Intelligent social chatting method and system based on large model, and server

Abstract

The invention discloses an intelligent social chat method based on a large model, and the method comprises the following steps: 1, collecting chat data, preference data and feedback data of a user, and forming training data; 2, inputting the training data into a recurrent neural network machine learning model for training to obtain a chat assistant model; 3, after the user receives the chat content, the chat assistant model generates a plurality of candidate reply contents according to the chat content and the context; and step 4, after the user selects the corresponding reply content, adding the chat content and the selected reply content to the chat data. Through the deep learning and natural language processing technology, the language habit and emotional state of the user can be analyzed in real time, so that the system can more accurately grasp the intention and demand of the user, thereby generating a reply better meeting the expectation of the user, improving the chatting efficiency, and enhancing the emotional connection between the user and the social contact object.

Description

Technical Field

[0001] The present application belongs to the field of social software, and in particular relates to an intelligent social chat method, system and server based on a large model. Background Art

[0002] With the popularity of social software, users have higher and higher requirements for chat experience. Traditional chat systems can only provide basic chat functions and are increasingly unable to meet users' diverse chat needs. Therefore, intelligent chat assistant systems have also emerged.

[0003] Intelligent chat assistant systems do have extremely important value and broad application prospects in the modern social environment. With the rapid development of big data and artificial intelligence technology, such systems can more deeply understand users' emotions, habits and other characteristics, and generate personalized replies, greatly improving the chat experience.

[0004] However, existing chat systems are all aimed at dialogue between people and systems, focusing on the human-computer interaction of the chat system, so that the system obtains different attributes of different users to make different responses to improve the experience of human-computer interaction. For the actual user environment, there are the following shortcomings:

[0005] It mainly focuses on providing information answers and basic interactive experience, but fails to effectively help users improve their social level;

[0006] It cannot deeply understand and analyze the user's emotions, habits and other characteristics, and cannot effectively generate prompts for user needs to help users communicate more effectively;

[0007] Existing systems all make one-time responses to questions, and fail to provide users with an effective candidate prompt sentence that conforms to user characteristics and habits in social scenarios.

[0008] Therefore, a new type of intelligent social chat method is needed to solve the above problems. Summary of the invention

[0009] In order to address the deficiencies of the prior art, the present application provides an intelligent social chat method, system and server based on a large model, which can analyze the user's language habits and emotional state in real time through deep learning and natural language processing technology, so that the system can more accurately grasp the user's intentions and needs, thereby generating replies that are more in line with user expectations, improving the efficiency of the chat, and enhancing the emotional connection between the user and the social object.

[0010] The technical effects to be achieved by this application are achieved through the following solutions:

[0011] According to a first aspect of the present application, there is provided a large model-based intelligent social chat method, comprising the following steps:

[0012] Step 1: Collect users’ chat data, preference data, and feedback data to form training data;

[0013] Step 2: Input the training data into a recursive neural network machine learning model for training to obtain a chat assistant model;

[0014] Step 3: When the user receives the chat content, the chat assistant model generates several candidate reply contents according to the chat content and context;

[0015] Step 4: After the user selects the corresponding reply content, the chat content and the selected reply content are added to the chat data.

[0016] Preferably, in step 1, the chat data includes current chat content, historical chat records and user information, and the chat data is pre-processed by natural language processing technology and then included in the training data.

[0017] Preferably, the preference data at least includes user interests and hobbies, areas of concern and current hot topics. The user interests and hobbies are obtained through the user's browsing history, search history and shopping behavior; the areas of concern are obtained through the user's follow objects, discussion groups and subscribed content on social media; the hot topics are obtained based on the interests and hobbies and the areas of concern.

[0018] Preferably, the feedback data is obtained by collecting feedback information from users, and the feedback information includes scores on the accuracy, relevance and interest of candidate reply contents.

[0019] Preferably, in step 2, the recursive neural network machine learning model includes an input layer, a hidden layer and an output layer,

[0020] The input layer is used to receive the training data and pass the training data to the hidden layer;

[0021] The hidden layer is used to encode the training data and model the time correlation so as to understand the contextual relationship and obtain the reply content;

[0022] The output layer is used to output the reply content.

