Psychological consultation method and system based on AI dialogue system

Abstract

The invention discloses a psychological counseling method and system based on an AI dialogue system, and relates to the technical field of artificial intelligence. By adopting a psychological professional data set construction algorithm, normalization, specialty, privacy and safety of model input are ensured, a multilevel filtering and evaluation method is adopted, a large model is used for data scoring and screening, a sensitive word filtering and manual proofreading mechanism and other related technologies are used, the quality of data required by the model can be improved, and the method is suitable for popularization and application. The performance of the constructed psychological dialogue model is improved, the reliable AI dialogue psychological large model is trained through technologies such as incremental pre-training, supervised fine tuning, reinforcement learning training and artificial feedback reinforcement, the model can more accurately understand user input, professional and high-quality consultation replies are generated, and the method is suitable for popularization and application. And the constructed psychological dialogue model is continuously optimized and improved according to user feedback, so that the consultation quality of the model and the user experience are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of photovoltaic grid-connected management, and in particular to a psychological consultation method and system based on an AI dialogue system. Background Art

[0002] At present, psychological counseling is considered an important way to help individuals solve psychological problems. However, the existing psychological counseling methods mainly rely on face-to-face communication between counselors, which is limited in time and space. In addition, the subjective evaluation of counselors may also lead to deviations in diagnostic results. In addition, there are also shortcomings such as high counseling fees and uneven regional coverage in the market.

[0003] With the development of artificial intelligence technology, AI-based dialogue systems have gradually attracted attention in the fields of medical and mental health. The working logic of the existing AI dialogue system is to automatically select highly matching replies from the constructed corpus based on the user dialogue. Therefore, the construction of the corpus is the key to the success of the AI ​​dialogue system. The data source of the corpus in the existing technology is mainly from literature, forums, etc., which does not meet the professional requirements for psychological counseling, and the construction process often only uses simple data preprocessing technology to process the original data, resulting in the standardization and professionalism of the constructed corpus cannot meet the requirements, making the constructed AI dialogue system model less effective in actual application, and it is difficult to provide users with high-quality dialogue services. To this end, we propose a psychological counseling method and system based on an AI dialogue system. Summary of the invention

[0004] The main purpose of the present invention is to provide a psychological counseling method and system based on an AI dialogue system, and to provide personalized psychological counseling services through natural language processing and sentiment analysis technology, thereby improving existing psychological counseling methods, enhancing the universality and convenience of counseling, and effectively solving the problems in the background technology.

[0005] To achieve the above object, the technical solution adopted by the present invention is:

[0006] The psychological consultation method based on the AI ​​dialogue system is characterized by including:

[0007] Generate dialogue tasks and obtain user input;

[0008] Through natural language processing technology, user input is understood in real time and a dialogue scenario is generated, where:

[0009] The conversation scenarios are classified according to the subject, emotion, and intention of the user input. Common classifications include learning problems, interpersonal communication, family relationships, emotional problems, behavioral problems, and self-cognition.

[0010] Encode user input through common natural language processing algorithms such as LSTM and Transformer to obtain its semantic representation;

[0011] Match the user's semantic representation with the pre-defined conversation scenario semantics, and determine the most matching conversation scenario by calculating similarities, such as cosine similarity and Euclidean distance;

[0012] Introducing contextual understanding, comprehensively considering the user's current conversation content and previous conversation history, improving matching accuracy, dynamically adjusting matching strategies, and continuously optimizing matching algorithms through user feedback and emotional fluctuation information;

[0013] Generate the best consulting strategy based on the dialogue scenario through the psychological dialogue model;

[0014] Monitor the emotional fluctuations and psychological state of users in the dialogue task in real time, and adjust the consultation strategy according to the monitoring results until the dialogue task is completed.

