Active question and answer interaction system and method based on large language model and personalized recommendation technology and electronic equipment

Abstract

The invention relates to an active question and answer interaction system and method based on a large language model and a personalized recommendation technology and electronic equipment. The method comprises the following steps: acquiring session data through a data processing module, and acquiring vector data corresponding to the session data; analyzing a historical dialogue between the vector data and a user through an active session scene triggering unit, calculating an active session coverage degree through an active session coverage degree dynamic control unit when a preset active session scene meeting a preset triggering condition exists, and if the active session coverage degree does not exceed a preset threshold value or not, sending the active session coverage degree dynamic control unit to the active session scene triggering unit; initiating an active dialogue to the user through the versations corresponding to the target scenes, and outputting the active dialogue through an output module; according to the method and the device, the real-time input information of the user and the historical dialogue information of the user are analyzed in real time, so that multiple interactions of active insertion can be realized, the enthusiasm and the duration of the dialogue of the user are effectively increased, and the continuity and the consistency of the dialogue are ensured.

Description

Technical Field

[0001] The present invention relates to the field of intelligent dialogue, and in particular to an active question-answering interaction system, method and electronic device based on a large language model and personalized recommendation technology. Background Art

[0002] In recent years, with the development of large language models (LLM), the field of human-computer dialogue has experienced significant technological innovation. Although the existing intelligent question-answering systems have been able to understand and generate natural language to a certain extent, they still face challenges in personalized interaction and proactive dialogue coverage. For example:

[0003] The existing technology (application number: 202410381555.9, publication number: CN 117972069 B) uses the traditional natural language understanding method combined with the Markov decision process (MDP) model to make decisions. This method has problems such as not being able to handle multi-round question-answering and dynamically changing environments well. In particular, it is difficult to maintain coherence and context consistency when dealing with rapidly changing dialogue scenarios. Specifically, in this method, the MDP assumes that all elements of the model (state, action, transition probability, reward function) are known, which is almost impossible in actual application. The user's interactions and intentions may vary greatly, and the system can only list limited user intentions and states. Secondly, the MDP model assumes that the future state depends only on the current state and action, and has nothing to do with past history. This assumption is particularly distorted in multi-round question-answering scenarios, and the user's subsequent conversation content is likely to be associated with historical conversations. Finally, the MDP assumes that the dynamic characteristics of the environment remain unchanged at each decision moment, which is called temporal consistency. This assumption is not valid in a dynamically changing environment. Users may change their conversation choices and interests due to changes in certain external conditions.

[0004] Existing technology (application number: 202210876585.8, publication number: CN 115527527 A) can actively initiate a conversation based on the user's behavior pattern. Although this technology can initiate a conversation based on the user's behavior, it is mainly limited to performing specific tasks, lacks flexibility and breadth, and cannot adapt to the needs of open domain conversations, that is, it has limited ability to handle diverse topics and non-preset scenarios.

[0005] The prior art (application number: 202410249210.8, publication number: CN 117972216 A) proposes a method based on deep learning to improve the performance of the dialogue management system. The active dialogue module in this method is only triggered in the first round of intelligent question and answer. In the subsequent user interaction process, it cannot be judged by context and other current environmental information, and thus it is impossible to perform multiple active interruptions in the middle of the entire dialogue process. Although this single-round active interaction capability improves the first-round recall rate and precision rate of intelligent question and answer to a certain extent, it is unable to switch and jump related topics in a timely and active manner in the subsequent question and answer process, and thus it is unable to effectively increase the user's dialogue enthusiasm and duration, and there is still a lot of room for improvement in user experience. In addition, the personalized recommendation technology used in this method is slightly single in terms of feature input. It only models the features on the user side, and does not take into account the surrounding environment information that can be obtained during the entire conversation process, such as geographic location, time, weather and other attributes, as well as the semantic information of the candidate set in the active conversation pool. The needs and preferences of modern people are more segmented and diverse. It is not enough to rely solely on partial features of users to make personalized recommendations. More features need to be input into the model to perform personalized recommendations and matching tasks for thousands of people.

