Large model-based training method and related device

Abstract

The embodiment of the invention relates to the field of data processing, and provides a training method based on a large model and a related device, and the method comprises the steps: carrying out the interaction of an evaluation stage with a user through the large model, obtaining evaluation interaction data, and carrying out the real-time evaluation analysis based on the evaluation interaction data, so as to obtain a personalized portrait of the user; generating a training scheme according to the personalized portrait; and carrying out interaction in a training stage with the user through the large model based on the training scheme, so that the user completes the training scheme. According to the embodiment of the invention, interactive intelligent guidance can be carried out through a large model, so that more real multi-dimensional assessment interaction data for assessing the psychology of the user is obtained, more accurate psychological characteristics are obtained, a targeted training scheme is conveniently designed for the user, and the user experience is improved. The psychological ability such as psychological toughness, emotion management and cognitive level of the user can be improved, and better experience is brought to the user.

Description

Technical Field

[0001] The embodiments of the present application relate to the field of data processing, and more specifically to a large model-based training method and related devices. Background Art

[0002] In recent years, the field of psychological ability training has received increasing attention. The improvement of psychological abilities, such as emotion regulation, stress management and cognitive ability optimization, is of great significance to personal health development and career achievement. However, traditional psychological ability training methods have obvious limitations. Traditional psychological ability training usually relies on the following technical means:

[0003] Standardized questionnaires: such as MMPI (Minnesota Multiphasic Personality Inventory) and 16PF (Cattell 16 Personality Factor Questionnaire), etc. are used for quantitative assessment of psychological state.

[0004] Modular training tools: preset static training content, covering stress management, emotion regulation and social skills training modules.

[0005] AI-assisted tools: Artificial intelligence systems for emotion recognition and behavior analysis have been introduced in recent years for simple interactions and feedback.

[0006] Although existing psychological ability training tools have achieved certain results, they are still insufficient in achieving personalization, interactivity and long-term adaptability. Summary of the invention

[0007] The embodiment of the present application provides a training method and related devices based on a large model, which can perform interactive intelligent guidance through the large model, so as to obtain more realistic and multi-dimensional evaluation interaction data for evaluating user psychology and obtain more accurate psychological characteristics, thereby facilitating the design of targeted training programs for users and bringing better experience to users.

[0008] In a first aspect, an embodiment of the present application provides a training method based on a large model, the method comprising:

[0009] The evaluation interaction data is obtained by interacting with the user in the evaluation phase, and real-time evaluation and analysis are performed based on the evaluation interaction data to obtain a personalized portrait of the user; the data modality of the evaluation interaction data includes at least one of the following: text data, voice data, image data and physiological data; the personalized portrait includes at least the multidimensional psychological state evaluation result and the multidimensional analysis result of psychological ability of the user; the multidimensional psychological state evaluation result includes at least one of the following: the emotional stability, stress tolerance and cognitive state of the user; the multidimensional analysis result of psychological ability includes at least the shortcomings of the user in different psychological ability dimensions;

[0010] Generate a training plan based on the personalized portrait; the training goal of the training plan is at least used to indicate how to improve the user's shortcomings in one psychological ability dimension;

[0011] The large model interacts with the user in a training phase based on the training solution so that the user can complete the training solution.

[0012] In a second aspect, an embodiment of the present application provides a large model-based training system, which has the function of implementing the large model-based training method provided in the first aspect above. The function can be implemented by hardware, or by hardware executing corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules can be software and / or hardware.

[0013] In one embodiment, the large model-based training system includes:

[0014] An interaction module is configured to interact with the user in the evaluation phase through the large model to obtain evaluation interaction data; the data modality of the evaluation interaction data includes at least one of the following: text data, voice data, image data and physiological data;

[0015] An evaluation module is configured to perform real-time evaluation and analysis based on the evaluation interaction data to obtain a personalized portrait of the user; the personalized portrait at least includes a multidimensional psychological state evaluation result and a multidimensional psychological ability analysis result of the user; the multidimensional psychological state evaluation result includes at least one of the following: the user's emotional stability, stress tolerance, and cognitive state; the multidimensional psychological ability analysis result includes at least the user's shortcomings in different psychological ability dimensions;

[0016] The training module is configured to generate a training plan according to the personalized portrait; and to interact with the user in the training phase based on the training plan through the big model so that the user can complete the training plan.

[0017] In a third aspect, an embodiment of the present application provides a computer-readable storage medium, which includes instructions, which, when executed on a computer, enables the computer to execute the large model-based training method as described in the first aspect.

[0018] In a fourth aspect, an embodiment of the present application provides a computing device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the large model-based training method described in the first aspect is implemented.

[0019] In a fifth aspect, an embodiment of the present application provides a chip, which includes a processor coupled to a transceiver, and is used to execute the technical solution provided in the first aspect of the embodiment of the present application.

[0020] In one possible design, the chip can also be a dedicated hardware structure for implementing the technical solution provided in the first aspect above. For example, processing involving a neural network model can be implemented by a dedicated neural network processor or a graphics processor.

[0021] In a sixth aspect, an embodiment of the present application provides a chip system, which includes a processor for implementing the functions involved in the first aspect above, for example, generating or processing the information involved in the large model-based training method provided in the first aspect above.

[0022] In a possible design, the chip system also includes a memory, which is connected to the processor through a circuit structure. The memory is used to store program instructions and data required by the terminal. The chip system can be composed of chips, or it can include chips and other discrete devices. Further optionally, the chip also includes a communication interface, and the processor is connected to the communication interface. The communication interface is used to receive data and / or information to be processed, the processor obtains the data and / or information from the communication interface, processes the data and / or information, and outputs the processing results through the communication interface. The communication interface can be an input and output interface.

[0023] In a seventh aspect, an embodiment of the present application provides a computer program product comprising instructions, which, when executed on a computer, enables the computer to execute the large model-based training method provided in the first aspect.

[0024] Compared with the prior art, in the embodiment of the present application, the evaluation interaction data is obtained by interacting with the user in the evaluation phase through the large model, and real-time evaluation analysis is performed based on the evaluation interaction data to obtain the personalized portrait of the user; a training program is generated based on the personalized portrait; and the large model interacts with the user in the training phase based on the training program so that the user can complete the training program. It can be seen that the embodiment of the present application can use the large model for interactive intelligent guidance, thereby obtaining more real and multi-dimensional evaluation interaction data for evaluating the user's psychology, and obtaining more accurate psychological characteristics, so as to facilitate the design of targeted training programs for users, which can help improve the user's psychological resilience, emotional management, cognitive level and other psychological abilities, and bring a better experience to the user. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] By reading the detailed description of the embodiments of the present application with reference to the accompanying drawings, the objectives, features and advantages of the embodiments of the present application will become easy to understand. Among them:

[0026] Figure 1 A schematic diagram of a system interaction scenario of a large model-based training method in an embodiment of the present application;

[0027] Figure 2 A schematic diagram of a flow chart of a large model-based training method according to an embodiment of the present application;

[0028] Figure 3 A schematic diagram of the structure of a large model-based training system according to an embodiment of the present application;

[0029] Figure 4 This is another structural schematic diagram of a large model-based training system according to an embodiment of the present application;

[0030] Figure 5 A schematic diagram of a structure of a computing device according to an embodiment of the present application;

[0031] Figure 6 A schematic diagram of the structure of a mobile phone in an embodiment of the present application;

[0032] Figure 7 A schematic diagram of the structure of a server in an embodiment of the present application.

[0033] In the drawings, the same or corresponding reference numerals represent the same or corresponding parts. DETAILED DESCRIPTION

[0034] The terms "first", "second", etc. in the specification and claims of the embodiments 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 data used in this way can be interchanged where appropriate, so that the embodiments described here can be implemented in an order other than the content illustrated or described here. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device including a series of steps or modules need not be limited to those steps or modules clearly listed, but may include other steps or modules that are not clearly listed or inherent to these processes, methods, products or devices. The division of the modules that appear in the embodiments of the present application is only a logical division. There may be other division methods when implemented in practical applications, such as multiple modules can be combined into or integrated in another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling between modules, and the communication connection can be electrical or other similar forms, which are not limited in the embodiments of the present application. Furthermore, the modules or sub-modules described as separate components may or may not be physically separated, may or may not be physical modules, or may be distributed in multiple circuit modules, and some or all of the modules may be selected according to actual needs to achieve the purpose of the embodiment of the present application.

[0035] The embodiments of the present application also provide a big model-based training method and related devices, which can be applied to a big model-based training system in a user's psychological ability training scenario. The big model-based training system may include a terminal device and a server that deploys the big model.

[0036] In some implementations, the terminal device and the server deploying the large model are deployed separately, referring to Figure 1 The training method based on a large model provided in the embodiment of the present application can be based on Figure 1 The system shown is implemented as follows. The system may include a server 01 and a terminal device 02.

[0037] The server 01 may be a server for deploying a large model.

[0038] The terminal device 02 can be a mobile phone or a computer.

[0039] The terminal device 02 may forward the received user interaction input data to the server 01. The server 01 may generate interaction output data based on the user input data, and then send it to the terminal device 02. The terminal device 02 may output the interaction output data to the user, and receive further feedback input data from the user, thereby completing the interaction.

[0040] It should be noted that the computing device involved in the embodiments of the present application may be a server and / or a terminal device.

[0041] The server involved in the embodiments of the present application can be an independent physical server, a server cluster or a distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, as well as big data and artificial intelligence platforms.

[0042] The terminal device involved in the embodiments of the present application may be a device that provides voice and / or data connectivity to a user, a handheld device with a wireless connection function, or other processing devices connected to a wireless modem. For example, a mobile phone (or "cellular" phone) and a computer with a mobile terminal, for example, a portable, pocket-sized, handheld, computer-built-in or vehicle-mounted mobile device, which exchanges voice and / or data with a wireless access network. For example, a personal communication service (English full name: Personal Communication Service, English abbreviation: PCS) phone, a cordless phone, a session initiation protocol (SIP) phone, a wireless local loop (Wireless Local Loop, English abbreviation: WLL) station, a personal digital assistant (English full name: Personal Digital Assistant, English abbreviation: PDA) and other devices.