[0015] Principles for determining when a user's emotions fluctuate or their psychological state changes:

[0016] Text sentiment analysis: Use natural language processing technology to perform sentiment analysis on user input text, identify positive and negative changes in sentiment, and use sentiment dictionaries and sentiment classification models, such as sentiment dictionaries, BERT, and RoBERTa, to make judgments;

[0017] The strategy for determining the degree of emotional fluctuation or the degree of change in psychological state is to classify the text input, regress and predict the amplitude of emotional fluctuation, and make a comprehensive judgment based on contextual information, such as the emotional changes of the previous and next sentences;

[0018] For the matching of the degree of emotional fluctuation or psychological state change with the adjustment of the consultation strategy, based on the reinforcement learning model, such as DQN-SIMPO, the consultation strategy is dynamically adjusted according to the real-time monitoring of emotional fluctuation data, and the optimal consultation response measures are selected, including:

[0019] Emotional adaptive strategy generation: Match different intensity counseling strategies according to the degree of emotional fluctuations. For example, soothing dialogues are used for mild fluctuations, supportive dialogues are provided for moderate fluctuations, and crisis intervention is performed for severe fluctuations;

[0020] User feedback loop: Get real-time feedback from users on the current consulting strategy, adjust the strategy model through feedback, and ensure continuous optimization of consulting results;

[0021] The construction process of the psychological dialogue model includes the following steps:

[0022] S1: Collect user conversation data through various methods for preprocessing, and use the preprocessed conversation data to build a psychological conversation dataset;

[0023] The specific process includes the following:

[0024] S11: Randomly extract a certain proportion of the original data from the psychological dialogue dataset, manually evaluate the quality of the extracted data, and remove noise and irrelevant content from the extracted data;

[0025] S12: predefined data filtering rules, clean the data according to the predefined rules to remove obvious noise and irrelevant content;

[0026] S13: Use text similarity algorithm to deduplicate the data and eliminate duplicate conversation data in the dataset;

[0027] S14: Use the pre-trained large language model to score the acquired conversation data, set a scoring threshold, and retain the data within the scoring threshold based on the scoring results;

[0028] S15: Establish a sensitive vocabulary list, perform matching filtering on the data, and remove or replace the conversation data containing sensitive words;

[0029] S16: manually proofreading the conversation data processed by the above steps, and constructing a psychological conversation data set based on the proofread conversation data;

[0030] S2: Using L l ama3-8B as a basic model, incrementally pre-training the psychological dialogue dataset to obtain a pre-trained psychological dialogue model;

[0031] S3: fine-tuning the pre-trained psychological dialogue model based on the psychological dialogue dataset using LoRA technology to obtain a fine-tuned psychological dialogue model;

[0032] S4: Build a simulation environment to simulate the interaction process between users and the system, and use DPO technology to directly optimize the strategy of the fine-tuned psychological dialogue model so that the model can obtain the best strategy in the simulation environment;

[0033] S5: Use SimPO technology to simulate user feedback and environmental changes in the dialogue task, optimize the simulation strategy of the fine-tuned psychological dialogue model, and obtain the enhanced psychological dialogue model;

[0034] S6: Obtain user feedback information, including user satisfaction ratings, suggestions, and opinions on system responses;

[0035] S7: Process and analyze the collected feedback information and construct a feedback data set;

[0036] S8: Use the feedback dataset for model training to obtain an iterative psychological dialogue model.

[0037] The psychological consultation system based on the AI ​​dialogue system includes a user input acquisition module, a dialogue task generation module, a dialogue scene generation module, and a psychological dialogue model construction module;

[0038] The dialogue task generation module is used to generate dialogue tasks for user psychological consultation;

[0039] The user input acquisition module is used to acquire user input content;

[0040] The dialogue scene generation module is used to understand user input in real time through natural language processing technology, and generate a dialogue scene according to the user input content;

[0041] The psychological dialogue model building module is used to build a psychological dialogue model for user psychological consultation, and generate the best consultation strategy according to the dialogue scenario through the psychological dialogue model;

[0042] The psychological dialogue model construction module includes a data acquisition module, a data preprocessing module, a dialogue model pre-training module, a dialogue model fine-tuning training module, and a dialogue model reinforcement training module;

[0043] The data collection module is used to collect conversation data of users from different regions, age groups, genders, and cultural backgrounds;

[0044] The data preprocessing module is used to preprocess the collected user dialogue data and construct a psychological dialogue data set using the preprocessed dialogue data;

[0045] The dialogue model pre-training module is used to obtain a pre-trained psychological dialogue model by performing incremental pre-training on the psychological dialogue dataset using L l ama3-8B as a basic model;

[0046] The dialogue model fine-tuning training module is used to fine-tune the pre-trained psychological dialogue model based on the psychological dialogue dataset using LoRA technology to obtain a fine-tuned psychological dialogue model;

[0047] The dialogue model reinforcement training module is used to perform reinforcement learning training on the constructed psychological dialogue model in a simulation environment;

[0048] The dialogue model intensive training module includes a simulation environment construction submodule, a direct strategy optimization submodule, and a simulation strategy optimization submodule;

[0049] The simulation environment construction submodule is used to construct a simulation environment for simulating the interaction process between the user and the system;

[0050] The direct strategy optimization submodule is used to perform direct strategy optimization on the consultation strategy in the simulation environment using the DPO technology, so that the model can obtain the best strategy in each dialogue scenario in the simulation environment;

[0051] The simulation strategy optimization submodule uses SimPO technology to perform simulation strategy optimization by simulating user feedback and environmental changes.