[0006] These existing technologies have made progress in some aspects, but there are still some key issues that need further research and development to overcome. Summary of the invention

[0007] In order to solve the above technical problems, the present invention proposes an active question-answering interactive system based on a large language model and personalized recommendation technology, comprising:

[0008] A data processing module is used to obtain session data and pre-process it to obtain vector data that conforms to a preset storage format; the session data includes user real-time input information;

[0009] An intention recognition module is used to analyze the vector data obtained by the data processing module to determine the user's intention, and to identify the conversation type corresponding to the current conversation according to the user's intention; the conversation type includes an open domain conversation and a task-based conversation;

[0010] A vector knowledge base module to determine the content items most relevant to the user’s intent;

[0011] A large language model module is used to generate system response content based on the content items determined by the vector knowledge base module, the context information of the current conversation, and the conversation type;

[0012] The active conversation corpus pool is used to store examples of various active conversations and corresponding conversations for each preset active conversation scenario;

[0013] A personalized recommendation module is used to obtain personalized information of the user and obtain the best candidate speech in the active conversation corpus based on system monitoring information; the system monitoring information is personalized information or personalized information, context information of the current conversation, historical conversations of the user and vector data obtained by the data processing module;

[0014] The active conversation module is used to obtain the target scenario and active conversation coverage, and when the active conversation coverage does not exceed the preset threshold, initiate an active conversation with the user through the speech corresponding to each target scenario; the target scenario is a preset active conversation scenario that meets the preset trigger condition; the active conversation coverage is the ratio of the number of active conversation initiations to the total number of interactions;

[0015] The output module is used to output the optimal candidate speech obtained by the personalized recommendation module or the system response content generated by the LLM module.

[0016] Furthermore, the session data also includes real-time environment data and pre-stored static data; the static data includes basic information of the user.

[0017] Furthermore, the data processing module includes:

[0018] An input processing unit, used to obtain session data and, when the data type of the user's real-time input information is voice or picture, convert it into text data and output it;

[0019] A data collection unit, used to collect real-time environmental data and pre-stored static data;

[0020] The feature processing unit is used to pre-process the session data to obtain vector data that conforms to a preset storage format, wherein the pre-processing includes: data cleaning, feature selection, and feature conversion.

[0021] Furthermore, the data cleaning refers to removing noise and irrelevant data in the session data; the irrelevant data includes null values ​​and repeated values;

[0022] The feature selection refers to scoring each feature in the cleaned session data through a feature scoring model, and screening out features with scores greater than a preset value; the feature scoring model is a decision tree, a random forest, or a LASSO regression model;

[0023] Feature conversion refers to converting the filtered features to obtain vector data corresponding to the session data; the vector data is a word frequency vector or an embedding vector.

[0024] Furthermore, the active conversation module includes:

[0025] An active conversation scenario triggering unit, used to obtain a preset active conversation scenario that meets a preset triggering condition;

[0026] Topic transition probability prediction unit, used to predict the next topic of interest using the user's historical conversations, preferences, and contextual information of the current conversation through an interest similarity model;

[0027] An active session coverage dynamic control unit, used to calculate the active session coverage and dynamically adjust the active session coverage according to preset rules;

[0028] The active conversation material pool management unit is used to update the active conversation material pool according to the user's conversation information; the conversation information includes the context information of the current conversation.

[0029] The present invention also proposes an active question-answering interaction method based on a large language model and personalized recommendation technology, comprising:

[0030] S1: Acquire session data through a data processing module, and acquire vector data corresponding to the session data; the session data includes real-time user input information; the real-time user input information is a spontaneous question asked by the user or an answer given by the user after an active session is initiated by an active session module;

[0031] S2: Analyze the vector data and the user's historical conversations through the active conversation scenario trigger unit to determine whether there is a preset active conversation scenario that meets the preset trigger condition. If so, proceed to the next step; if not, jump to step S4;

[0032] S3; calculate the active conversation coverage through the active conversation coverage dynamic control unit, and determine whether the active conversation coverage is greater than the preset threshold. If not, initiate an active conversation with the user through the words corresponding to each target scene, and output it through the output module; if so, proceed to the next step; the target scene is a preset active conversation scene that meets the preset trigger condition;

[0033] S4: obtaining the user intent corresponding to the vector data through the intent recognition module, identifying the conversation type corresponding to the vector data according to the user intent, and determining whether the conversation type is an open domain conversation. If so, proceed to the next step; if not, jump to step S6;

[0034] S5: Determine the content items most relevant to the user's intent through the vector knowledge base module;

[0035] S6: Generate system reply content by using relevant information through the large language model module; the relevant information is the context information of the current conversation and the content item determined by the vector knowledge base module or the context information of the current conversation;

[0036] S7: Output the system reply content through the output module.

[0037] Furthermore, in the step S6:

[0038] If the dialogue type is an open domain dialogue, the large language model module uses the context information of the current dialogue and the content items determined by the vector knowledge base module to generate the system response content. If the dialogue type is a task-based dialogue, the large language model module uses the context information of the current dialogue to call the corresponding external program, and generates the system response content based on the feedback data of the external program.