[0043] Reference Figure 2 , Figure 2 A flowchart of a training method based on a large model provided in an embodiment of the present application. The method can be executed by a training system based on a large model, can be applied to a user's psychological ability training scenario, interacts with the user through a terminal device, and completes the establishment of a user's personalized portrait and psychological ability training through interaction. The method includes steps 101-103:

[0044] Step 101, the big model interacts with the user in the evaluation phase to obtain evaluation interaction data, and performs real-time evaluation analysis based on the evaluation interaction data to obtain a personalized portrait of the user.

[0045] In the embodiments of the present application, a large model refers to a machine learning model with large-scale parameters and complex computing structure. A large model can be constructed by a deep neural network, with billions or even hundreds of billions of parameters, requiring a large amount of data and computing resources for training and reasoning. Typical large models are, for example, GPT.

[0046] The embodiment of the present application uses a large model to interact with the user, so that during the interaction process, a dialogue is conducted based on the interaction situation, user information is obtained, and a psychological assessment is completed, so as to formulate a targeted training plan that fits the user's own situation, thereby completing psychological ability training more efficiently. Since the large model has strong understanding and analysis capabilities and content generation capabilities, through the interaction with the user through the large model, a more humane response or guidance can be made based on the user's feedback during the interaction process, and information acquisition and subsequent psychological ability training can be completed in an interactive, guided, and motivational manner that is more emotionally caring and similar to that of a professional psychologist.

[0047] Considering that the communication between natural persons, such as between a psychological counselor and a user, is not mechanical or fixed, but gradually deepens with the communication between the two. Therefore, in the embodiment of the present application, in order to obtain user information in an anthropomorphic manner and evaluate the user's psychology, the interaction between the big model and the user during the evaluation process can be carried out in stages, for example, including the basic information acquisition stage, the psychological evaluation stage, and the ability evaluation stage. Specifically, the interaction between the big model and the user in the evaluation stage to obtain the evaluation interaction data includes the following stages ac:

[0048] Phase a: interacting with the user in the basic information acquisition phase to acquire the user's basic information.

[0049] It is considered that the inherent information of the user itself may have a certain impact on the subsequent psychological assessment and interaction content. For example, if the user is a student, the source of his stress may be mainly study pressure, so in the subsequent conversation, the main communication is on issues related to study pressure. Therefore, in order to facilitate the interaction in the psychological assessment stage and the interaction in the ability assessment stage. In an embodiment of the present application, the interaction in stage a can occur in the initial stage of the interaction between the user and the large model. Specifically, if a user interacts with the large model through a terminal device, it can first be determined whether the user has basic information on historical records, that is, whether the user is a new user. If so, it is necessary to obtain basic information.

[0050] It is understandable that the basic information may include multiple different types of information. For example, in an embodiment of the present application, the basic information includes at least one of the following information of the user: age, gender, occupation, education and living status. Considering that the communication between two people under natural circumstances is often in the form of alternating dialogues, for example, in a psychological counseling scenario, it is often the psychological counselor who asks questions and the user who answers. Therefore, in order to simulate natural real dialogues and bring users a more practical experience, in an embodiment of the present application, the interaction between the large model and the user at each stage includes multiple rounds of dialogue, and each round of dialogue is completed alternately by the large model and the user.

[0051] For example, in stage a, the large model interacts with the user in the basic information acquisition stage by generating an interactive exploration dialogue, so as to obtain the basic information of the user. In the embodiment of the present application, the interactive exploration dialogue refers to the interaction of the large model based on the basic information to be obtained and the dialogue of the user. For example, the user first outputs the content of the dialogue: "I am a college student and I am a little troubled recently." The large model can understand based on the content of the dialogue, obtain the occupation (primary school student) as basic information, and generate an interactive exploration dialogue based on the key concepts (primary school student and trouble): "Oh, that's really distressing, let me guess, are you having some difficulties in learning?". As a result, the large model realizes an interactive exploration dialogue by generating further exploration dialogue content based on the user's discourse. In the embodiment of the present application, the large model and the user can interact through dialogue to facilitate the large model to obtain the basic information of the user, and the large model can interact by generating an interactive exploration dialogue during the interaction process.

[0052] Phase b: interacting with the user in the psychological assessment phase to obtain psychological assessment interaction data of the user.

[0053] After completing the basic information collection of the user, a psychological assessment phase can be performed to assess the user's psychological state, thereby facilitating identification of the user's current psychological state. Specifically, in the embodiment of the present application, the large model interacts with the user in the psychological assessment phase by generating open-ended questions to obtain the user's psychological assessment interaction data.

[0054] For example, in the basic information acquisition stage, the user expressed that he had a heavy workload recently, often worked overtime until late, felt very stressed, and his sleep quality was affected. Then, in the psychological assessment stage, the large model can generate a series of targeted open-ended questions: "Can you describe in detail the work scene where you feel the most stress?", "How long has this situation lasted?", "How do you usually relieve this stress?", etc.

[0055] In an embodiment of the present application, the large model obtains psychological assessment interaction data for evaluating the user's psychology by generating open-ended questions. Since open-ended questions allow users to answer freely according to their actual situation and feelings, without being restricted by preset answers; and usually require respondents to provide detailed responses, rather than simple "yes" or "no", it can promote deeper and more extensive exchanges and discussions, and help obtain richer and more flexible information. For example, it can help get a variety of different answers, not be affected by preset answers, and is conducive to obtaining real and diverse information, so that users can fully express their opinions and views, which is conducive to exerting initiative and imagination, avoiding the induction of fixed answers to respondents, and can obtain more real information.

[0056] Phase c: interacting with the user in the capability assessment phase to obtain the user's capability assessment interaction data.

[0057] In the embodiment of the present application, the large model interacts with the user in the capability assessment stage by generating a scenario simulation test to obtain the user's capability assessment interaction data. Specifically, the scenario simulation test that the large model can generate can be a scenario simulation test that assumes that the user is a certain role and encounters certain events in a preset situation, so as to obtain the user's handling methods in the face of various different situations, so as to evaluate the user's emotional response patterns and various psychological ability conditions.

[0058] It is understandable that emotional response patterns can include fighting mode: this mode is manifested as fiercely fighting against the unpleasant emotions in the heart, and common behaviors include swearing, throwing things, roaring, fighting, etc. This mode is the instinctive stress response of human beings, which usually puts the emotional nerves in a state of tension. Escape mode: When facing difficult emotions, the escape mode is manifested as escape and diversion, and common behaviors include eating and drinking, crazy shopping, smoking, playing games, drinking, etc. Although this mode can temporarily relieve emotions, it cannot really solve the problem. Suppression mode: People in this mode usually look peaceful, not easy to get angry, and often deal with emotions by swallowing their anger. They often take the attitude of less trouble is worse than more trouble. Although they appear calm on the surface, they may be under great pressure in their hearts.

[0059] Based on this, in the embodiment of the present application, several different examples of obtaining evaluation interaction stages are given by way of example through stages ac. Based on this example, it can be obtained that the evaluation interaction data includes the basic information of the user, psychological evaluation interaction data and / or ability evaluation interaction data. It can be understood that the evaluation interaction stage and evaluation interaction data in the embodiment of the present application are not limited to the examples of stages ac. For example, stages ac may not appear so clearly in the entire interaction process, that is, in the process of obtaining basic information, if the interaction with the user reaches a time to obtain psychological ability or evaluate psychological state, the interaction of the psychological evaluation stage or the interaction of the ability evaluation stage can be inserted, and the interaction of the basic information acquisition stage can be continued thereafter.

[0060] Considering that in a real psychological counseling scenario, the user's psychological state may not only be reflected based on the content of his or her speech, but also based on his or her all-round performance throughout the process, such as the tone of speech, facial expressions and body movements. In order to more comprehensively evaluate the user's psychological condition, in one embodiment of the present application, the evaluation interaction data may include data of multiple modalities. Specifically, in an embodiment of the present application, the data modality of the evaluation interaction data includes at least one of the following: text data, voice data, image data and physiological data. Based on this, in an embodiment of the present application, the interaction between the large model and the user can be performed in a variety of different ways to complete the evaluation and training. It is understandable that the large model can be deployed on a server or a server cluster (hereinafter referred to as a server), and the server can interact with the terminal device through a communication connection for data interaction, and the terminal device can interact with the user through the input and output devices that it can control.

[0061] The input device that can be controlled by the terminal device may include one or more of an image acquisition device, a voice acquisition device, a physiological data acquisition device, a handwriting device, etc. Based on this, the interaction data type acquired by the large model when interacting with the user includes one of the following: text, voice, image, and physiological;

[0062] When interacting with the user through the large model, the user's interactive input method includes at least one of the following:

[0063] Using an image acquisition device to acquire a face image and / or a posture image of the user during the interaction process;

[0064] Using a physiological data acquisition device to obtain the user's physiological data during the interaction process; the physiological data includes one of the following: heart rate, blood pressure, body temperature, respiratory rate;

[0065] Using a handwriting device to obtain handwriting input data of the user during the interaction process;

[0066] A voice acquisition device is used to acquire the voice data of the user during the interaction process.

[0067] It can be understood that the image acquisition device can be a camera, the physiological data acquisition device can be a wearable device (such as a smart watch or bracelet) that integrates heart rate, blood pressure and body temperature detection, the handwriting device can be a touch screen or a stylus, and the voice acquisition device can be a microphone.

[0068] Similar to the fact that a variety of different input methods can be provided to the user during the interaction process in order to obtain evaluation interaction data of different modalities, the large model can also output interactive content through a variety of different output devices. Specifically, the output device that can be controlled by the terminal device may include one or more of a display device, a voice playback device, a somatosensory device, and the like. Based on this, the type of interaction data output by the large model when interacting with the user includes one of the following: text, voice, image, and somatosensory;

[0069] The interactive output mode when interacting with the user through the large model includes at least one of the following:

[0070] Using a display device to output the text dialogue content and / or image dialogue content of the large model;

[0071] Using a voice playback device to output the voice dialogue content of the large model;

[0072] A somatosensory device is used to output the somatosensory interaction content of the large model.

[0073] It is understandable that the display device may be a display, the voice playback device may be a speaker, and the somatosensory device may be a handle, a seat, or a massager including a vibration device. In some embodiments, the somatosensory device may also be a smart home device that can change the state of the environment, such as an air conditioner, a humidifier, a lighting device, or a curtain control device. Based on this, the somatosensory interaction content may be vibration or massage of a corresponding frequency and intensity to the user, or may be to increase or decrease the temperature, humidity, and brightness in the environment.