[0052] The system also includes an artificial feedback reinforcement learning module, which optimizes the decision-making and response capabilities of the psychological dialogue model by introducing artificial feedback, including a user feedback information acquisition submodule, a user feedback information processing submodule, and an enhanced training submodule;

[0053] The user feedback information acquisition submodule is used to acquire feedback information including user satisfaction scores, suggestions, and opinions on system responses;

[0054] The user feedback information processing submodule is used to process and analyze the collected feedback information and construct a feedback data set;

[0055] The reinforcement training submodule is used to use the feedback data set for model training to obtain an iterative psychological dialogue model.

[0056] The data preprocessing module includes a sampling evaluation submodule, a rule filtering submodule, a data deduplication submodule, a data evaluation submodule, a sensitive word filtering submodule, and a manual proofreading and screening submodule;

[0057] The sampling evaluation submodule is used to randomly extract a certain proportion of the original data in the psychological dialogue data set, and manually evaluate the quality of the extracted data to remove noise and irrelevant content in the extracted data;

[0058] The rule filtering submodule is used to predefine data filtering rules and clean the data according to the predefined rules to remove obvious noise and irrelevant content;

[0059] The data deduplication submodule is used to perform deduplication processing on the data using a text similarity algorithm to eliminate duplicate conversation data in the data set;

[0060] The data evaluation submodule is used to score the acquired dialogue data using a pre-trained large language model, set a scoring threshold, and retain data within the scoring threshold based on the scoring result;

[0061] The sensitive word filtering submodule is used to establish a sensitive word list, perform matching filtering on the data, and remove or replace the conversation data containing sensitive words;

[0062] The manual proofreading and screening submodule is used to manually proofread the dialogue data and construct a psychological dialogue data set with the proofread dialogue data.

[0063] The system further includes a memory, a processor, and an electronic program stored in the memory and capable of running on the processor.

[0064] The present invention has the following beneficial effects:

[0065] 1) Compared with the prior art, the technical solution of the present invention ensures the standardization, professionalism, privacy and security of the model input by adopting a psychology professional data set construction algorithm. In the data preprocessing step, a multi-level filtering and evaluation method is adopted, and the technology of data scoring and screening using a large model, sensitive word filtering and manual proofreading mechanism and other related technologies can improve the quality of the data required by the model, thereby facilitating the improvement of the performance of the constructed psychological dialogue model;

[0066] 2) Compared with the prior art, the technical solution of the present invention trains a reliable AI dialogue psychology model through incremental pre-training, supervised fine-tuning, reinforcement learning training and manual feedback reinforcement, so that the model can understand user input more accurately and generate professional and high-quality consultation responses;

[0067] 3) Compared with the prior art, the technical solution of the present invention can instantly understand user needs and provide corresponding responses during the conversation through real-time natural language processing technology, thereby improving the user interaction experience, and can also adjust the consulting strategy according to user feedback to provide high-quality consulting services;

[0068] 4) Compared with the prior art, the technical solution of the present invention collects users' opinions and suggestions on consultation effects, system performance, etc., and continuously optimizes and improves the constructed psychological dialogue model based on user feedback and conversation records, thereby improving the consultation quality and user experience of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a schematic diagram of the construction process of the psychological dialogue model of the present invention;

[0070] Figure 2 This is a flow chart of the conversation data preprocessing process of the present invention;

[0071] Figure 3 This is a structural block diagram of the psychological counseling system based on the AI ​​dialogue system of the present invention. DETAILED DESCRIPTION

[0072] The present invention will be further described below in conjunction with specific implementation methods, wherein the accompanying drawings are only used for exemplary descriptions and represent only schematic diagrams rather than actual drawings, and should not be understood as limiting the present invention. In order to better illustrate the specific implementation methods of the present invention, some parts of the accompanying drawings may be omitted, enlarged or reduced, and do not represent the size of the actual product.