[0039] Furthermore, in step S3, an active dialogue is initiated to the user through the corresponding words of each target scenario, specifically:

[0040] The personalized information of the user is obtained through the personalized recommendation module, and the speech of all target scenarios is sorted by the personalized recommendation algorithm using the system monitoring information, and the speech ranked first is obtained as the optimal candidate speech; the system monitoring information is personalized information or personalized information, context information of the current conversation, historical conversations of the user and vector data obtained by the data processing module;

[0041] Initiate active dialogue with users through the optimal candidate dialogue and output it through the output module.

[0042] Furthermore, the session data also includes real-time environment data and pre-stored static data; the static data includes basic information of the user;

[0043] The personalized recommendation algorithm is a collaborative filtering algorithm, a matrix decomposition algorithm, an extreme gradient boosting algorithm, a deep factorization machine algorithm, or a deep structured semantic model.

[0044] The present invention further proposes an electronic device, comprising: a processor, and a memory for storing a program, wherein the program comprises instructions, and when the instructions are executed by the processor, the processor executes the method described above.

[0045] The beneficial effects of the embodiments of the present invention include:

[0046] (1) In the present invention, the conversation data is obtained through the data processing module, and the vector data corresponding to the conversation data is obtained; the vector data and the user's historical conversation are analyzed through the active conversation scene trigger unit, and when there is a preset active conversation scene that meets the preset trigger condition, the active conversation coverage is calculated through the active conversation coverage dynamic control unit, and when the active conversation coverage does not exceed the preset threshold, the active conversation is initiated to the user through the speech corresponding to each target scene, and output through the output module; the present invention can realize multiple interactions of active interruption by real-time analysis of the user's real-time input information and the user's historical conversation information, effectively increasing the enthusiasm and duration of the user's conversation, and ensuring the coherence and consistency of the conversation;

[0047] (2) In the present invention, the input of the personalized recommendation module includes personalized information, context information of the current conversation, the user's historical conversations and vector data obtained by the data processing module, which solves the problem in the prior art that only the user's features are modeled, and the surrounding environment information that can be obtained during the entire conversation is not considered, which leads to the inability to achieve personalized recommendations for different people;

[0048] (3) In the present invention, the system reply content is generated by using the relevant information through the large language model module; the relevant information is the context information of the current conversation and the content items determined by the vector knowledge base module or the context information of the current conversation; that is, the present invention uses the large language model (LLM technology) to replace the traditional MDP technology for natural language understanding, and its ultra-long context memory capacity solves the shortcomings of the above-mentioned MDP, so that the intelligent question and answer can better follow the real-time status and information of the user;

[0049] (4) The present invention combines LLM with personalized recommendation technology and proactive question-and-answering, which can simultaneously meet the open domain and task-oriented conversation intentions, ensuring that users have a more natural, smooth, intimate, accurate and easy-to-understand intelligent question-and-answer experience.

[0050] The details of one or more embodiments of the invention are set forth in the following drawings and description so that other features, objects, and advantages of the invention are more readily apparent. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the prior art descriptions. Obviously, the drawings described below are only some embodiments of the present invention, and for ordinary technicians in this field, other embodiments can be obtained based on these drawings without creative work.

[0052] Figure 1 This is a flow chart of an active question-answering interaction method based on a large language model and personalized recommendation technology according to an embodiment of the present invention;

[0053] Figure 2 Schematic diagram of the structure of the electronic device of this embodiment. DETAILED DESCRIPTION

[0054] Embodiments of the present embodiment will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present embodiment are shown in the accompanying drawings, it should be understood that the present embodiment can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein, which are instead provided for a more thorough and complete understanding of the present embodiment. It should be understood that the drawings and embodiments of the present embodiment are only for exemplary purposes and are not intended to limit the scope of protection of the present embodiment.

[0055] In order to better understand the embodiments of the present application, the technical terms involved in the embodiments of the present application are explained as follows:

[0056] LLM (Large Language Model): LLM is a natural language processing model based on deep learning, containing hundreds of billions or more parameters, capable of learning language knowledge and patterns from massive text data. It can understand and generate human language and has strong semantic understanding and generation capabilities. Representative LLMs include the GPT series (such as GPT-3, ChatGPT, ChatGPT4o), BERT, GLM, BLOOM, and LLaMA models. The emergence of LLM has fundamentally changed the field of human-computer dialogue. Its emergence, ultra-long context memory, and generative capabilities have brought the fluency and richness of intelligent question-answering to a new level, and it has been widely used in translation, text generation, article summarization, intelligent customer service, education, and other fields.

[0057] Personalized recommendation: It is a technology that recommends content, goods or services that meet the needs of users based on their historical behaviors and interest preferences. It analyzes user data, explores users' potential needs, and improves the accuracy of recommendations and user satisfaction. After years of technical exploration and application, personalized recommendation algorithms have become mature. From traditional collaborative filtering and matrix decomposition algorithms to machine learning and deep learning algorithms such as XGBoost, DeepFM, and DSSM, they have been widely used in different fields.