[0074] The above methods exemplify the interaction between the large model and the user in various modes, but are not limited to them. For example, the interaction scene can also be constructed through virtual reality (VR) or augmented reality (AR) devices, so that the user has a more immersive experience, interacts more realistically, and obtains more accurate evaluation interaction data, so that the user's personalized psychological portrait is more realistic.

[0075] In order to provide better overall interactive performance and user satisfaction, the interaction is more natural and humane. In one embodiment of the present application, the large model uses a dynamic adaptive interaction strategy to interact with the user in the evaluation phase; the dynamic adaptive interaction strategy is pre-built based on a reinforcement learning method; for example, the dynamic adaptive interaction strategy can optimize the interaction strategy based on multi-objective reinforcement learning. Specifically, the dynamic adaptive interaction strategy is also configured to be pre-trained based on a multi-objective reinforcement learning method to balance different interaction goals; the interaction goals include the efficiency of interaction intention realization, the user's interaction satisfaction and interaction cost. In detail, multi-objective reinforcement learning (MORL) aims to optimize several potentially conflicting goals. In intelligent interaction, MORL can be used to balance goals in multiple dimensions such as satisfaction, efficiency, and cost. For example, it can train the agent by designing a suitable reward function and strategy representation so that it can automatically select the optimal interaction strategy when facing different users and scenarios. After sufficient training, the system can make trade-offs between multiple goals and improve overall interactive performance and user satisfaction.

[0076] Specifically, the interaction between the large model and the user during the evaluation phase can include the following steps (1)-(3):

[0077] Step (1) generates and outputs an initial dialogue based on the dynamic adaptive interaction strategy to obtain dialogue feedback from the user.

[0078] In an embodiment of the present application, the initial dialogue can be generated by the dynamic adaptive interaction strategy at each different evaluation stage based on historical dialogue information. For example, the initial dialogue of the interaction in the psychological evaluation stage can be generated by the dynamic adaptive interaction strategy based on the dialogue of the interaction in the basic information acquisition stage. In this dynamic adaptive interaction strategy, the large model also uses a pre-built graph neural network-based dialogue state tracking module to perform state estimation based on historical dialogues, and the state estimation is used to guide the update of the dynamic adaptive interaction strategy.

[0079] Specifically, the dynamic adaptive interaction strategy can provide accurate state estimation based on the pre-built graph neural network (GNN), namely the dialogue state tracking module, to capture the complex relationship between related concepts in the dialogue process, making the dialogue coherent and natural. Specifically, the graph neural network is used in the dialogue state tracking (DST) to understand and represent the historical information of the dialogue and various relationships between entities, that is, the dialogue content can be represented in the form of a graph, where nodes may represent entities or concepts in the dialogue history, and edges represent the association or semantic flow between nodes, thereby capturing the complex relationship between objects, so as to update the node information through the message passing mechanism on the graph, and infer the state of the current dialogue based on the dialogue context, such as the user's intention and key slot value information. As a result, the key information in the historical dialogue can be retained by the system, so as to guide the update of the dynamic adaptive interaction strategy and generate context-related dialogue content.

[0080] Step (2), adjusting the dynamic adaptive interaction strategy according to the user's dialogue feedback.

[0081] In an embodiment of the present application, the dialogue feedback includes evaluation interaction data and a psychological feature fusion vector acquired in real time based on the evaluation interaction data, and the psychological feature fusion vector is updated in real time based on the interaction.

[0082] Step (3), generating and outputting a new dialogue based on the adjusted dynamic adaptive interaction strategy to obtain new dialogue feedback from the user until the interaction in the evaluation phase is completed.

[0083] In order to make the dialogue process more natural and humane, in one embodiment of the present application, during the interaction between the large model and the user, the response emotional style is obtained based on the emotional characteristics of the user, and the interactive dialogue content for output to the user is generated based at least on the response emotional style. Specifically, the dynamic adaptive interaction strategy can use an adaptive response generation mechanism that integrates emotional computing to guide the generation of each round of dialogue. For example, in some embodiments, the dialogue content can be generated based on historical dialogues and dialogue intentions. After obtaining the dialogue content, the dialogue content can be polished based on the adaptive response generation mechanism that integrates emotional computing so that it has an emotional style that meets the requirements.

[0084] In the adaptive response generation mechanism, the user's emotional state is first identified and understood, and then the content and style of the response are adjusted according to the emotional state. This mechanism may rely on natural language processing and machine learning techniques, such as sentiment classifiers, to analyze the user's input and generate responses with corresponding emotions. Systems that integrate this computing power can better connect with users and provide AI experiences with emotional interactions.

[0085] The above steps (1)-(3) exemplify an intelligent interactive decision-making method that can track the interactive process and generate context-coherent dialogue interaction content with a humanized emotional style.

[0086] In the embodiment of the present application, after obtaining the evaluation interaction data, real-time evaluation analysis can be performed based on the evaluation interaction data to obtain the personalized psychological portrait of the user. Specifically, the real-time evaluation analysis based on the evaluation interaction data can include the following steps I-III:

[0087] Step I: based on the data of different modes in the evaluation interaction data, feature extraction is performed respectively to obtain the features of each mode.

[0088] In the embodiment of the present application, the evaluation interaction data may include data in text mode, data in image mode, data in speech mode, data in physiological mode, and data in non-standard mode (such as structured data), that is, the basic information. Based on this, step I may include the following ways to obtain features of different modes:

[0089] Feature acquisition of text-based data

[0090] If the evaluation interaction data includes text data, the text data is subjected to syntactic analysis, semantic analysis and sentiment analysis to obtain multi-dimensional text features.

[0091] In the embodiment of the present application, the interaction between the large model and the user is mainly achieved through dialogue, and the content of the dialogue is the main basis for analyzing the user's psychological condition. Therefore, in the embodiment of the present application, different types of analysis will be performed based on the dialogue content, that is, the data of the text modality, to obtain multi-dimensional text features, so as to deeply understand the semantics of the dialogue, thereby more accurately obtaining the user's psychological characteristics.

[0092] Considering that the dialogue interaction between the large model and the user is continuous, the complete dialogue between the two parties is similar to a long text, that is, the dialogue in the previous time sequence may have semantic relevance with the dialogue in the later time sequence, that is, the two rounds of dialogue before and after with multiple rounds of dialogue in between may have long dependencies such as semantic relevance. Therefore, in order to capture the hidden semantic features in the dialogue process, when analyzing and processing the text data, the embodiment of the present application will vectorize each word in the text data, and then perform weighted processing to obtain the dependency relationship of data at different positions without relying on the original fixed sentence order, and obtain hidden semantic features.

[0093] Feature acquisition of speech modality data

[0094] If the evaluation interaction data includes voice data, feature extraction is performed on the voice data to obtain voice features, so as to obtain emotional features based on the voice features.

[0095] It is understandable that if the user interacts with the large model in the form of voice, then in the user's voice, in addition to the specific voice content that can reflect the user's psychological condition, the form of voice expression can also reflect the user's state. For example, when the user is angry, the volume of the voice may be very loud, the pitch may be very high, and the speaking speed may be very fast. Therefore, in the embodiment of the present application, in order to more comprehensively evaluate and analyze the user's psychological condition, when the evaluation interaction data includes voice data, voice signal performance feature extraction is performed, and the voice signal performance feature includes sound signal features such as pitch, volume, speaking speed and frequency, so as to obtain the user's emotional characteristics based on the voice signal performance feature.

[0096] Feature acquisition of image modality data

[0097] Taking into account that the user's facial expressions and body movements during the interaction process can also reflect the user's psychological condition, therefore, in some embodiments of the present application, facial images and body images of the user during the interaction process can be captured by an image capture device, so that the psychological condition can be obtained based on the user's facial expressions and body movements.

[0098] Therefore, in an embodiment of the present application, if the evaluation interaction data includes image data, feature extraction is performed on the face image and the posture image in the image data respectively to obtain facial expression features and posture features.

[0099] Feature acquisition of physiological modality data

[0100] If the evaluation interaction data includes physiological data, the physiological data is vectorized and physiological characteristics are obtained.

[0101] In an embodiment of the present application, the physiological data may include heart rate, blood pressure, body temperature, respiratory rate, etc. During the entire interaction process, various physiological data of the user can be acquired in real time, and time series data sets can be formed based on the type of physiological data. For example, heart rate time series data may include the heart rate values ​​of the user at each moment during the interaction process, thereby perceiving the psychological changes of the user during the interaction process based on the changes in heart rate values ​​at different moments. Other types of physiological data can obtain physiological characteristics based on similar processing methods, which will not be repeated here.

[0102] Feature acquisition of basic information modality data

[0103] Acquiring the evaluation interaction data includes basic information, and acquiring basic information features based on different types of data in the basic information respectively.

[0104] In some embodiments of the present application, basic information may include age, gender, occupation, education and living status, among which age, gender, occupation, and education are relatively fixed features and will not change much. They have little impact on the short-term formation of the user's psychological state, or the psychological changes they show are small. Therefore, they can be converted into fixed features through vectorization and other methods (for example, unique hot encoding, label encoding or embedding technology), which will not be repeated here. Living status may be a relatively dynamic feature. For example, living status may include the user's living environment, work environment, daily routine, behavioral data or social media activities, etc., which may change. Based on this, basic information features are obtained based on different types of data in the basic information, and the following steps 1-3 may be included:

[0105] Step 1: Acquire the initial psychological orientation characteristics of the user according to the static information in the basic information.

[0106] In the embodiment of the present application, the static information may be the working environment, living environment, long-term daily routine, etc. It is understandable that the user's living environment and working environment may have a greater impact on the user's psychology and play a certain role in determining the user's basic psychological condition.

[0107] For example, if a user lives and works in a narrow environment without sunlight for a long time, the user may develop a depressed and gloomy psychological tendency, and his or her views on various things tend to be gloomy; if the user lives and works in a wide environment with bright sunlight for a long time, the user may develop a positive and optimistic psychological tendency, and his or her views on various things tend to be bright. Based on this, in an embodiment of the present application, static information in the basic information is obtained, and the initial psychological orientation features of the user are determined based on the static information. The initial psychological orientation features may include positive, negative, optimistic, pessimistic, etc. The initial psychological orientation features can be used to assist in obtaining the user's psychological situation based on other evaluation interaction data, thereby realizing the establishment of a personalized psychological portrait of the user.