[0023] Preferably, the recursive neural network machine learning model adopts the following algorithm:

[0024] ht=tanh(Whxxt+Whhht-1+bh)

[0025] Among them, W hx is the weight matrix input to the hidden layer, W hh is the weight matrix from hidden layer to hidden layer, b h is the bias vector of the hidden layer, h t is the representation of the hidden state at time step t, and the hidden state is the context information of the chat content; x t is the representation of the input at time step t, where the input is the chat content; tanh is the hyperbolic tangent function.

[0026] Preferably, in step 3, the chat content is input into the chat assistant model and then forward propagated. The chat assistant model receives the output of the unit at the previous moment and the input at the current moment, and generates the output at the current moment to form a number of candidate reply contents.

[0027] Preferably, in step 4, the user replies by himself or selects the reply content to reply, and after replying, the historical chat record and the current chat content are updated, and step 2 is repeated to update the chat assistant model.

[0028] According to a second aspect of the present application, there is provided an intelligent social chat system using the above-mentioned large model-based intelligent social chat method, including an information collection module, a chat assistant module and a feedback update module, wherein:

[0029] The information collection module is used to collect users' chat data, preference data and feedback data, and generate training data;

[0030] The chat assistant module is used to train according to the training data and generate a number of reply contents according to the chat content;

[0031] The feedback update module is used to update the chat data according to the user's reply and collect the user's feedback information, so as to optimize the chat assistant module.

[0032] According to a third aspect of the present application, there is provided a server, comprising: a memory and at least one processor;

[0033] The memory stores a computer program, and the at least one processor executes the computer program stored in the memory to implement the above-mentioned large model-based intelligent social chatting method.

[0034] According to an embodiment of the present application, the beneficial effects of using the intelligent social chat method based on the large model are: providing users with an efficient, convenient and personalized communication method, bringing users a more fluent and satisfactory communication experience, and assisting users to improve communication efficiency; being able to understand the user's emotions and feelings according to the context, helping users to complete social interaction, and generating replies that match the user's emotional state; and increasing the data dimensions of users' interests and hot topics. This is of great significance for improving the naturalness and richness of social interaction. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] In order to more clearly illustrate the embodiments of the present application or the existing technical solutions, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0036] Figure 1 This is a flow chart of an intelligent social chat method based on a large model in one embodiment of the present application;

[0037] Figure 2 This is a structural block diagram of an intelligent social chat system based on a large model in one embodiment of the present application;

[0038] Figure 3 This is a structural block diagram of a server in an embodiment of the present application. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present application clearer, the technical solution of the present application will be clearly and completely described below in conjunction with specific embodiments and corresponding drawings. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0040] The intelligent social chat method based on the big model includes the following steps:

[0041] Step 1: Collect users’ chat data, preference data, and feedback data to form training data;

[0042] In this step, the chat data includes the current chat content, historical chat records and user information. The chat data is pre-processed by natural language processing technology and then included in the training data.

[0043] Users chat with friends through social tools and agree to use the social content as training data for their intelligent chat assistants. The chat assistant system first starts to collect the user's historical chat records. This includes the previous conversation content, the context of the user's chat, the user's personal information and preferences, corpus, and sentiment tags. These data are used to train machine learning models. By collecting and analyzing these historical chat records and information, the chat assistant system can more accurately understand the user's needs and intentions, thereby providing smarter and more personalized services.

[0044] The chat assistant system uses natural language processing (NLP) technology, including word segmentation, part-of-speech tagging, entity recognition, grammatical analysis, etc., to understand the specific meaning and intention of user input. It analyzes contextual information, identifies keywords, phrases, and patterns to infer user emotions and needs, such as:

[0045] User input: "Today's weather is great, the sun is shining, and my mood is getting better!"

[0046] The system identifies keywords: "great", "sunny", "improved mood". These keywords are usually associated with positive emotions. Therefore, the system determines that the user's emotion is "positive / happy".

[0047] Another example: User input: "Every meeting goes overtime, it's really annoying!"

[0048] The system matches the fixed expression pattern of "I'm so annoyed", which is often used in the Chinese context to express strong boredom or dissatisfaction. Although the previous sentence "Every meeting is overtime" provides background information, the pattern of "I'm so annoyed" is strong enough to express the user's emotion. Therefore, the system determines that the user's emotion is "negative / annoyed".