[0058] Active conversation technology: As an important form of conversation technology, its core lies in the system or machine's ability to actively initiate conversations based on the user's context, behavior or preset rules, and to exchange information and interact with the user, rather than just responding to the user's request. By combining natural language processing and personalized recommendation technology, active conversation technology can deeply understand the user's intentions and provide accurate and personalized services.

[0059] In order to solve the problems in the prior art, an embodiment of the present invention provides an active question-answering interactive system based on a large language model and personalized recommendation technology, including:

[0060] The data processing module is used to obtain session data and pre-process it to obtain vector data that conforms to a preset storage format; the session data includes real-time user input information; the session data also includes real-time environment data and pre-stored static data; the static data includes basic user information, preference settings, etc.

[0061] The data processing module comprises:

[0062] An input processing unit, used to obtain session data, and when the data type of the user's real-time input information is voice or picture, convert it into text data and output it to the feature processing unit;

[0063] In this embodiment, no additional processing is required for text input. For voice input, the audio needs to be converted into text using voice recognition technology (ASR). For image input, the text information in the image needs to be extracted and converted into text using optical character recognition technology (OCR).

[0064] A data collection unit, used to collect real-time environmental data and pre-stored static data;

[0065] The feature processing unit is used to pre-process the session data to obtain vector data that conforms to a preset storage format, wherein the pre-processing includes: data cleaning, feature selection, and feature conversion.

[0066] The data cleaning refers to removing noise and irrelevant data in the session data; the irrelevant data includes null values ​​and repeated values;

[0067] The feature selection refers to scoring each feature in the cleaned session data through a feature scoring model, and screening out features with scores greater than a preset value; the feature scoring model is a decision tree, a random forest, or a LASSO regression model;

[0068] The pre-processing also includes:

[0069] Feature crossover refers to creating new features to capture the relationship or interaction between the selected features. This can be achieved through feature combination, crossover, nonlinear transformation, etc., which helps improve the prediction performance of the model. Specifically, for example, create a crossover feature of age and city - "age*city".

[0070] Feature conversion refers to converting the filtered features with new features to obtain vector data corresponding to the session data; the vector data is a word frequency vector or an embedding vector.

[0071] In this embodiment, each piece of text data corresponds to an embedding vector, which represents the semantic information of the entire sentence or paragraph; each piece of text data corresponds to a word frequency vector, which represents the frequency of occurrence of each word in the text data.

[0072] An intention recognition module is used to analyze the vector data obtained by the data processing module to determine the user's intention, and to identify the conversation type corresponding to the current conversation according to the user's intention; the conversation type includes an open domain conversation and a task-based conversation;

[0073] A vector knowledge base module to determine the content items most relevant to the user’s intent;

[0074] In this embodiment, once the intent recognition module determines the user's intent type, the vector knowledge base module can retrieve the most relevant information based on this information. For example, in an open domain conversation, if the intent recognition module determines that the user's intent is to chat about travel, the vector knowledge base module will search for content items related to travel.

[0075] It should be noted that the vector knowledge base (Vector DB) is a commonly used component in modern intelligent systems, especially in LLM-based applications. It can store knowledge or information in vector form. This representation makes it possible to perform efficient similarity search in large-scale data. Through the distance measurement in the vector space, the content items most relevant to the user's intention or preference can be quickly found.

[0076] A large language model module is used to generate system response content based on the content items determined by the vector knowledge base module, the context information of the current conversation, and the conversation type;

[0077] The active conversation corpus pool is used to store examples of various active conversations and corresponding conversations for each preset active conversation scenario;

[0078] A personalized recommendation module is used to obtain personalized information of the user and obtain the best candidate speech in the active conversation material pool according to the system monitoring information; the system monitoring information is personalized information or personalized information, context information of the current conversation, historical conversations of the user and vector data obtained by the data processing module; in this embodiment, the personalized information includes but is not limited to the user's gender, age, geographical location, etc.;

[0079] It should be noted that when the current conversation is the first conversation between the user and the system, the system monitoring information is personalized information. When it is not the first conversation, the system monitoring information is personalized information, context information of the current conversation, the user's historical conversations and vector data obtained by the data processing module (because it is an active conversation, when no data of "context information of the current conversation" is generated, this item is empty).