[0108] It can be understood that the initial psychological guidance features in the embodiments of the present application mainly reflect the psychological impact of the external environment or the internal environment on the user, and the static information may be other manifestations of the user's own psychological activities or environmental conditions.

[0109] Step 2: Based on the temporal dynamic information in the basic information, obtain the dynamic psychological state characteristics of the user.

[0110] In the embodiment of the present application, the time series dynamic information can be the user's short-term daily routine, behavior data, social media activities, etc., that is, data that can show the behavior that the user chooses to do based on his or her own psychological condition. For example, if a user does something happy one day, he or she may post a dynamic on social media to explain what kind of happy thing he or she did; or if the user's daily routine does not conform to the long-term routine, it may be because the user's psychology has fluctuated greatly and it is difficult to fall asleep normally.

[0111] Based on this, in the embodiments of the present application, the dynamic psychological state characteristics of the user can be analyzed based on different types of temporal dynamic information of the user.

[0112] Step 3: Fusing the initial psychological guidance feature and the dynamic psychological state feature to obtain the basic information feature.

[0113] In an embodiment of the present application, after obtaining the initial psychological orientation feature and the dynamic psychological state feature, the two features can be integrated to obtain a feature that describes the user's mind from different dimensions. For example, the initial psychological orientation feature and the dynamic psychological state feature can be combined in a multidimensional array to obtain a feature that can express both long-term stable psychological tendencies and short-term dynamic current psychological fluctuations.

[0114] The above steps 1-3 exemplarily provide a method for obtaining basic information features, but are not limited to this. Those skilled in the art can also set a suitable feature extraction method according to actual needs and specific data.

[0115] In some embodiments of the present application, a handwriting device is used to obtain the handwriting input data of the user, and then the handwriting features of the user can be obtained, and the psychological state of the user can be assisted in the assessment based on the handwriting features. For example, the handwriting features can also be integrated into the psychological feature fusion vector, so as to be used for psychological assessment. It can be understood that sloppy handwriting may indicate that the user is impatient or in an impatient mood; neat handwriting may indicate that the user is calm or in a calm mood.

[0116] In step II, the attention mechanism is used to fuse the features obtained from data analysis based on different modalities to obtain a psychological feature fusion vector.

[0117] In the embodiment of the present application, in order to more comprehensively evaluate the user's psychology, data of different modalities are collected through data collection devices of different modalities. Considering that the data of different modalities react to the user's psychological condition to different degrees, different attention levels can be applied when obtaining the user's psychological condition, that is, psychological characteristics, based on the data of each modality. That is, the fusion of features of different modalities can be achieved through the attention mechanism, thereby obtaining a psychological feature fusion vector.

[0118] In the embodiment of the present application, the concepts of query, key and value are introduced through the attention mechanism, the input features are evaluated for relevance, and the attention weights between the features are calculated. These weights are used to estimate which part of the input information is more important when constructing the psychological feature fusion vector. Through weighted summation, these input features are integrated to form a highly refined feature expression, which becomes the user's psychological feature fusion vector, thereby highlighting the meaningful part of the data, ignoring noisy information, and achieving more accurate feature fusion, which helps to improve the accuracy of the user's psychological state assessment.

[0119] Specifically, in an embodiment of the present application, the features of one or more modalities can be first selected as a query vector. For example, if the interactive evaluation data includes data of two modalities, image and text, the goal is to perform sentiment analysis based on text and image, and the feature vectors of text and image can be combined as a query vector. Then, the features of each modality can be used as a key vector. For example, the features of modalities such as image and text can be used as key vectors respectively. The attention score between the query vector and each key vector is calculated using methods such as dot product, bilinear transformation or additive attention, and the value vector (the value vector usually corresponds to the key vector, that is, the feature of each modality can be used as a value vector) is weighted and summed according to the attention score to obtain a fused feature vector, that is, a psychological feature fusion vector. That is, in an embodiment of the present application, the features of each modality can be first combined as a query vector, and then the weight of attention is calculated when the features of each modality are used as key vectors respectively, and finally the weighted summation is performed based on each weight and the feature vector of each modality to obtain a psychological feature fusion vector.

[0120] In order to make fuller use of data of different modalities and obtain a more comprehensive and accurate user psychological state. In one embodiment of the present application, an adaptive multi-layer attention mechanism can be used to further capture the complex relationship between different modalities. Specifically. In each layer of attention calculation, the query, key and value can be redistributed to gradually refine the fused feature vector. For example, assuming that the interactive evaluation data includes data of image modality and text modality, three attention calculation layers can be set based on the possible combination of different modality data, one of which uses the data of image modality as the query, and then uses the data of image modality and the data of text modality as the key to calculate the attention score, and then multiply the attention score by the value to obtain the weighted fusion psychological feature; another attention calculation layer uses the data of text modality as the query, and the subsequent calculation can refer to the first attention calculation layer, which will not be repeated here; in the last attention calculation layer, the data of image modality and the data of text modality can be combined as the query vector, and the subsequent calculation process is as described above, which will not be repeated here.

[0121] Based on this, it can be seen that in one embodiment of the present application, the adaptive multiple attention mechanism can automatically configure the attention calculation layer according to the possible modal combination, and then calculate a candidate psychological feature fusion vector in multiple attention calculation layers, and finally the candidate psychological feature fusion vectors can be fused to obtain the final psychological feature fusion vector.

[0122] Step III: analyzing and evaluating the psychological feature fusion vector to update the historical personalized psychological portrait of the user and obtain a real-time personalized psychological portrait of the user.

[0123] In the embodiment of the present application, after obtaining the user's psychological feature fusion vector, real-time analysis and evaluation can be performed to obtain the personalized psychological portrait of the user. It can be understood that in the embodiment of the present application, the psychological evaluation and analysis of the user can be carried out in real time as the interaction progresses, that is, each time a relatively complete evaluation interaction data is obtained, the corresponding evaluation and analysis can be carried out. For example, assuming that the user and the big model have the first round of dialogue in the basic information acquisition stage, the evaluation and analysis can be carried out when the first round of dialogue is completed, and based on the evaluation and analysis results, the generation of the second round of dialogue of the big model is guided, and the second round of dialogue is carried out. It can be understood that during the interaction process of the entire evaluation stage, the personalized portrait of the user is constantly updated and improved, that is, as the interaction proceeds, more and more interactive evaluation data can be obtained, so that more and more user psychological conditions can be analyzed and evaluated, and the personalized psychological portrait of the user can be gradually improved.

[0124] In the embodiment of the present application, the personalized portrait at least includes the multidimensional psychological state assessment result and the multidimensional analysis result of psychological ability of the user; the multidimensional psychological state assessment result includes at least one of the following: the emotional stability, stress tolerance, and cognitive state of the user; the multidimensional analysis result of psychological ability at least includes the shortcomings of the user in different psychological ability dimensions. It is understandable that since the personalized portrait of the user is gradually established during the interaction process, it is not possible to include the characteristics of each dimension in the initial stage. For example, in the basic information acquisition stage, data that can analyze and evaluate the user's psychological ability may not be obtained. In this stage, the personalized psychological portrait may not include the multidimensional analysis result of psychological ability.

[0125] In order to obtain the user's psychological characteristics more effectively and accurately, the personalized psychological portrait is constructed. In one embodiment of the present application, a real-time evaluation analysis can be performed based on the evaluation interaction data through a pre-constructed dynamic evaluation model; specifically, the dynamic evaluation model can include a time series evaluation model constructed using a recurrent neural network, and the recurrent neural network can process sequence data, and through the cyclic transmission of its hidden layer state, it can capture and learn the time series correlation in the data. For example, in some embodiments, the recurrent neural network can effectively simulate and capture the dynamic change process of the user's psychological state based on the input user behavior sequence or other time series data. In this way, the system can evaluate and predict the user's psychological state based on historical and current data. It can be seen that the time series evaluation model in the embodiment of the present application can be used to perform time series psychological state analysis based on the data of each modality, and obtain the psychological state change characteristics of the data of different modalities; the psychological change characteristics can be used to join the process of feature evaluation analysis based on different modalities in the above embodiment, and jointly construct a psychological feature fusion vector, or psychological evaluation analysis can be performed separately, and the personalized psychological portrait of the user can be improved based on the analysis results.

[0126] Furthermore, in an embodiment of the present application, the dynamic evaluation model can generate an adaptive evaluation strategy based on reinforcement learning, so as to adjust the evaluation strategy based on changes in the interaction environment and user feedback during the interaction process, so as to improve the effectiveness and accuracy of the evaluation strategy.

[0127] Based on this, after evaluating and analyzing the user's psychological state, the dynamic evaluation model in the embodiment of the present application can also construct the psychological state characteristics of each dimension into a psychological feature tensor, and update it based on real-time evaluation and analysis results, thereby helping to capture subtle changes in the user's state and facilitating the construction and improvement of the personalized psychological portrait.

[0128] Step 102: Generate a training plan based on the personalized portrait.

[0129] In an embodiment of the present application, the personalized portrait may include a multidimensional analysis result of the user's psychological ability, and the multidimensional analysis result of the psychological ability includes at least the shortcomings of the user in different psychological ability dimensions. Thus, one or more psychological ability dimensions that need to be improved can be determined based on the multidimensional analysis result of the psychological ability. Thus, the training goal of the training program is at least used to indicate the improvement of the shortcomings of the user in one psychological ability dimension. For example, if a user has shortcomings in pressure-bearing ability, the pressure-bearing ability of the user can be improved.

[0130] The following specifically describes that generating a training plan based on the personalized portrait may include:

[0131] Based on the personalized portrait, the shortcomings of the user in each psychological ability dimension are obtained, and the target psychological ability dimension is obtained therefrom.

[0132] Based on the target psychological ability dimensions, establish task objectives for each stage.