[0049] Through the understanding and analysis of historical data, the processed data will be used as training data.

[0050] Preference data at least includes user interests, areas of interest, and current hot topics. User interests are obtained through user browsing history, search history, and shopping behavior; areas of interest are obtained through users’ followers, discussion groups, and subscribed content on social media; hot topics are obtained based on interests and areas of interest. For these areas, the system collects relevant expertise, industry trends, and the latest research results. Current hot topic event data is also added to the system. Training data is enriched through user interests and hot event data.

[0051] Feedback data is obtained by collecting user feedback information, which includes scores on the accuracy, relevance, and interestingness of candidate responses. By collecting user feedback information, the system can understand user satisfaction with candidate responses and identify existing problems. The system will conduct in-depth analysis and processing of this data. By mining key information and patterns in user feedback, the system can identify existing problems and deficiencies, and adjust algorithm parameters, optimize model structure, or improve response strategies accordingly.

[0052] During the use process, users found some areas where the system did not meet their needs, or had some new functional requirements. These suggestions are of great guiding significance for the optimization and upgrade of the system. The system will actively collect and analyze these suggestions and incorporate them into the subsequent improvement plan.

[0053] Step 2: Input the training data into the recursive neural network machine learning model for training to obtain the chat assistant model;

[0054] In the training process of this step, a large amount of labeled training data is input for training, the training data is cleaned, word segmented, and the text is converted into a numerical vector. The mapping relationship between user input and intent is automatically learned, so that the long-term dependencies and complex semantic information in the sentence can be captured, so as to more accurately understand the user's true intent.

[0055] By using the recurrent neural network (RNN) machine learning model, we can refer to data such as the user's historical questions, the system's previous responses, the topic and direction of the conversation, etc., to accurately understand the user's needs and provide appropriate candidate responses.

[0056] Step 3: When the user receives the chat content, the chat assistant model generates several candidate replies based on the chat content and context;

[0057] The chat assistant model in this step uses intent recognition technology to determine the purpose or intent of the chat content so that the system can correctly understand the user's needs and provide the user with candidate chat replies. Intent recognition uses a machine learning model, recursive neural network (RNN). RNN has continuity and context awareness when processing conversations, and can better understand and respond to user utterances. Compared with other algorithms, it can effectively capture this sequence dependency and enhance the coherence and comprehension of the conversation. Users may ask complex questions or have continuous conversations. RNN has the ability to remember and apply previous information to current input data, so it can handle longer conversation sequences.

[0058] The recurrent neural network machine learning model uses the following algorithm:

[0059] h t =tanh(Whx x t +W hh h t-1 +b h )

[0060] Among them, W hx is the weight matrix input to the hidden layer, W hh is the weight matrix from hidden layer to hidden layer, b h is the bias vector of the hidden layer;

[0061] h t is the representation of the hidden state at time step t. The hidden state is the core of RNN, which captures the sequence information up to the current time step. In RNN, the hidden state is passed from the previous time step to the next time step in a recursive way, allowing the model to keep the memory of past information; the hidden state is the context information of the chat conversation, including the content of the previous conversation, the user's intention, emotion, etc.;

[0062] x t is the representation of the input at time step t. In the context of intelligent social chat applications, x t Represents the text or voice data entered by the user at time step t (after appropriate preprocessing and encoding), and also includes the user's metadata such as age, gender, geographic location, etc.

[0063] Tanh is the hyperbolic tangent function, and its mathematical expression is:

[0064] tanh(x)=ex+e-xex-ex

[0065] The hyperbolic tangent function is an activation function that maps any real value to between -1 and 1. In RNN and other types of neural networks, activation functions are used to introduce nonlinearity, enabling the network to learn and represent complex patterns and helping to maintain the dynamic range of the hidden state within a controllable range.

[0066] The recursive neural network machine learning model includes an input layer, a hidden layer and an output layer, wherein: the input layer is used to receive training data and pass the training data to the hidden layer; the hidden layer is used to encode the training data and model the time correlation to understand the contextual relationship and obtain the response content; the output layer is used to output the response content.