[0080] The active conversation module is used to obtain the target scenario and active conversation coverage, and when the active conversation coverage does not exceed the preset threshold, initiate an active conversation with the user through the speech corresponding to each target scenario; the target scenario is a preset active conversation scenario that meets the preset trigger condition; the active conversation coverage is the ratio of the number of active conversation initiations to the total number of interactions;

[0081] It should be noted that the preset trigger conditions can be defined based on a variety of factors such as user input, historical behavior, environmental information, etc. For example, if the user turns on the active question-and-answer interaction system at a certain time in the evening, the system may trigger a preset active conversation scene that asks the user whether he needs to adjust the indoor light. Once the preset active conversation scene is triggered, the system also needs to perform corresponding actions based on the user's feedback. For example, if the user answers "yes", the system may automatically adjust the lighting; if the user answers "no", the system will not make any changes.

[0082] The active conversation module includes:

[0083] An active conversation scenario triggering unit, used to obtain a preset active conversation scenario that meets a preset triggering condition;

[0084] The topic transfer probability prediction unit is used to predict the next topic of interest by using the user's historical conversations, preferences and context information of the current conversation through an interest similarity model. When the current topic ends (that is, the active conversation material pool management unit does not receive feedback from the user) and the active conversation coverage does not exceed a preset threshold, the predicted topic of interest is output through the output module. In this embodiment, the purpose of predicting the next topic of interest is to increase the continuity and novelty of the user conversation and make the conversation more natural and smooth.

[0085] An active session coverage dynamic control unit, used to calculate active session coverage and dynamically adjust the active session coverage according to preset rules to ensure that active sessions are neither too frequent nor too rare;

[0086] It should be noted that the preset rules include:

[0087] Time-based rules:

[0088] Adjust coverage based on the time of day, for example, users are more likely to engage in active sessions in the morning or evening. Increase active session coverage during times when users are more active, and reduce it during times when users are less active.

[0089] Rules based on user behavior:

[0090] Adjust the active session coverage based on the user's interaction frequency. For example, if a user has not actively initiated a session for a period of time, increase the frequency of active sessions appropriately. If a user frequently participates in active sessions, reduce the frequency of active sessions appropriately to avoid disturbing the user.

[0091] Rules based on user feedback:

[0092] Adjust the coverage based on the user's feedback on the active session. For example, if the user often gives positive feedback, the frequency of the active session can be increased. If the user gives negative feedback or ignores the active session, the frequency of the active session can be reduced.

[0093] The active conversation material pool management unit is used to update the active conversation material pool according to the user's conversation information; the conversation information includes the context information of the current conversation.

[0094] It should be noted that the active conversation corpus management unit will collect user feedback, which can be direct feedback from users (for example, the user explicitly expresses that he does not like a certain topic or rhetoric) or indirect feedback (for example, the user did not continue the conversation after hearing a certain topic).

[0095] Specifically, the functions of the active conversation pool management unit include:

[0096] Collect direct feedback data, indirect feedback data and system performance data: Direct feedback data: obtained through user evaluation or direct user interaction records. Indirect feedback data: infer the user's attitude towards a certain speech by monitoring the user's interaction behavior. System performance data: the overall interaction situation of the system, including the length of the session, user activity, etc.

[0097] Analyze feedback data and system performance data: Regularly analyze user feedback and system performance data to determine which scripts need improvement.

[0098] Adjust the script: Modify the script content based on the analysis results to make it more in line with user preferences and needs.

[0099] Remove ineffective scripts: Regularly review the active conversation pool and delete scripts that are ineffective (for example, scripts that are used less frequently than the preset value).

[0100] The output module is used to output the optimal candidate speech obtained by the personalized recommendation module or the system response content generated by the LLM module.

[0101] In this embodiment, the output content of the output module is multimodal. By accessing the text-to-speech technology (TTS), end-to-end voice output can be achieved, and the text-to-image model can also be accessed for image output display.

[0102] In order to explain an active question-answering interactive system based on a large language model and personalized recommendation technology in an embodiment of the present invention, Figure 1 As shown, this embodiment also proposes an active question-answering interaction method based on a large language model and personalized recommendation technology, including:

[0103] S1: Acquire session data through a data processing module, and acquire vector data corresponding to the session data; the session data includes real-time user input information; the real-time user input information is a spontaneous question asked by the user or an answer given by the user after an active session is initiated by an active session module; the session data also includes real-time environment data and pre-stored static data; the static data includes basic information of the user;

[0104] S2: Analyze the vector data and the user's historical conversations through the active conversation scenario trigger unit to determine whether there is a preset active conversation scenario that meets the preset trigger condition. If so, proceed to the next step; if not, jump to step S4;

[0105] It should be noted that there may be one or more preset active conversation scenarios that meet the preset trigger condition.