[0133] In the embodiment of the present application, after determining the target psychological ability dimension that needs to be improved, the improvement target can be established in stages. For example, if the target psychological ability dimension is stress tolerance, and the user's current stress tolerance is poor, the user's stress tolerance can be improved in multiple stages; for example, in the first training stage, the user's stress tolerance can be improved to fair; in the second training stage, the user's stress tolerance can be improved to good.

[0134] Based on the task objectives of each stage, training tasks for different stages are generated respectively.

[0135] In an embodiment of the present application, a variational autoencoder can be used to generate training tasks for different stages based on the task objectives of each stage. A variational autoencoder is a deep learning model that consists of an encoder and a decoder. The encoder is responsible for mapping the input data to a potential representation space, while the decoder maps the points in this potential space back to the data space. In an embodiment of the present application, VAE can be pre-trained and learn a complex data distribution based on a training sample set during training, so that the model can generate new content similar to the training samples. Furthermore, by optimizing the variational lower bound (ELBO), VAE can learn the probability distribution of training samples and make the generated content diverse.

[0136] In order to generate training tasks that meet the task objectives, in an embodiment of the present application, the variational autoencoder uses the task objectives of each stage as conditional constraints to generate training tasks at different stages. Specifically, it is necessary to introduce conditional constraints in the VAE model. This can usually be achieved by embedding additional information (such as labels or descriptions) into the latent space to guide the generation process. For example, if the task goal is to generate text on a specific topic, the topic label can be added as a condition. Conditional VAE (CVAE) learns to generate content that conforms to these conditions by considering this conditional information during the training process.

[0137] Furthermore, in order to make the generated content have a higher relevance and diversity with the task objectives, in one embodiment of the present application, the variational autoencoder is trained using a contrastive learning method. Contrastive learning is a machine learning technique that aims to learn the similarities and differences between data points. Applying contrastive learning in VAE can further optimize the relevance and diversity of generated content. Through contrastive learning, VAE is trained to be able to distinguish and pull together potential representations of similar content, while pushing away irrelevant or non-compliant content. Thereby, the consistency of the generated content with the training objectives can be strengthened, while maintaining the diversity and richness of the generated content.

[0138] In one embodiment of the present application, when generating training tasks, role-playing training tasks can be constructed, that is, different social or test scenarios are pre-constructed and the user is given a role in the scenario, so that the user can train more immersively and train in different social scenarios.

[0139] In one embodiment of the present application, when generating training tasks, a teamwork mode can be developed and set. For example, if the training task of a user A is to overcome his communication barriers with others, a social training task can be generated, and the social training task can simultaneously generate a help task, and push the help task to other users. Other users can socialize with user A based on the help task and help user A to practice socializing, so that multiple people can interact and improve their psychological training abilities.

[0140] In one embodiment of the present application, when generating training tasks, a points or reward mechanism may be set in some training tasks. When a user completes a certain number of training tasks or achieves a certain achievement in the process of performing training tasks, based on the user's points or other rewards, explicit positive feedback is provided to the user to motivate the user to continue training. In some possible embodiments, a leaderboard may be set to introduce a competition mechanism to help users generate training motivation.

[0141] In some embodiments of the present application, offline psychological counseling resources can be introduced into the big model. During the interaction or training process, if the big model determines that the user needs artificial psychological intervention, a matching psychological counselor can be recommended. It is understandable that in order to ensure the quality of the psychological ability training of the big model, the training quality of the big model can be supervised by experts.

[0142] In some embodiments of the present application, different template training programs can be pre-set, such as a training program for workplace stress management, a training program for adolescent psychological growth, or a psychological training program for a specific group of people (such as postpartum anxiety for pregnant women). This can help save time for interactive evaluation and resource consumption for generating tasks.

[0143] In one embodiment of the present application, the large model can also be converted into an intelligent assistant, which is resident in the user's terminal device in the form of an intelligent assistant to provide training services to the user. The image and personality of the intelligent assistant can be defined by the user, and the conversation style (such as humor, rigor or gentleness, etc.) can be adjusted according to the interaction with the user, so as to bring a better interactive experience to the user. It is understandable that the intelligent assistant can support switching between multiple different languages, so as to provide services for users with different cultural backgrounds.

[0144] After the training tasks of each stage are generated in the above manner, the training tasks of each stage may be integrated to obtain the training plan.

[0145] Step 103: Interact with the user in a training phase through the large model based on the training plan so that the user can complete the training plan.

[0146] In an embodiment of the present application, the server deploying the large model can interact with the terminal device to provide the user with training tasks for each stage of the training program so that the user can complete the training program.

[0147] In order to make the user's training more efficient and in line with the user's actual psychological condition, in the embodiment of the present application, in the process of executing the training task, the user's task completion feedback is obtained in real time so as to adjust and optimize the training task in a timely manner. Specifically, in the embodiment of the present application, in the process of interacting with the user in the training phase through the large model based on the training scheme, the method also includes:

[0148] Get the execution feedback of the user's current training task.

[0149] In an embodiment of the present application, the execution feedback includes at least the user's psychological state, training effect, completion progress of the training task and / or difficulty evaluation.

[0150] Wherein, obtaining the training effect of the current training task of the user includes:

[0151] The user's training task completion status is evaluated by using a pre-established training effect evaluation model based on a gradient boosting tree to obtain the training effect of the current training task.

[0152] Among them, Gradient Boosting Decision Trees (GBDT) is a machine learning technique for regression and classification problems that builds an integrated model. When implementing this technique, an initial weak prediction model, such as a decision tree, is first created. Subsequently, the algorithm iteratively adds new trees, each of which is trained based on the residual gradient predicted by the previous tree. In this way, each step attempts to reduce the error of the previous round of predictions. In the context of evaluating the effectiveness of training methods based on large models, gradient boosting trees can be used to predict the performance of training methods and quantify the degree of influence of each parameter to help researchers adjust and optimize the parameters of large models.

[0153] Considering that there may be multiple shortcomings in the personalized psychological profile of the user, that is, the user may have multiple psychological ability dimensions that need to be improved at the same time, that is, there are multiple training programs and simultaneous training tasks. Therefore, in one embodiment of the present application, if the user performs different training tasks at the same time, the interaction data and features of the user during the execution of different training tasks are also shared through multi-task learning, and the prospects of the user on different training tasks are predicted, so as to adjust the training program.

[0154] Multi-task Learning (MTL) is a learning paradigm in which the model learns multiple related tasks at the same time, sharing some information and patterns between them. This method can improve the performance of the model on each individual task. For user progress prediction, by sharing the feature representation of a large model, a unified model can be built to perform a variety of different but related prediction tasks. This may include the user's current position on the learning path, possible problems in the future, and the degree of mastery of the learning materials. Based on this, the correlation between tasks is used by MTL in the embodiment of the present application to improve model accuracy and generalization ability.

[0155] Next, the training scheme can be adjusted based on the execution feedback so that the subsequent unexecuted training tasks meet the preset requirements. Specifically, the training scheme can be adjusted based on the pre-built knowledge graph-based dynamic adjustment method for training schemes according to the execution feedback. It can be understood that the preset requirements can refer to making the user's expectations of executing the training task meet the target, that is, predicting the efficiency and situation of the user executing the training task, so that the prediction result meets the preset efficiency or user satisfaction.

[0156] A knowledge graph is a structure for storing and managing knowledge that can represent entities, concepts, and the relationships between them. In large models, knowledge graphs can be used to dynamically adjust training schemes. This is achieved by analyzing the performance of the model under different training sample sets and parameters, and combining this information with the relationships in the knowledge graph. Then, based on the performance of the model and the insights in the knowledge graph, the algorithm dynamically adjusts the selection of training data, the configuration of parameters, or the order of learning tasks to optimize the overall performance of the model. This closed-loop optimization feedback mechanism can more finely adjust the model's learning process, making it more efficient.

[0157] In summary, in the embodiments of the present application, a psychological state modeling method based on multimodal feature fusion is proposed, which realizes the dynamic and accurate construction of user psychological portraits; an innovative adaptive evaluation algorithm is designed, and dynamic optimization of evaluation strategies is realized through reinforcement learning to improve evaluation accuracy; a conditionally constrained content generation model is developed to ensure that the generated content not only meets the training objectives but also has personalized characteristics; an interactive decision-making system based on graph neural networks is implemented, which provides intelligent training guidance and emotional support; a complete closed-loop optimization mechanism is constructed, and continuous optimization of training plans is realized through multi-task learning and knowledge graphs.

[0158] The following is an explanation of the large model-based training method of this application in combination with a specific example:

[0159] Xiao Wang is a professional who is about to face an important promotion defense. Recently, he has developed severe anxiety symptoms due to stress. He decided to try to use a large model-based training method to help himself.

[0160] First, the training method based on the big model uses multiple rounds of conversational evaluation to understand Xiao Wang's specific situation. During the interaction, the training method based on the big model found that Xiao Wang had shortcomings in the dimension of psychological ability:

[0161] Extreme fear of public speaking

[0162] Frequent negative self-talk

[0163] Poor sleep quality

[0164] Decreased work efficiency

[0165] Based on this information, the large model-based training method generates a personalized training plan:

[0166] 1. In the speech ability training session:

[0167] We simulated various speech scenarios, from simple 2-3-person group reports to large-scale conference speeches. After each simulation, we analyzed Xiao Wang’s expression and emotional state, and gave specific suggestions for improvement.

[0168] 2. In emotion management training:

[0169] When Xiao Wang has negative thoughts like "I might mess up the entire defense", we will identify and guide him to conduct cognitive reconstruction in time, helping him to turn negative thoughts into positive action plans.

[0170] 3. In terms of stress management:

[0171] A series of relaxation exercises and meditation guidance are designed based on Xiao Wang’s schedule, and real-time emotional support and stress-relieving suggestions are provided when he feels stressed.

[0172] After two weeks of continuous training, data analysis revealed that:

[0173] Xiao Wang’s public speaking anxiety level was reduced by 40%

[0174] Significant reduction in negative self-talk

[0175] Improved sleep quality

[0176] Improved work concentration

[0177] In the final promotion defense, Xiao Wang performed well and successfully got the promotion opportunity. In this process, the training method based on the big model not only helped him overcome the specific speech anxiety, but also improved his overall psychological resilience, allowing him to learn to better manage stress and emotions.

