under certain circumstances.

[0141] It should be understood that the various steps described in the method embodiments of the present invention may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present invention is not limited in this respect.

[0142] Corresponding to the above method embodiment, the embodiment of the present invention also provides an intelligent interaction device based on a large model, whose structure is as follows: Figure 3 As shown, it includes: a query intention information obtaining unit 10, an item candidate list obtaining unit 20, a first model prompt word generating unit 30, a target business item screening unit 40, an item content retrieval unit 50, a second model prompt word generating unit 60, a query reply information generating unit 70 and a query reply information returning unit 80.

[0143] The query intention information obtaining unit 10 is used to obtain the user's query intention information based on the query content input by the user.

[0144] The item candidate list obtaining unit 20 is used to perform knowledge retrieval on the query intention information to obtain an item candidate list of the query intention information, wherein the item candidate list includes at least one business item associated with the query intention information.

[0145] The first model prompt word generating unit 30 is used to generate the first model prompt word by using the query intention information and the item candidate list.

[0146] The target business item screening unit 40 is used to input the first model prompt word into the large model, so that the large model screens out at least one target business item from the item candidate list according to the first model prompt word.

[0147] The item content retrieval unit 50 is used to retrieve the item content related to the target business item in the database in response to the user's confirmation operation on any target business item.

[0148] The second model prompt word generating unit 60 is used to generate the second model prompt word by using the matter content and the query intention information.

[0149] The query reply information generating unit 70 inputs the second model prompt word into the large model, so that the large model generates the query reply information according to the second model prompt word.

[0150] The query reply information returning unit 80 is used to return the query reply information to the user.

[0151] Optionally, the query intent information includes user query intent and query keywords.

[0152] Optionally, the issue candidate list obtaining unit 20 may include: a query vector generating subunit, a document set identifying subunit, a target document screening subunit and an issue candidate list constructing subunit.

[0153] The query vector generation subunit is used to preprocess and vectorize the query intent information to generate a query vector.

[0154] The document set identification subunit is used to identify a document set containing the query keyword through a preset inverted index, wherein the document set includes document vectors of multiple documents, and the documents are associated with business items.

[0155] The target document screening subunit is used to screen out target documents associated with the query intent information by at least using the similarity between the query vector and each document vector.

[0156] The item candidate list construction subunit is used to construct an item candidate list of query intent information based on the business items associated with the target document.

[0157] Optionally, the large model-based intelligent interaction device may further include: a vectorization result obtaining unit and an inverted index building unit.

[0158] The vectorization result obtaining unit is used for the item candidate list obtaining unit 20 to perform knowledge retrieval on the query intention information. Before obtaining the item candidate list of the query intention information, the documents associated with multiple business items are preprocessed and vectorized to obtain vectorization results, wherein the vectorization results include document vectors of multiple documents.

[0159] The inverted index building unit is used to build an inverted index using the vectorization result, so that each word group in the document vector is associated with a corresponding document.

[0160] Optionally, preprocessing includes cleaning the text, standardizing the format, and removing all non-text elements.

[0161] Optionally, the target document screening subunit can also be used to specifically screen out target documents associated with the query intent information by utilizing the similarity between the query vector and each document vector and the document metadata of each document, wherein the document metadata includes the document's timeliness data, authority data, and user preference data.

[0162] Optionally, the large model-based intelligent interaction device may further include: a logic verification unit.

[0163] The logic verification unit is used to perform a logic verification on the query reply information before the query reply information returning unit 80 returns the query reply information to the user, and trigger the query reply information returning unit 80 after the verification passes.

[0164] The present invention provides an intelligent interaction device based on a large model, which is used to: obtain the user's query intention information based on the query content input by the user; perform knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, wherein the candidate list of items includes at least one business item associated with the query intention information; generate a first model prompt word using the query intention information and the candidate list of items; input the first model prompt word into the large model so that the large model selects at least one target business item from the candidate list of items based on the first model prompt word; retrieve the item content related to the target business item from the database in response to the user's confirmation operation on any target business item; generate a second model prompt word using the item content and the query intention information; input the second model prompt word into the large model so that the large model generates query reply information based on the second model prompt word; and return the query reply information to the user. The present invention uses a large-scale language model and knowledge retrieval to provide an efficient and accurate intelligent interaction method, thereby improving the service efficiency and service quality of the intelligent service assistant based on artificial intelligence, and improving user service satisfaction.