[0106] S3; calculate the active conversation coverage through the active conversation coverage dynamic control unit, and determine whether the active conversation coverage is greater than the preset threshold. If not, initiate an active conversation with the user through the words corresponding to each target scene, and output it through the output module; if so, proceed to the next step; the target scene is a preset active conversation scene that meets the preset trigger condition;

[0107] In step S3, an active dialogue is initiated to the user through the corresponding words of each target scenario, specifically:

[0108] The personalized information of the user is obtained through the personalized recommendation module, and the speech of all target scenarios is sorted by the personalized recommendation algorithm using the system monitoring information, and the speech ranked first is obtained as the optimal candidate speech; the system monitoring information is personalized information or personalized information, context information of the current conversation, historical conversations of the user and vector data obtained by the data processing module;

[0109] Initiate active dialogue with users through the optimal candidate dialogue and output it through the output module.

[0110] In another embodiment of the present invention, when the active conversation scenario triggering unit obtains multiple preset active conversation scenarios that meet the preset trigger conditions, it can randomly select a preset active conversation scenario as the target scenario, and then sort the words of the target scenario through the personalized recommendation module, and obtain the words ranked first as the optimal candidate words; this method can increase the diversity of conversations and avoid always selecting the same scenario.

[0111] The personalized recommendation algorithm is a collaborative filtering algorithm, a matrix decomposition algorithm, an extreme gradient boosting algorithm, a deep factorization machine algorithm, or a deep structured semantic model.

[0112] S4: obtaining the user intent corresponding to the vector data through the intent recognition module, identifying the conversation type corresponding to the vector data according to the user intent, and determining whether the conversation type is an open domain conversation. If so, proceed to the next step; if not, jump to step S6;

[0113] S5: Determine the content items most relevant to the user's intent through the vector knowledge base module;

[0114] S6: Generate system reply content by using relevant information through the large language model module; the relevant information is the context information of the current conversation and the content item determined by the vector knowledge base module or the context information of the current conversation;

[0115] In the step S6:

[0116] If the dialogue type is an open domain dialogue, the large language model module uses the context information of the current dialogue and the content items determined by the vector knowledge base module to generate the system response content. If the dialogue type is a task-based dialogue, the large language model module uses the context information of the current dialogue to call the corresponding external program, and generates the system response content based on the feedback data of the external program.

[0117] Open domain dialogue usually refers to small talk or dialogue without a specific purpose, while task-based dialogue refers to the user's desire to complete a specific task through dialogue, such as querying information, booking services, etc. In the intelligent question-answering system, the large language model module can simulate human dialogue, and it can interact naturally and smoothly based on the context of the dialogue and the data obtained from the vector knowledge base module. Agents can also be built in this module to call various downstream tools, i.e. external programs, such as search engines, weather queries, stock consultations, and other interfaces.

[0118] S7: Output the system reply content through the output module.

[0119] In the present invention, the conversation data is obtained through the data processing module, and the vector data corresponding to the conversation data is obtained; the vector data and the user's historical conversations are analyzed through the active conversation scenario trigger unit, and when there is a preset active conversation scenario that meets the preset trigger condition, the active conversation coverage is calculated through the active conversation coverage dynamic control unit, and when the active conversation coverage does not exceed the preset threshold, an active conversation is initiated to the user through the speech corresponding to each target scenario, and outputted through the output module; the present invention can realize multiple interactions of active interruptions by performing real-time analysis on the user's real-time input information and the user's historical conversation information, effectively increasing the enthusiasm and duration of the user's conversation, and ensuring the continuity and consistency of the conversation.

[0120] An embodiment of the present invention further provides an electronic device, comprising: at least one processor; and a memory in communication with the at least one processor. The memory stores a computer program executable by the at least one processor, and when the computer program is executed by the at least one processor, the electronic device executes the method of the embodiment of the present invention.

[0121] An embodiment of the present invention further provides a non-transitory machine-readable medium storing a computer program, wherein the computer program, when executed by a processor of a computer, is used to cause the computer to execute the method of the embodiment of the present invention.

[0122] An embodiment of the present invention further provides a computer program product, including a computer program, wherein the computer program, when executed by a processor of a computer, is used to enable the computer to execute the method of the embodiment of the present invention.

[0123] refer to Figure 2 , a block diagram of an electronic device that can be used as a server or client of an embodiment of the present invention will now be described, which is an example of a hardware device that can be applied to various aspects of the present invention. Electronic devices are intended to represent various forms of digital electronic computer devices, such as laptop computers, desktop computers, workbenches, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic devices can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present invention described and / or required herein.