[0178] This case shows how the big model-based training method can help users solve practical psychological problems through dynamic evaluation, personalized training content generation, intelligent interactive guidance, and continuous optimization feedback. The system does not simply provide standardized solutions, but provides continuous and adaptive support based on the user's specific situation and progress.

[0179] In the embodiment of the present application, the training method based on the big model interacts with the user in the evaluation phase through the big model to obtain evaluation interaction data, and performs real-time evaluation analysis based on the evaluation interaction data to obtain a personalized portrait of the user; generates a training plan based on the personalized portrait; and interacts with the user in the training phase based on the training plan through the big model so that the user completes the training plan. It can be seen that the embodiment of the present application can use the big model for interactive intelligent guidance, thereby obtaining more real and multi-dimensional evaluation interaction data for evaluating the user's psychology, and obtaining more accurate psychological characteristics, so as to facilitate the design of targeted training plans for users, which can help improve the user's psychological resilience, emotional management, cognitive level and other psychological abilities, and bring a better experience to the user.

[0180] The above describes a large model-based training method in an embodiment of the present application. The following introduces a large model-based training system (such as a server) that executes the above large model-based training method.

[0181] See also Figure 3 ,like Figure 3 The schematic diagram of a large model-based training system is shown, which can be applied to the user's psychological ability training scenario, interact with the user through the terminal device, and complete the user's personalized portrait establishment and psychological ability training through interaction. The large model-based training in the embodiment of the present application can achieve the corresponding Figure 2 The steps of the method based on the big model in the corresponding embodiment are as follows. The functions implemented by the training based on the big model can be implemented by hardware, or by hardware executing the corresponding software. The hardware or software includes one or more modules corresponding to the above functions, and the modules can be software and / or hardware.

[0182] The large model-based training system may include:

[0183] The interaction module 601 is configured to interact with the user in the evaluation phase through the large model to obtain evaluation interaction data; the data modality of the evaluation interaction data includes at least one of the following: text data, voice data, image data and physiological data;

[0184] The evaluation module 602 is configured to perform real-time evaluation and analysis based on the evaluation interaction data to obtain a personalized portrait of the user; the personalized portrait at least includes a multi-dimensional psychological state evaluation result and a multi-dimensional psychological ability analysis result of the user; the multi-dimensional psychological state evaluation result includes at least one of the following: the user's emotional stability, stress tolerance, and cognitive state; the multi-dimensional psychological ability analysis result includes at least the user's shortcomings in different psychological ability dimensions;

[0185] The training module 603 is configured to generate a training plan according to the personalized portrait; and to interact with the user in the training phase based on the training plan through the large model so that the user can complete the training plan.

[0186] In some embodiments, the assessment interaction data includes basic information of the user, psychological assessment interaction data and / or ability assessment interaction data;

[0187] The interaction module 601 is further configured to interact with the user in the basic information acquisition phase to acquire the basic information of the user; the basic information includes at least one of the following information of the user: age, gender, occupation, education level and living status; and

[0188] Interacting with the user in a psychological assessment phase to obtain psychological assessment interaction data of the user; and

[0189] Interact with the user in the capability assessment phase to obtain capability assessment interaction data of the user.

[0190] In some embodiments, the interaction between the large model and the user at each stage includes multiple rounds of dialogue, and each round of dialogue is completed alternately by the large model and the user;

[0191] The large model interacts with the user in the basic information acquisition phase by generating an interactive exploration dialogue, so as to acquire the basic information of the user;

[0192] The large model interacts with the user in the psychological assessment stage by generating open questions to obtain the user's psychological assessment interaction data;

[0193] The large model interacts with the user in the capability assessment phase by generating a scenario simulation test to obtain the user's capability assessment interaction data.

[0194] In some embodiments, the type of interaction data output by the large model when interacting with the user includes one of the following: text, voice, image, and somatosensory;

[0195] The interactive output mode when interacting with the user through the large model includes at least one of the following:

[0196] Using a display device to output the text dialogue content and / or image dialogue content of the large model;

[0197] Using a voice playback device to output the voice dialogue content of the large model;

[0198] Using a somatosensory device to output the somatosensory interaction content of the large model;

[0199] The type of interaction data acquired by the large model when interacting with the user includes one of the following: text, voice, image, and physiology;

[0200] When interacting with the user through the large model, the user's interactive input method includes at least one of the following:

[0201] Using an image acquisition device to acquire a face image and / or a posture image of the user during the interaction process;

[0202] Using a physiological data acquisition device to obtain the user's physiological data during the interaction process; the physiological data includes one of the following: heart rate, blood pressure, body temperature, respiratory rate;

[0203] Using a handwriting device to obtain handwriting input data of the user during the interaction process;

[0204] A voice acquisition device is used to acquire the voice data of the user during the interaction process.

[0205] In some embodiments, the evaluation module 602 is further configured to perform feature extraction based on data of different modalities in the evaluation interaction data to obtain features of each modality; wherein each word in the data of the text modality in the evaluation interaction data is vectorized and then weighted to obtain the dependency relationship of data at different positions without relying on the original fixed order to obtain hidden semantic features; and

[0206] Using the attention mechanism to fuse the features obtained from data analysis based on different modalities to obtain a psychological feature fusion vector; and

[0207] An analysis and evaluation is performed based on the psychological feature fusion vector to update the historical personalized psychological portrait of the user and obtain a real-time personalized psychological portrait of the user.

[0208] In some implementations, the evaluation module 602 is further configured to, if the evaluation interaction data includes text data, perform syntactic analysis, semantic analysis, and sentiment analysis on the text data to obtain multi-dimensional text features;

[0209] If the evaluation interaction data includes voice data, feature extraction is performed on the voice data to obtain voice signal performance features, so as to obtain emotional features based on the voice signal performance features; the voice signal performance features include pitch, volume and frequency;

[0210] If the evaluation interaction data includes image data, then feature extraction is performed on the face image and the posture image in the image data to obtain the face expression feature and the posture feature;

[0211] If the evaluation interaction data includes physiological data, vectorizing the physiological data and acquiring physiological characteristics;

[0212] Acquiring the evaluation interaction data includes basic information, and acquiring basic information features based on different types of data in the basic information respectively.

[0213] In some embodiments, the living status includes at least one of the following: the user's living environment, work environment, daily routine, behavioral data, and social media activities;

[0214] The evaluation module 602 is further configured to obtain the initial psychological orientation characteristics of the user according to the static information in the basic information; and

[0215] Based on the time-series dynamic information in the basic information, obtaining the dynamic psychological state characteristics of the user; and

[0216] The initial psychological guidance feature and the dynamic psychological state feature are fused to obtain the basic information feature.

[0217] In some embodiments, the evaluation module 602 is further configured to perform real-time evaluation analysis based on the evaluation interaction data through a pre-built dynamic evaluation model;

[0218] The dynamic evaluation model includes a time series evaluation model constructed by using a recurrent neural network, which is used to perform time series psychological state analysis based on data of each modality to obtain psychological state change characteristics of data of different modalities;

[0219] The dynamic evaluation model generates an adaptive evaluation strategy based on reinforcement learning so as to adjust the evaluation strategy based on changes in the interaction environment and user feedback during the interaction process;

[0220] After evaluating and analyzing the user's psychological state, the dynamic evaluation model also constructs the psychological state characteristics of each dimension into a psychological characteristic tensor, and updates it based on the real-time evaluation and analysis results; the psychological characteristic tensor is used to construct the personalized psychological portrait.

[0221] In some embodiments, the interaction module 601 is further configured to use a dynamic adaptive interaction strategy to interact with the user in the evaluation phase; the dynamic adaptive interaction strategy is pre-constructed based on a reinforcement learning method;

[0222] The interaction module 601 is further specifically configured to generate and output an initial dialogue based on the dynamic adaptive interaction strategy to obtain dialogue feedback from the user; and

[0223] adjusting the dynamic adaptive interaction strategy according to the user's dialogue feedback; the dialogue feedback includes evaluation interaction data and a psychological feature fusion vector acquired in real time based on the evaluation interaction data, the psychological feature fusion vector being updated in real time based on the interaction; and

[0224] A new dialogue is generated and output based on the adjusted dynamic adaptive interaction strategy to obtain new dialogue feedback from the user until the interaction in the evaluation phase is completed.

[0225] In some embodiments, the large model further uses a pre-built graph neural network-based dialogue state tracking module to perform state estimation based on historical dialogues, and the state estimation is used to guide the update of the dynamic adaptive interaction strategy;

[0226] The dynamic adaptive interaction strategy is also configured to be pre-trained based on a multi-objective reinforcement learning method to balance different interaction objectives; the interaction objectives include interaction intention realization efficiency, user interaction satisfaction and interaction cost;

[0227] During the interaction between the large model and the user, a reply emotional style is acquired based on the emotional characteristics of the user, and interactive dialogue content for output to the user is generated at least based on the reply emotional style.

[0228] In some implementations, the training module 603 is further configured to obtain the user's shortcomings in various psychological ability dimensions based on the personalized portrait, and obtain a target psychological ability dimension therefrom; and

[0229] Establishing task objectives for each stage based on the target psychological ability dimensions; and

[0230] Generate training tasks for different stages based on the task objectives of each stage; and

[0231] The training plan is obtained by integrating the training tasks of each stage.

[0232] In some implementations, the training module 603 is further configured to generate training tasks for different stages based on the task objectives of each stage through a variational autoencoder;

[0233] The variational autoencoder generates training tasks at different stages based on the task objectives at each stage as conditional constraints;

[0234] The variational autoencoder is trained by using a contrastive learning method.

[0235] Reference Figure 4 , Figure 4Another structural diagram of a training system based on a large model according to an embodiment of the present application. In some embodiments, the system 60 further includes a feedback optimization module 604, which is configured to obtain execution feedback of the user's current training task; the execution feedback includes at least the user's psychological state, training effect, and completion progress and / or difficulty evaluation of the training task; and

[0236] The training plan is adjusted based on the execution feedback so that subsequent unexecuted training tasks meet preset requirements.

[0237] In some implementations, obtaining the training effect of the user's current training task includes:

[0238] The user's training task completion status is evaluated by using a pre-established training effect evaluation model based on a gradient boosting tree to obtain the training effect of the current training task.

[0239] In some embodiments, the feedback optimization module 604 is further configured to adjust the training scheme based on a pre-built knowledge graph-based training scheme dynamic adjustment method according to the execution feedback.