[0165] Regarding the device in the above embodiment, the specific manner in which each unit performs the operation has been described in detail in the embodiment of the method, and will not be elaborated here.

[0166] The large model-based intelligent interaction device includes a processor and a memory. The query intention information obtaining unit 10, the item candidate list obtaining unit 20, the first model prompt word generating unit 30, the target business item screening unit 40, the item content retrieval unit 50, the second model prompt word generating unit 60, the query reply information generating unit 70 and the query reply information returning unit 80 are all stored in the memory as program units, and the processor executes the above program units stored in the memory to realize corresponding functions.

[0167] The processor contains a kernel, which calls the corresponding program unit from the memory. One or more kernels can be set, and by adjusting the kernel parameters, large-scale language models and knowledge retrieval can be used to improve the service efficiency and quality of the intelligent service assistant and improve user satisfaction.

[0168] The embodiment of the present invention also provides an intelligent interactive system based on a large model, the structure of which is as follows: Figure 4 As shown, it may include: a user module 100 , a control module 200 , a knowledge retrieval module 400 , an application algorithm module 500 and a large model module 300 .

[0169] The user module 100 is used to obtain the user's query intention information based on the query content input by the user, and send the query intention information to the control module 200; request the user to confirm the target business item, and return the user confirmation result to the control module 200; display the query reply information.

[0170] The control module 200 is used to receive the query intention information sent by the user module 100 and forward the query intention information to the knowledge retrieval module 400; send the item candidate list and query intention information returned by the knowledge retrieval module 400 to the application algorithm module 500; send the target business item to the user module 100; retrieve the item content related to the target business item in the database based on the user confirmation result sent by the user module 100, and send the item content to the application algorithm module 500; forward the query reply information to the user module 100.

[0171] The knowledge retrieval module 400 is used to perform knowledge retrieval on the query intention information sent by the control module 200, obtain a candidate list of items of the query intention information, and return the candidate list of items to the control module 200; and forward the query reply information to the control module 200.

[0172] The application algorithm module 500 is used to generate a first model prompt word using query intent information and a candidate list of items, and send the first model prompt word to the large model module 300; send the target business item returned by the large model module 300 to the control module 200; generate a second model prompt word using the item content and query intent information, and send the second model prompt word to the large model module 300; and forward the query reply information to the control module 200.

[0173] The large model module 300 is used to filter out at least one target business item from the item candidate list based on the first model prompt word, and return the target business item to the application algorithm module 500; generate query reply information based on the second model prompt word, and send the query reply information to the application algorithm module 500.

[0174] Optionally, the user module 100 provides multi-channel access, natural language understanding and context retention functions. Multi-channel access: The system supports cross-platform operation, and users can access services through the Web, mobile devices or social media. This requires the system to have a high degree of compatibility and adaptability. Natural language understanding: The user module 100 integrates advanced NLP technology and can parse the user's natural language input, including colloquial expressions, slang and industry terms. Context retention: The system can record and understand the context of user interactions, making the conversation more fluent and providing relevant answers based on the context.

[0175] Optionally, the control module 200 provides request distribution. Status monitoring and event exception handling functions. Request distribution: The control module 200 evaluates the user's request and assigns it to the most appropriate processing module, such as knowledge retrieval or direct application algorithm module 500 for processing. Status monitoring: Real-time monitoring of the operating status of each module, including response time, system load and error rate, to ensure efficient operation of the system. Event exception handling: When the system encounters an unexpected situation or the user's request cannot be met, the control module 200 can take alternative processes, such as providing error information or requesting further clarification from the user.

[0176] Optionally, the knowledge retrieval module 400 provides vectorization processing. Large-scale index construction and relevance scoring functions. Vectorization processing: converting the user's query and the information in the knowledge base into vector form to use the vector space model for efficient retrieval. Large-scale index construction: building and maintaining an efficient index system to support fast retrieval of large-scale data sets. Relevance scoring: scoring the retrieval results and sorting them according to the relevance, timeliness and accuracy of the content to ensure that the most useful information is returned first.