[0124] like Figure 2As shown, the electronic device includes a computing unit 401, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 402 or a computer program loaded from a storage unit 408 into a random access memory (RAM) 403. In RAM 403, various programs and data required for the operation of the electronic device can also be stored. The computing unit 401, ROM 402, and RAM 403 are connected to each other via a bus 404. An input / output (I / O) interface 405 is also connected to the bus 404.

[0125] Multiple components in the electronic device are connected to the I / O interface 405, including: an input unit 406, an output unit 407, a storage unit 408, and a communication unit 409. The input unit 406 can be any type of device that can input information to the electronic device, and the input unit 406 can receive input digital or character information, and generate key signal input related to user settings and / or function control of the electronic device. The output unit 407 can be any type of device that can present information, and can include but is not limited to a display, a speaker, a video / audio output terminal, a vibrator, and / or a printer. The storage unit 408 can include but is not limited to a disk, an optical disk. The communication unit 409 allows the electronic device to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks, and can include but is not limited to a modem, a network card, an infrared communication device, a wireless communication transceiver, and / or a chipset, such as a Bluetooth device, a WiFi device, a WiMax device, a cellular communication device, and / or the like.

[0126] The computing unit 401 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, a CPU, a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 401 performs the methods and processes described above. For example, in some embodiments, the method embodiments of the present invention may be implemented as a computer program, which is tangibly contained in a machine-readable medium, such as a storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on an electronic device via ROM 402 and / or communication unit 409. In some embodiments, the computing unit 401 may be configured to perform the above-described method in any other appropriate manner (e.g., by means of firmware).

[0127] The computer programs for implementing the methods of the embodiments of the present invention may be written in any combination of one or more programming languages. These computer programs may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that when the computer programs are executed by the processor or controller, the functions / operations specified in the flow chart and / or block diagram are implemented. The computer programs may be executed entirely on the machine, partially on the machine, partially on the machine as a stand-alone software package and partially on a remote machine, or entirely on a remote machine or server.

[0128] In the context of an embodiment of the present invention, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable signal medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0129] It should be noted that the term "including" and its variations used in the embodiments of the present invention are open inclusions, that is, "including but not limited to". The term "based on" means "based at least in part on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one other embodiment"; the term "some embodiments" means "at least some embodiments". The modifications of "one" and "multiple" mentioned in the embodiments of the present invention are illustrative rather than restrictive. Those skilled in the art should understand that unless the context clearly indicates otherwise, it should be understood as "one or more".

[0130] The term "embodiment" in this specification refers to specific features, structures or characteristics described in conjunction with the embodiment that can be included in at least one embodiment of the present invention. The appearance of this phrase in various places in the specification does not necessarily mean the same embodiment, nor does it mean that it is mutually exclusive with other embodiments and is independent or optional. The various embodiments in this specification are described in a related manner, and the same or similar parts between the various embodiments refer to each other. In particular, for the device, equipment, and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts refer to the partial description of the method embodiment.

[0131] The above-mentioned embodiments only express several implementation methods of the present invention, and the description thereof is relatively specific and detailed, but it cannot be understood as limiting the scope of patent protection. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the scope of protection of the present invention. Therefore, the scope of protection of the present invention shall be subject to the attached claims.

Claims

1. An active question-answering interactive system based on a large language model and personalized recommendation technology, characterized in that: include: A data processing module is used to obtain session data and pre-process it to obtain vector data that conforms to a preset storage format; the session data includes user real-time input information; An intention recognition module is used to analyze the vector data obtained by the data processing module to determine the user's intention, and to identify the conversation type corresponding to the current conversation according to the user's intention; the conversation type includes an open domain conversation and a task-based conversation; A vector knowledge base module to determine the content items most relevant to the user’s intent; A large language model module is used to generate system response content based on the content items determined by the vector knowledge base module, the context information of the current conversation, and the conversation type; The active conversation corpus pool is used to store examples of various active conversations and corresponding conversations for each preset active conversation scenario; A personalized recommendation module is used to obtain personalized information of the user and obtain the best candidate speech in the active conversation corpus based on system monitoring information; the system monitoring information is personalized information or personalized information, context information of the current conversation, historical conversations of the user and vector data obtained by the data processing module; An active conversation module is used to obtain target scenarios and active conversation coverage, and when the active conversation coverage does not exceed a preset threshold, initiate an active conversation with the user through the speech corresponding to each target scenario; the target scenario is a preset active conversation scenario that meets a preset trigger condition; The active session coverage is the ratio of the number of active session initiations to the total number of interactions; The output module is used to output the optimal candidate speech obtained by the personalized recommendation module or the system response content generated by the LLM module.

2. According to claim 1, an active question-answering interactive system based on a large language model and personalized recommendation technology is characterized in that: The session data also includes real-time environment data and pre-stored static data; the static data includes basic information of the user.