[0067] Each node in the hidden layer (called a gate unit) is a nonlinear function that encodes the input and models the temporal correlation. The input data (i.e., the chat content) is used for forward propagation through the RNN model. At each moment, the RNN receives the output of the unit at the previous moment (the last reply to the other party in the social chat) and the input at the current moment (the content entered by the user when typing on the keyboard, and the input content before sending), and generates the output at the current moment (the reply content). This process will continue until the entire input sequence is processed. The input sequence refers to the entire time sequence from the user typing on the keyboard to clicking to send.

[0068] Then, the backpropagation through time (BPTT) algorithm is used to calculate the gradient. Since the user input content has a sequence, the backpropagation algorithm can ensure the consistency and correctness of the sequence of user data. BPTT is a time backpropagation algorithm used for training in RNN, which takes into account the temporal dependency, that is, the sequence of chat conversation content (continuous conversation content).

[0069] Finally, candidate replies are generated. Replies that match the context are generated, and candidate sentences of possible replies are automatically generated for users to choose based on the user's chat content and context.

[0070] Step 4: After the user selects the corresponding reply content, the chat content and the selected reply content are added to the chat data.

[0071] In this step, the user replies by himself or chooses one of the above replies. After the user replies to the other party's message, the new conversation content is seamlessly added to the user's chat history to ensure the continuity and integrity of the conversation. Update the current training data, enrich the user's emotional state and needs, etc., and retrain the chat assistant model to make it more in line with the user's real needs.

[0072] like Figure 2 As shown, in one embodiment of the present application, an intelligent social chat system using the above-mentioned large model-based intelligent social chat method includes an information collection module, a chat assistant module and a feedback update module, wherein:

[0073] Information collection module, used to collect users' chat data, preference data and feedback data, and generate training data;

[0074] The chat assistant module is used to train according to the training data and generate a number of reply contents according to the chat content;

[0075] The feedback update module is used to update the chat data according to user replies and collect user feedback information, so as to optimize the chat assistant module.

[0076] like Figure 3 As shown, a server in an embodiment of the present application includes: a memory 301 and at least one processor 302;

[0077] The memory 301 stores a computer program, and the at least one processor 302 executes the computer program stored in the memory 301 to implement the above-mentioned large model-based intelligent social chat method.

[0078] The beneficial effects of the above method are: assisting users to improve communication efficiency: providing users with efficient, convenient and personalized communication methods, and bringing users a smoother and more satisfactory communication experience. In addition to meeting personal chat needs, it can also expand the application market, provide users with personalized services, cope with complex scenarios, and enhance user experience. It can be applied to assisting manual customer service, online education, medical consultation, e-commerce shopping guides and other scenarios. In these scenarios, intelligent chat assistants can help users quickly find answers, solve problems or complete purchases, thereby improving overall service quality and efficiency.

[0079] The system understands the user's emotions and feelings based on the context and helps the user complete social interaction: it can generate replies that match the user's emotional state. This resonance can make the user feel understood and cared for, thereby enhancing the emotional connection in social interactions. When the user expresses negative emotions, the system can actively provide emotional support. For example, by sending comforting words, sharing relevant resources or suggesting professional help, it can help users relieve stress, reduce anxiety or deal with other emotional problems. Recommend relevant social activities, groups or topics based on the user's preferences to help users connect with like-minded people and enrich the user's social life. It helps the system better grasp the tone and intention of the conversation and reduce the occurrence of misunderstandings and conflicts. When the system detects potential conflict points, it can intervene in time by providing neutral points of view, suggesting that both parties think calmly or guiding the conversation in a positive direction.

[0080] Added data dimensions of users' interests and hot topics: This is of great significance in improving the naturalness and richness of social interactions. By sharing and discussing interests and hot topics, users can more easily find common interests with other users. This helps to establish closer connections and friendships and promote the in-depth development of social relationships.

[0081] It should be noted that the above detailed descriptions are exemplary and are intended to provide further explanation of the present application. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the art to which the present application belongs.

[0082] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present application. As used herein, unless the context clearly indicates otherwise, the singular form is also intended to include the plural form. In addition, it should also be understood that when the terms "comprise" and / or "include" are used in this specification, it indicates the presence of features, steps, operations, devices, components and / or combinations thereof.

[0083] It should be noted that the terms "first", "second", etc. in the specification and claims of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein.

[0084] In addition, the terms "include" and "have" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product, or apparatus that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products, or apparatuses.