[0177] Optionally, the algorithm module 500 provides data preprocessing. Prompt writing and result post-processing functions. Data preprocessing: Clean the data input by the user, remove noise and irrelevant information, and ensure the quality and consistency of the data. Prompt writing: Write different prompts for the big model according to different tasks. Result post-processing: Proofread the answers generated by the big model, including logical consistency checking, language polishing and format adjustment, to ensure the professionalism and readability of the answers.

[0178] Optionally, the large model module 300 provides deep semantic understanding and multi-tasking capabilities. Deep semantic understanding: The large model can deeply understand the semantics of user queries, including implicit meanings, emotional colors, and specific needs. Multi-tasking: The large model is designed to handle multiple tasks simultaneously, such as information retrieval, question answering, etc.

[0179] Optionally, the user module 100, the control module 200, the knowledge retrieval module 400, the application algorithm module 500 and the large model module 300 interact through a security protocol to ensure the privacy and security of user data.

[0180] The embodiment of the present invention provides an intelligent interactive system based on a big model, which uses a big model and advanced algorithms to provide fast, accurate and user-friendly government services. Its main features include adaptive learning, efficient retrieval and modular design, which enables it to achieve automatic response around the clock. This not only reduces operating costs, but also significantly improves service efficiency and user satisfaction.

[0181] An embodiment of the present invention provides a computer-readable storage medium on which a program is stored. When the program is executed by a processor, the intelligent interaction method based on a large model is implemented.

[0182] An embodiment of the present invention provides a processor, which is used to run a program, wherein the intelligent interaction method based on a large model is executed when the program is running.

[0183] like Figure 5 As shown, an embodiment of the present invention provides an electronic device 1000, which includes at least one processor 1001, at least one memory 1002 connected to the processor 1001, and a bus 1003; wherein the processor 1001 and the memory 1002 communicate with each other through the bus 1003; the processor 1001 is used to call the program instructions in the memory 1002 to execute the above-mentioned intelligent interaction method based on the large model. The electronic device in this article can be a server, a PC, a PAD, a mobile phone, etc.

[0184] The present invention also provides a computer program product, which, when executed on an electronic device, is suitable for executing a program that initializes the steps of a large model-based intelligent interaction method.

[0185] The present invention is described with reference to flowcharts and / or block diagrams of methods, apparatuses, electronic devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable device to generate a machine, so that the instructions executed by the processor of the computer or other programmable device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0186] In a typical configuration, an electronic device includes one or more processors (CPU), a memory, and a bus. The electronic device may also include an input / output interface, a network interface, and the like.

[0187] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM, and the memory includes at least one memory chip. The memory is an example of a computer-readable medium.

[0188] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0189] 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, stored data, displayed data, etc.) involved in the present invention are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0190] In the description of the present invention, it should be understood that the terms "up", "down", "front", "back", "left" and "right" etc. indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings. They are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the positions or elements referred to must have specific directions, be constructed and operate in specific directions. Therefore, they should not be understood as limitations of the present invention.

[0191] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. It should also be noted that the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0192] It should be understood by those skilled in the art that the embodiments of the present invention may be provided as methods, systems or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware aspects. Moreover, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0193] The above are only embodiments of the present invention and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent substitution, improvement, etc. made within the spirit and principle of the present invention should be included in the scope of the present invention.

Claims

1. An intelligent interaction method based on a large model, characterized in that: include: Based on the query content input by the user, obtaining the query intention information of the user; Performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, wherein the candidate list of items includes at least one business item associated with the query intention information; Generate a first model prompt word using the query intention information and the item candidate list; Inputting the first model prompt word into the large model, so that the large model selects at least one target business item from the item candidate list according to the first model prompt word; In response to the user's confirmation operation on any of the target business items, searching the database for item contents related to the target business item; Generate a second model prompt word using the matter content and the query intention information; Inputting the second model prompt word into the large model so that the large model generates query response information according to the second model prompt word; The query reply information is returned to the user.

2. The method according to claim 1, characterized in that The query intention information includes the user's query intention and query keywords, and the step of performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information includes: Preprocessing and vectorizing the query intent information to generate a query vector; By presetting an inverted index, a document set containing the query keyword is identified, wherein the document set includes document vectors of a plurality of documents, and the documents are associated with business items; At least using the similarity between the query vector and each of the document vectors to screen out target documents associated with the query intent information; Based on the business items associated with the target document, a candidate list of items of the query intention information is constructed.