3. The active question-answering interactive system based on a large language model and personalized recommendation technology according to claim 2, characterized in that: The data processing module comprises: An input processing unit, used to obtain session data and, when the data type of the user's real-time input information is voice or picture, convert it into text data and output it; Data collection unit, used to collect real-time environmental data and pre-stored static data; The feature processing unit is used to pre-process the session data to obtain vector data that conforms to a preset storage format, wherein the pre-processing includes: data cleaning, feature selection, and feature conversion.

4. The active question-answering interactive system based on a large language model and personalized recommendation technology according to claim 3, characterized in that: The data cleaning refers to removing noise and irrelevant data in the session data; the irrelevant data includes null values ​​and repeated values; The feature selection refers to scoring each feature in the cleaned session data through a feature scoring model, and screening out features with scores greater than a preset value; the feature scoring model is a decision tree, a random forest, or a LASSO regression model; Feature conversion refers to converting the filtered features to obtain vector data corresponding to the session data; the vector data is a word frequency vector or an embedding vector.

5. The active question-answering interactive system based on a large language model and personalized recommendation technology according to claim 4, characterized in that: The active conversation module includes: An active conversation scenario triggering unit, used to obtain a preset active conversation scenario that meets a preset triggering condition; Topic transition probability prediction unit, used to predict the next topic of interest using the user's historical conversations, preferences, and contextual information of the current conversation through an interest similarity model; An active session coverage dynamic control unit, used to calculate the active session coverage and dynamically adjust the active session coverage according to preset rules; The active conversation material pool management unit is used to update the active conversation material pool according to the user's conversation information; the conversation information includes the context information of the current conversation.

6. An active question-answering interactive method based on a large language model and personalized recommendation technology, characterized in that: include: S1: Acquire session data through a data processing module, and acquire vector data corresponding to the session data; the session data includes real-time user input information; the real-time user input information is a spontaneous question asked by the user or an answer given by the user after an active session is initiated by an active session module; S2: Analyze the vector data and the user's historical conversations through the active conversation scenario trigger unit to determine whether there is a preset active conversation scenario that meets the preset trigger condition. If so, proceed to the next step; if not, jump to step S4; S3； The active conversation coverage is calculated by the active conversation coverage dynamic control unit, and it is determined whether the active conversation coverage is greater than a preset threshold. If not, an active conversation is initiated to the user through the corresponding words of each target scenario, and the result is output through the output module; If yes, proceed to the next step; the target scenario is a preset active conversation scenario that meets the preset trigger condition; S4: obtaining the user intent corresponding to the vector data through the intent recognition module, identifying the conversation type corresponding to the vector data according to the user intent, and determining whether the conversation type is an open domain conversation. If so, proceed to the next step; if not, jump to step S6; S5: Determine the content items most relevant to the user's intent through the vector knowledge base module; S6: Generate system reply content by using relevant information through the large language model module; the relevant information is the context information of the current conversation and the content item determined by the vector knowledge base module or the context information of the current conversation; S7: Output the system reply content through the output module.

7. The active question-answering interactive method based on a large language model and personalized recommendation technology according to claim 6, characterized in that: In the step S6: If the dialogue type is an open domain dialogue, the large language model module uses the context information of the current dialogue and the content items determined by the vector knowledge base module to generate the system response content. If the dialogue type is a task-based dialogue, the large language model module uses the context information of the current dialogue to call the corresponding external program, and generates the system response content based on the feedback data of the external program.

8. The active question-answering interactive method based on a large language model and personalized recommendation technology according to claim 7, characterized in that: In step S3, an active dialogue is initiated to the user through the corresponding words of each target scenario, specifically: The personalized information of the user is obtained through the personalized recommendation module, and the speech of all target scenarios is sorted by the personalized recommendation algorithm using the system monitoring information, and the speech ranked first is obtained as the optimal candidate speech; the system monitoring information is personalized information or personalized information, context information of the current conversation, historical conversations of the user and vector data obtained by the data processing module; Initiate active dialogue with users through the optimal candidate dialogue and output it through the output module.

9. The active question-answering interactive method based on a large language model and personalized recommendation technology according to claim 8, characterized in that: The session data also includes real-time environment data and pre-stored static data; the static data includes basic information of the user; The personalized recommendation algorithm is a collaborative filtering algorithm, a matrix decomposition algorithm, an extreme gradient boosting algorithm, a deep factorization machine algorithm, or a deep structured semantic model.

10. An electronic device, comprising: A processor, and a memory storing a program, wherein the program comprises instructions, which, when executed by the processor, cause the processor to perform the method according to any one of claims 6 or 9.

Images