[0240] In some embodiments, if the user performs different training tasks at the same time, the interaction data and features of the user during the execution of different training tasks are shared through multi-task learning to predict the user's prospects on different training tasks in order to adjust the training plan.

[0241] In the embodiment of the present application, the training system based on the big model interacts with the user in the evaluation phase through the big model to obtain evaluation interaction data, and performs real-time evaluation analysis based on the evaluation interaction data to obtain a personalized portrait of the user; generates a training plan based on the personalized portrait; and interacts with the user in the training phase based on the training plan through the big model so that the user can complete the training plan. It can be seen that the embodiment of the present application can use the big model for interactive intelligent guidance, thereby obtaining more real and multi-dimensional evaluation interaction data for evaluating the user's psychology, and obtaining more accurate psychological characteristics, so as to facilitate the design of targeted training plans for users, which can help improve the user's psychological resilience, emotional management, cognitive level and other psychological abilities, and bring a better experience to the user.

[0242] The above describes the large model-based training system 60 in the embodiment of the present application from the perspective of modular functional entities. The following describes the large model-based training system in the embodiment of the present application from the perspective of hardware processing.

[0243] Figure 3 , Figure 4 The systems shown can all have Figure 5The structure shown, when Figure 3 , Figure 4 The large model-based training system 60 shown has the following features: Figure 5 When the structure shown is Figure 5 The processor and transceiver in the system can realize the above-mentioned system embodiment corresponding to the system and provide the same or similar functions, Figure 6 The memory in the storage processor is used to store computer programs that need to be called when executing the above-mentioned large model-based training method.

[0244] An embodiment of the present application also relates to a chip system, which includes at least one processor and an interface circuit, wherein the processor includes multiple vector storage units, and the processor is used to execute instructions and / or data interaction through the interface circuit, so that the chip system executes the method of any of the above embodiments.

[0245] In a possible implementation, the chip system may also directly include a memory, in which a computer program or computer instructions are stored.

[0246] For example, the memory may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of RAM are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link dynamic random access memory (SLDRAM), and direct rambus RAM (DRRAM).

[0247] An embodiment of the present application further relates to a processor, which includes multiple storage units for calling a computer program or computer instruction stored in the memory so that the processor executes the method described in any of the above embodiments.

[0248] For example, in an embodiment of the present application, the processor is an integrated circuit chip with signal processing capabilities. For example, the processor can be an FPGA, a general-purpose processor, a DSP, an ASIC or other programmable logic device, a discrete gate or transistor logic device, a discrete hardware component, a SoC, a CPU, a network processor (network processor, NP), a microcontroller (micro controller unit, MCU), a PLD or other integrated chip, which can implement or execute the disclosed methods, steps and logic block diagrams in the embodiments of the present application. In a possible implementation, the embodiment of the present application also provides a computer-readable storage medium, the computer-readable storage medium stores a program code, and when the program code is run on the computer, the computer executes the above method embodiment.

[0249] The present application also provides a terminal device, such as Figure 6 For the sake of convenience, only the parts related to the embodiments of the present application are shown. For specific technical details not disclosed, please refer to the method part of the embodiments of the present application. The terminal device can be any terminal device including a mobile phone, a tablet computer, a personal digital assistant (PDA), a point of sales (POS), a car computer, etc., taking the mobile phone as an example:

[0250] Figure 6 FIG. 1 is a block diagram showing a partial structure of a mobile phone related to a terminal device provided in an embodiment of the present application. Figure 6 The mobile phone includes: a radio frequency (RF) circuit 1010, a memory 1020, an input unit 1030, a display unit 1040, a sensor 1050, an audio circuit 1060, a wireless fidelity (WiFi) module 1070, a processor 1080, and a power supply 1090. Those skilled in the art will understand that Figure 6 The mobile phone structure shown in the figure does not constitute a limitation on the mobile phone, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.

[0251] Combine the following Figure 6 A detailed introduction to the various components of the mobile phone:

[0252] The RF circuit 1010 can be used for receiving and sending signals during information transmission or calls. In particular, after receiving the downlink information of the base station, it is sent to the processor 1080 for processing; in addition, the designed uplink data is sent to the base station. Generally, the RF circuit 1010 includes but is not limited to an antenna, at least one amplifier, a transceiver, a coupler, a low noise amplifier (Low Noise Amplifier, LNA), a duplexer, etc. In addition, the RF circuit 1010 can also communicate with the network and other devices through wireless communication. The above wireless communication can use any communication standard or protocol, including but not limited to Global System of Mobile communication (Global System of Mobile communication, GSM), General Packet Radio Service (General Packet Radio Service, GPRS), Code Division Multiple Access (Code Division Multiple Access, CDMA), Wideband Code Division Multiple Access (Wideband Code Division Multiple Access, WCDMA), Long Term Evolution (Long Term Evolution, LTE), email, Short Messaging Service (SMS), etc.

[0253] The memory 1020 can be used to store software programs and modules. The processor 1080 executes various functional applications and data processing of the mobile phone by running the software programs and modules stored in the memory 1020. The memory 1020 can mainly include a program storage area and a data storage area, wherein the program storage area can store an operating system, an application required for at least one function (such as a sound playback function, an image playback function, etc.), etc.; the data storage area can store data created according to the use of the mobile phone (such as audio data, a phone book, etc.), etc. In addition, the memory 1020 can include a high-speed random access memory, and can also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0254] The input unit 1030 can be used to receive input digital or character information, and to generate key signal input related to the user settings and function control of the mobile phone. Specifically, the input unit 1030 may include a touch panel 1031 and other input devices 1032. The touch panel 1031, also known as a touch screen, can collect the user's touch operation on or near it (such as the user's operation on the touch panel 1031 or near the touch panel 1031 using any suitable object or accessory such as a finger, stylus, etc.), and drive the corresponding connection device according to a pre-set program. Optionally, the touch panel 1031 may include two parts: a touch detection device and a touch controller. Among them, the touch detection device detects the user's touch orientation, detects the signal brought by the touch operation, and transmits the signal to the touch controller; the touch controller receives the touch information from the touch detection device, converts it into contact coordinates, and then sends it to the processor 1080, and can receive and execute commands sent by the processor 1080. In addition, the touch panel 1031 can be implemented in various types such as resistive, capacitive, infrared, and surface acoustic waves. In addition to the touch panel 1031, the input unit 1030 may further include other input devices 1032. Specifically, the other input devices 1032 may include but are not limited to one or more of a physical keyboard, function keys (such as volume control keys, switch keys, etc.), a trackball, a mouse, a joystick, and the like.

[0255] The display unit 1040 can be used to display information input by the user or information provided to the user and various menus of the mobile phone. The display unit 1040 may include a display panel 1041. Optionally, the display panel 1041 may be configured in the form of a liquid crystal display (LCD), an organic light-emitting diode (OLED), etc. Further, the touch panel 1031 may cover the display panel 1041. When the touch panel 1031 detects a touch operation on or near it, it is transmitted to the processor 1080 to determine the type of touch event. Subsequently, the processor 1080 provides a corresponding visual output on the display panel 1041 according to the type of touch event. Although in Figure 6 In the embodiment, the touch panel 1031 and the display panel 1041 are used as two independent components to realize the input and output functions of the mobile phone, but in some embodiments, the touch panel 1031 and the display panel 1041 can be integrated to realize the input and output functions of the mobile phone.

[0256] The mobile phone may also include at least one sensor 1050, such as a light sensor, a motion sensor, and other sensors. Specifically, the light sensor may include an ambient light sensor and a proximity sensor, wherein the ambient light sensor may adjust the brightness of the display panel 1041 according to the brightness of the ambient light, and the proximity sensor may turn off the display panel 1041 and / or the backlight when the mobile phone is moved to the ear. As a type of motion sensor, the accelerometer sensor can detect the magnitude of acceleration in all directions (generally three axes), and can detect the magnitude and direction of gravity when stationary. It can be used for applications that identify the posture of the mobile phone (such as horizontal and vertical screen switching, related games, magnetometer posture calibration), vibration recognition related functions (such as pedometer, tapping), etc.; as for other sensors that can be configured in the mobile phone, such as gyroscopes, barometers, hygrometers, thermometers, infrared sensors, etc., they will not be repeated here.

[0257] The audio circuit 1060, the speaker 1061, and the microphone 1062 can provide an audio interface between the user and the mobile phone. The audio circuit 1060 can transmit the received audio data to the speaker 1061 after converting the received audio data into an electrical signal, which is converted into a sound signal for output; on the other hand, the microphone 1062 converts the collected sound signal into an electrical signal, which is received by the audio circuit 1060 and converted into audio data, and then the audio data is output to the processor 1080 for processing, and then sent to another mobile phone through the RF circuit 1010, or the audio data is output to the memory 1020 for further processing.

[0258] Wi-Fi is a short-range wireless transmission technology. The mobile phone can help users send and receive emails, browse web pages, and access streaming media through the Wi-Fi module 1070. It provides users with wireless broadband Internet access. Figure 6 A Wi-Fi module 1070 is shown, but it is understandable that it is not an essential component of the mobile phone and can be omitted as needed without changing the essence of the invention.

[0259] The processor 1080 is the control center of the mobile phone. It uses various interfaces and lines to connect various parts of the entire mobile phone. It executes various functions of the mobile phone and processes data by running or executing software programs and / or modules stored in the memory 1020, and calling data stored in the memory 1020, so as to monitor the mobile phone as a whole. Optionally, the processor 1080 may include one or more processing units; optionally, the processor 1080 may integrate an application processor and a modem processor, wherein the application processor mainly processes the operating system, user interface and application programs, etc., and the modem processor mainly processes wireless communications. It is understandable that the above-mentioned modem processor may not be integrated into the processor 1080.

[0260] The mobile phone also includes a power supply 1090 (such as a battery) for supplying power to various components. Optionally, the power supply can be logically connected to the processor 1080 through a power management system, thereby managing functions such as charging, discharging, and power consumption management through the power management system.

[0261] Although not shown, the mobile phone may also include a camera, a Bluetooth module, etc., which will not be described in detail here.