3. The method according to claim 2, characterized in that Before performing knowledge retrieval on the query intention information to obtain a candidate list of items of the query intention information, the method further includes: Preprocessing and vectorizing documents associated with a plurality of business items to obtain vectorization results, wherein the vectorization results include document vectors of the plurality of documents; An inverted index is constructed using the vectorization result so that each word group in the document vector is associated with the corresponding document.

4. The method according to claim 3, characterized in that The pre-processing includes cleaning the text, standardizing the format, and removing all non-text elements.

5. The method according to claim 2, characterized in that: The step of at least using the similarity between the query vector and each of the document vectors to select a target document associated with the query intent information includes: Utilizing the similarity between the query vector and each of the document vectors and the document metadata of each of the documents, target documents associated with the query intent information are screened out, wherein the document metadata includes timeliness data, authority data and user preference data of the document.

6. The method according to claim 1, characterized in that Before returning the query reply information to the user, the method further includes: Performing a logic check on the query reply information, and after the check passes, executing the step of returning the query reply information to the user.

7. An intelligent interactive device based on a large model, characterized in that: include: A query intention information obtaining unit, an item candidate list obtaining unit, a first model prompt word generating unit, a target business item screening unit, an item content retrieval unit, a second model prompt word generating unit, a query reply information generating unit and a query reply information returning unit, The query intention information obtaining unit is used to obtain the query intention information of the user based on the query content input by the user; The item candidate list obtaining unit is used to perform knowledge retrieval on the query intention information to obtain an item candidate list of the query intention information, wherein the item candidate list includes at least one business item associated with the query intention information; The first model prompt word generating unit is used to generate a first model prompt word by using the query intention information and the item candidate list; The target business item screening unit is used to input the first model prompt word into the large model, so that the large model screens at least one target business item from the item candidate list according to the first model prompt word; The item content retrieval unit is used to retrieve the item content related to the target business item in the database in response to the user's confirmation operation on any of the target business items; The second model prompt word generating unit is used to generate a second model prompt word by using the matter content and the query intention information; The query reply information generating unit is used to input the second model prompt word into the large model, so that the large model generates query reply information according to the second model prompt word; The query reply information returning unit is used to return the query reply information to the user.

8. An intelligent interactive system based on a large model, characterized in that: include: User module, control module, knowledge retrieval module, application algorithm module and large model module, The user module is used to obtain the user's query intention information based on the query content input by the user, and send the query intention information to the control module; request the user to confirm the target business item, and return the user confirmation result to the control module; display the query reply information; The control module is used to receive the query intention information sent by the user module and forward the query intention information to the knowledge retrieval module; send the item candidate list and query intention information returned by the knowledge retrieval module to the application algorithm module; send the target business item to the user module; retrieve the item content related to the target business item in the database based on the user confirmation result sent by the user module, and send the item content to the application algorithm module; forward the query reply information to the user module; The knowledge retrieval module is used to perform knowledge retrieval on the query intention information sent by the control module, obtain a candidate list of items of the query intention information, and return the candidate list of items to the control module; forwarding the query reply information to the control module; The application algorithm module is used to generate a first model prompt word by using the query intention information and the item candidate list, and send the first model prompt word to the large model module; Sending the target business item returned by the large model module to the control module; Generate a second model prompt word using the matter content and the query intent information, and send the second model prompt word to the large model module; forwarding the query reply information to the control module; The large model module is used to screen out at least one target business item from the item candidate list based on the first model prompt word, and return the target business item to the application algorithm module; generate query reply information based on the second model prompt word, and send the query reply information to the application algorithm module.

9. A computer-readable storage medium having a program stored thereon, characterized in that: When the program is executed by a processor, the large model-based intelligent interaction method as described in any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: The electronic device includes at least one processor, and at least one memory and a bus connected to the processor; wherein the processor and the memory communicate with each other via the bus; the processor is used to call program instructions in the memory to execute the large model-based intelligent interaction method as described in any one of claims 1 to 6.

Images