[0073] A psychological consultation method and system based on an AI dialogue system, wherein the psychological consultation method comprises the following steps:

[0074] 1) Generate a dialogue task and obtain user input;

[0075] 2) Understand user input in real time through natural language processing technology and generate dialogue scenarios;

[0076] 3) Generate the best consulting strategy based on the dialogue scenario through the psychological dialogue model;

[0077] 4) Monitor the emotional fluctuations and psychological state of users in the dialogue task in real time, and adjust the consultation strategy according to the monitoring results until the dialogue task is completed.

[0078] The process of constructing the psychological dialogue model in the technical solution of the present invention includes the following steps:

[0079] Step 1: Collect user conversation data through various methods;

[0080] Among them, psychological dialogue data comes from books, interviews, consultations, question-and-answer communities, videos, group counseling, open source data sets, etc. The system collects user dialogue data in a variety of ways. The collected data should cover users from different regions, age groups, genders, and cultural backgrounds to ensure the universality and robustness of the model. Moreover, the dialogue data covers multiple psychological schools, such as CBT, humanistic, dynamic, narrative, short-focus, Gestalt and other major psychological schools, ensuring the quality and professionalism of the data.

[0081] Step 2: Preprocess the collected user conversation data and use the preprocessed conversation data to build a psychological conversation dataset. The specific process includes the following steps:

[0082] Step 21: Manual Sampling Evaluation

[0083] A certain percentage of data is randomly selected from the data set and manually evaluated to check whether the data contains noise, irrelevant content or other quality issues. The main purpose is to conduct a quality assessment on the initially collected data to ensure the representativeness, diversity and standardization of the data.

[0084] Step 22: Rule filtering

[0085] Preliminary cleaning of data is performed using predefined rules to remove obvious noise and irrelevant content. By setting a series of filtering rules, such as removing data containing specific keywords, conversations that are too short or too long, and data with incorrect formats, rule filtering can quickly remove data that obviously does not meet the requirements and improve the overall quality of the data set.

[0086] Step 23: Algorithmic deduplication

[0087] Eliminate duplicate conversations in the data set to avoid overfitting during training. Use text similarity algorithms such as Jaccard similarity and cosine similarity to deduplicate the data and ensure that each conversation appears only once in the data set.

[0088] Step 24: Scoring the large model data

[0089] Use pre-trained large models to score the data and filter out high-quality data. Use pre-trained large language models such as BERT, GPT, etc. to score each conversation and evaluate its quality and relevance. Based on the scoring results, retain high-quality data and remove low-quality data.

[0090] Step 25: Sensitive word filtering

[0091] Remove sensitive words from conversations to ensure the security and compliance of data sets. This mainly involves establishing a sensitive word list, matching and filtering data, removing conversations containing sensitive words, or processing them appropriately, such as replacing or blocking them.

[0092] Step 26: Manual Proofreading and Screening

[0093] Manually proofread the data after the previous steps to further improve the accuracy and consistency of the data. The data is manually checked and proofread one by one to ensure the accuracy and consistency of the data content. In particular, the data related to sentiment analysis and consultation questions and answers are proofread and marked.

[0094] Step 3: Train the AI ​​psychological dialogue model. The specific training process is as follows:

[0095] Step 31: Incremental pre-training

[0096] The goal is to further improve the model's language understanding and generation capabilities by training on a large-scale psychological dialogue dataset, using L l ama3-8B as the base model. L l ama3-8B is a pre-trained large language model with powerful natural language processing capabilities. Incremental pre-training is performed on the psychological dialogue dataset to further optimize the model's parameters so that it can better adapt to dialogue tasks in specific fields.

[0097] Step 32: Supervised Fine-tuning

[0098] The goal is to enable the model to perform specific tasks, namely dialogue consultation, more accurately by fine-tuning it on a psychological dialogue dataset. LoRA technology is used for fine-tuning. LoRA is a low-rank adaptation technology that can achieve task adaptability of the model by adjusting a small number of parameters while maintaining the original performance of the model. Fine-tuning training is performed on a specific task dataset to optimize the model parameters so that it can better perform specific tasks.