[0262] The present application also provides a server. Figure 7 , Figure 7 1 is a schematic diagram of a server structure provided in an embodiment of the present application. The server 1100 may have relatively large differences due to different configurations or performances, and may include one or more central processing units (full name in English: central processing units, abbreviated in English: CPU) 1122 (for example, one or more processors) and memory 1132, one or more storage media 1130 (for example, one or more mass storage devices) storing application programs 1142 or data 1144. Among them, the memory 1132 and the storage medium 1130 can be short-term storage or permanent storage. The program stored in the storage medium 1130 may include one or more modules (not shown in the figure), and each module may include a series of instruction operations on the server. Furthermore, the central processing unit 1122 can be configured to communicate with the storage medium 1130 to execute a series of instruction operations in the storage medium 1130 on the server 1100.

[0263] The server 1100 may also include one or more power supplies 1126, one or more wired or wireless network interfaces 1150, one or more input and output interfaces 1158, and / or one or more operating systems 1141, such as Windows Server, Mac OS X, Unix, Linux, FreeBSD, etc.

[0264] The steps performed by the server in the above embodiment can be based on the Figure 7 The structure of the server 1100 shown in FIG. Figure 3 , Figure 4 The steps performed by the large model-based training system 60 shown can be based on the Figure 7 For example, the central processor 1122 performs the following operations by calling the instructions in the memory 1132:

[0265] The input / output interface 1158 is used to interact with the user in the evaluation phase through the big model to obtain evaluation interaction data, and real-time evaluation and analysis are performed based on the evaluation interaction data to obtain a personalized portrait of the user; the data modality of the evaluation interaction data includes at least one of the following: text data, voice data, image data, and physiological data; the personalized portrait includes at least the multidimensional psychological state evaluation result and the multidimensional analysis result of psychological ability of the user; the multidimensional psychological state evaluation result includes at least one of the following: the emotional stability, stress tolerance, and cognitive state of the user; the multidimensional analysis result of psychological ability includes at least the shortcomings of the user in different psychological ability dimensions;

[0266] Generate a training plan based on the personalized portrait; the training goal of the training plan is at least used to indicate how to improve the user's shortcomings in one psychological ability dimension;

[0267] The large model interacts with the user in a training phase based on the training solution so that the user can complete the training solution.

[0268] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0269] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and modules described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0270] In the several embodiments provided in the embodiments of the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.

[0271] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0272] In addition, each functional module in each embodiment of the present application can be integrated into a processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The above-mentioned integrated module can be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0273] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0274] The computer program product includes one or more computer instructions. When the computer program is loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from a computer-readable storage medium to another computer-readable storage medium, for example, the computer instructions may be transmitted from a website site, a computer, a server, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a server or a data center that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state hard disk Solid State Disk (SSD)), etc.

[0275] The technical solutions provided in the embodiments of the present application are introduced in detail above. The principles and implementation methods of the embodiments of the present application are explained by using specific examples in the embodiments of the present application. The description of the above embodiments is only used to help understand the methods and core ideas of the embodiments of the present application. At the same time, for those skilled in the art, according to the ideas of the embodiments of the present application, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the embodiments of the present application.

Claims

1. A training method based on a large model, characterized in that: The method comprises: The evaluation interaction data is obtained by interacting with the user in the evaluation phase, and real-time evaluation and analysis are performed based on the evaluation interaction data to obtain a personalized portrait of the user; the data modality of the evaluation interaction data includes at least one of the following: text data, voice data, image data and physiological data; the personalized portrait includes at least the multidimensional psychological state evaluation result and the multidimensional analysis result of psychological ability of the user; the multidimensional psychological state evaluation result includes at least one of the following: the emotional stability, stress tolerance and cognitive state of the user; the multidimensional analysis result of psychological ability includes at least the shortcomings of the user in different psychological ability dimensions; Generate a training plan based on the personalized portrait; the training goal of the training plan is at least used to indicate how to improve the user's shortcomings in one psychological ability dimension; The large model interacts with the user in a training phase based on the training solution so that the user can complete the training solution.

2. The method according to claim 1, characterized in that The evaluation interaction data includes basic information of the user, psychological evaluation interaction data and / or ability evaluation interaction data; The evaluation interaction data obtained by the interaction between the large model and the user during the evaluation phase includes: Interacting with the user in the basic information acquisition phase to acquire the user's basic information; the basic information includes at least one of the following information of the user: age, gender, occupation, education level, and living status; Interacting with the user in a psychological assessment phase to obtain psychological assessment interaction data of the user; Interact with the user in the capability assessment phase to obtain capability assessment interaction data of the user.

3. The method according to claim 2, characterized in that The interaction between the large model and the user at each stage includes multiple rounds of dialogue, and each round of dialogue is completed alternately by the large model and the user; The large model interacts with the user in the basic information acquisition phase by generating an interactive exploration dialogue, so as to acquire the basic information of the user; The large model interacts with the user in the psychological assessment stage by generating open questions to obtain the user's psychological assessment interaction data; The large model interacts with the user in the capability assessment phase by generating a scenario simulation test to obtain the user's capability assessment interaction data.

4. The method according to claim 1, characterized in that The type of interaction data output by the large model when interacting with the user includes one of the following: text, voice, image, and body sensation; The interactive output mode when interacting with the user through the large model includes at least one of the following: Using a display device to output the text dialogue content and / or image dialogue content of the large model; Using a voice playback device to output the voice dialogue content of the large model; Using a somatosensory device to output the somatosensory interaction content of the large model; The type of interaction data acquired by the large model when interacting with the user includes one of the following: text, voice, image, and physiology; When interacting with the user through the large model, the user's interactive input method includes at least one of the following: Using an image acquisition device to acquire a face image and / or a posture image of the user during the interaction process; Using a physiological data acquisition device to obtain the user's physiological data during the interaction process; the physiological data includes one of the following: heart rate, blood pressure, body temperature, respiratory rate; Using a handwriting device to obtain handwriting input data of the user during the interaction process; A voice acquisition device is used to acquire the voice data of the user during the interaction process.

5. The method according to claim 3, characterized in that The real-time evaluation analysis based on the evaluation interaction data includes: Based on the data of different modes in the evaluation interaction data, feature extraction is performed respectively to obtain the features of each mode; wherein each word in the data of the text mode in the evaluation interaction data is vectorized and then weighted to obtain the dependency relationship of data at different positions without relying on the original fixed order, so as to obtain the hidden semantic features; The attention mechanism is used to fuse the features obtained from data analysis based on different modalities to obtain a psychological feature fusion vector; An analysis and evaluation is performed based on the psychological feature fusion vector to update the historical personalized psychological portrait of the user and obtain a real-time personalized psychological portrait of the user.

6. The method according to claim 5, characterized in that Based on the data of different modes in the evaluation interaction data, feature extraction is performed respectively to obtain the features of each mode, including: If the evaluation interaction data includes text data, then the text data is subjected to syntactic analysis, semantic analysis and sentiment analysis to obtain multidimensional text features; If the evaluation interaction data includes voice data, feature extraction is performed on the voice data to obtain voice signal performance features, so as to obtain emotional features based on the voice signal performance features; the voice signal performance features include pitch, volume and frequency; If the evaluation interaction data includes image data, feature extraction is performed on the face image and the posture image in the image data to obtain the face expression feature and the posture feature; If the evaluation interaction data includes physiological data, vectorizing the physiological data and acquiring physiological characteristics; Acquiring the evaluation interaction data includes basic information, and acquiring basic information features based on different types of data in the basic information respectively.

7. The method according to claim 6, characterized in that The living status includes at least one of the following: the user's living environment, work environment, daily routine, behavioral data, and social media activities; Obtaining basic information features based on different types of data in the basic information respectively, including: Acquiring the initial psychological orientation characteristics of the user according to the static information in the basic information; Based on the time-series dynamic information in the basic information, obtaining the dynamic psychological state characteristics of the user; The initial psychological guidance feature and the dynamic psychological state feature are fused to obtain the basic information feature.

8. The method according to claim 5, characterized in that Performing real-time evaluation analysis based on the evaluation interaction data through a pre-built dynamic evaluation model; The dynamic evaluation model includes a time series evaluation model constructed by using a recurrent neural network, which is used to perform time series psychological state analysis based on data of each modality to obtain psychological state change characteristics of data of different modalities; The dynamic evaluation model generates an adaptive evaluation strategy based on reinforcement learning, so as to adjust the evaluation strategy based on changes in the interaction environment and user feedback during the interaction process; After evaluating and analyzing the user's psychological state, the dynamic evaluation model also constructs the psychological state characteristics of each dimension into a psychological characteristic tensor, and updates it based on the real-time evaluation and analysis results; the psychological characteristic tensor is used to construct the personalized psychological portrait.

9. The method according to claim 1, characterized in that The large model uses a dynamic adaptive interaction strategy to interact with the user in the evaluation phase; The dynamic adaptive interaction strategy is pre-constructed based on reinforcement learning; Interact with users during the evaluation phase through a large model, including: Generate and output an initial dialogue based on the dynamic adaptive interaction strategy to obtain dialogue feedback from the user; The dynamic adaptive interaction strategy is adjusted according to the user's dialogue feedback; the dialogue feedback includes evaluation interaction data and a psychological feature fusion vector acquired in real time based on the evaluation interaction data, and the psychological feature fusion vector is updated in real time based on the interaction; A new dialogue is generated and output based on the adjusted dynamic adaptive interaction strategy to obtain new dialogue feedback from the user until the interaction in the evaluation phase is completed.

10. A training system based on a large model, characterized in that: The system includes: An interaction module is configured to interact with the user in the evaluation phase through the large model to obtain evaluation interaction data; the data modality of the evaluation interaction data includes at least one of the following: text data, voice data, image data and physiological data; An evaluation module is configured to perform real-time evaluation and analysis based on the evaluation interaction data to obtain a personalized portrait of the user; the personalized portrait at least includes a multidimensional psychological state evaluation result and a multidimensional psychological ability analysis result of the user; the multidimensional psychological state evaluation result includes at least one of the following: the user's emotional stability, stress tolerance, and cognitive state; the multidimensional psychological ability analysis result includes at least the user's shortcomings in different psychological ability dimensions; The training module is configured to generate a training plan according to the personalized portrait; and to interact with the user in the training phase based on the training plan through the big model so that the user can complete the training plan.

Images