[0085] For ease of description, spatially relative terms, such as "above", "above", "on the upper surface of", "above", etc., may be used herein to describe the spatial positional relationship between a device or feature and other devices or features as shown in the figure. It should be understood that spatially relative terms are intended to include different orientations of the device in use or operation in addition to the orientation described in the figure. For example, if the device in the accompanying drawings is inverted, the device described as "above other devices or structures" or "above other devices or structures" will be positioned as "below other devices or structures" or "below other devices or structures". Thus, the exemplary term "above" may include both "above" and "below". The device may also be positioned in other different ways, such as rotated 90 degrees or in other orientations, and the spatially relative descriptions used herein are interpreted accordingly.

[0086] In the above detailed description, reference is made to the accompanying drawings, which form a part of this document. In the accompanying drawings, similar symbols typically identify similar components unless the context indicates otherwise. The illustrated embodiments described in the detailed description, drawings, and claims are not meant to be limiting. Other embodiments may be used, and other changes may be made, without departing from the spirit or scope of the subject matter presented herein.

[0087] The above are only preferred 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 replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent social chat method based on a large model, characterized in that: The steps include: Step 1: Collect users’ chat data, preference data, and feedback data to form training data; Step 2: Input the training data into a recursive neural network machine learning model for training to obtain a chat assistant model; Step 3: When the user receives the chat content, the chat assistant model generates several candidate reply contents according to the chat content and context; Step 4: After the user selects the corresponding reply content, the chat content and the selected reply content are added to the chat data.

2. The intelligent social chat method based on a large model according to claim 1 is characterized in that: In step 1, the chat data includes current chat content, historical chat records and user information. The chat data is pre-processed by natural language processing technology and then included in the training data.

3. The intelligent social chat method based on a large model according to claim 1 is characterized in that: The preference data at least includes user interests and hobbies, areas of concern and current hot topics. The user interests and hobbies are obtained through the user's browsing history, search history and shopping behavior; the areas of concern are obtained through the user's follow objects, discussion groups and subscribed content on social media; the hot topics are obtained based on the interests and hobbies and the areas of concern.

4. The intelligent social chat method based on a large model according to claim 1 is characterized in that: The feedback data is obtained by collecting feedback information from users, and the feedback information includes scores on the accuracy, relevance, and interest of candidate reply contents.

5. The large model-based intelligent social chat method according to claim 1, characterized in that: In step 2, the recursive neural network machine learning model includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the training data and pass the training data to the hidden layer; The hidden layer is used to encode the training data and model the time correlation so as to understand the contextual relationship and obtain the reply content; The output layer is used to output the reply content.

6. The large model-based intelligent social chat method according to claim 5, characterized in that: The recursive neural network machine learning model uses the following algorithm: h t =tanh(W hx x t +W hh h t-1 +b h ) Among them, W hx is the weight matrix input to the hidden layer, W hh is the weight matrix from hidden layer to hidden layer, b h is the bias vector of the hidden layer, h t is the representation of the hidden state at time step t, and the hidden state is the context information of the chat content; x t is the representation of the input at time step t, where the input is the chat content; tanh is the hyperbolic tangent function.

7. The large model-based intelligent social chat method according to claim 6, characterized in that: In step 3, the chat content is input into the chat assistant model and then forward propagated. The chat assistant model receives the output of the unit at the previous moment and the input at the current moment, and generates the output at the current moment to form several candidate reply contents.

8. The large model-based intelligent social chat method according to claim 2, characterized in that: In step 4, the user replies by himself or selects the reply content to reply, and after replying, the historical chat record and current chat content are updated, and step 2 is repeated to update the chat assistant model.

9. An intelligent social chat system using the large model-based intelligent social chat method according to any one of claims 1 to 8, characterized in that: It includes information collection module, chat assistant module and feedback update module, among which: The information collection module is used to collect users' chat data, preference data and feedback data, and generate training data; The chat assistant module is used to train according to the training data and generate a number of reply contents according to the chat content; The feedback update module is used to update the chat data according to user replies and collect user feedback information, so as to optimize the chat assistant module.

10. A server, characterized in that: include: memory and at least one processor; The memory stores a computer program, and the at least one processor executes the computer program stored in the memory to implement the large model-based intelligent social chat method as described in any one of claims 1 to 8.

Images