[0099] Step 33: Reinforcement Learning Training

[0100] The goal is to enable the model to self-adjust and optimize in a dynamic environment through policy optimization in a simulated environment. The main steps include:

[0101] Step 331: Build a simulation environment to simulate the interaction between the user and the system. The environment should include a variety of scenarios and tasks to ensure that the model can adapt to different dialogue scenarios;

[0102] Step 332: Direct policy optimization is performed using DPO technology. DPO enables the model to obtain the best policy in a simulation environment by directly optimizing the policy function.

[0103] Step 333: Use SimPO technology to perform simulation strategy optimization. SimPO further optimizes the model strategy by simulating user feedback and environmental changes.

[0104] Reinforcement learning training is carried out in a simulated environment. The model continuously adjusts and optimizes its own strategy through interaction with the environment to improve the dynamic adaptability and response quality of the dialogue system.

[0105] Step 4: Artificial Feedback Reinforcement Learning

[0106] The purpose of artificial feedback reinforcement learning is to further optimize the decision-making and response capabilities of the model by introducing artificial feedback. The specific steps mainly include:

[0107] Step 41: Collect user feedback information in actual applications, including user satisfaction scores, suggestions and opinions on system responses;

[0108] Step 42: Process and analyze the collected feedback information, extract valuable feedback content, and construct a feedback data set.

[0109] Step 43: Use the feedback dataset for model training and optimize the model’s decision-making and response capabilities by introducing an artificial feedback mechanism.

[0110] Reinforcement learning algorithms are used during the training process to enable the model to continuously adjust and optimize its own strategies based on feedback information. By continuously collecting and processing user feedback, the model is continuously optimized and adjusted to ensure that the model can adapt to changes in user needs and provide higher quality conversation services.

[0111] The above shows and describes the basic principles and main features of the present invention and the advantages of the present invention. It should be understood by those skilled in the art that the present invention is not limited to the above embodiments. The above embodiments and descriptions are only for explaining the principles of the present invention. Without departing from the spirit and scope of the present invention, the present invention may have various changes and improvements, which fall within the scope of the present invention to be protected. The scope of protection of the present invention is defined by the attached claims and their equivalents.

Claims

1. A psychological consultation method based on an AI dialogue system, characterized in that: include: Generate dialogue tasks and obtain user input; Use natural language processing technology to understand user input in real time and generate conversation scenarios; Generate the best consulting strategy based on the dialogue scenario through the psychological dialogue model; Monitor the emotional fluctuations and psychological state of users in the dialogue task in real time, and adjust the consultation strategy according to the monitoring results until the dialogue task is completed; The construction process of the psychological dialogue model includes the following steps: S1: Collect user conversation data through various methods for preprocessing, and use the preprocessed conversation data to build a psychological conversation dataset; S2: using Llama3-8B as a basic model, performing incremental pre-training on the psychological dialogue dataset to obtain a pre-trained psychological dialogue model; S3: fine-tuning the pre-trained psychological dialogue model based on the psychological dialogue dataset using LoRA technology to obtain a fine-tuned psychological dialogue model; S4: Build a simulation environment to simulate the interaction process between users and the system, and use DPO technology to directly optimize the strategy of the fine-tuned psychological dialogue model so that the model can obtain the best strategy in the simulation environment; S5: Use SimPO technology to simulate user feedback and environmental changes in dialogue tasks, optimize the simulation strategy of the fine-tuned psychological dialogue model, and obtain the enhanced psychological dialogue model.

2. The psychological consultation method based on the AI ​​dialogue system according to claim 1, characterized in that: The construction process of the psychological dialogue model also includes the following steps: S6: Obtain user feedback information, including user satisfaction ratings, suggestions, and opinions on system responses; S7: Process and analyze the collected feedback information and construct a feedback data set; S8: Use the feedback dataset for model training to obtain an iterative psychological dialogue model.

3. The psychological consultation method based on the AI ​​dialogue system according to claim 1 is characterized in that: The specific process of step S1 includes: the following: S11: Randomly extract a certain proportion of the original data from the psychological dialogue dataset, manually evaluate the quality of the extracted data, and remove noise and irrelevant content from the extracted data; S12: predefined data filtering rules, clean the data according to the predefined rules to remove obvious noise and irrelevant content; S13: Use text similarity algorithm to deduplicate the data and eliminate duplicate conversation data in the dataset; S14: Use the pre-trained large language model to score the acquired conversation data, set a scoring threshold, and retain the data within the scoring threshold based on the scoring results; S15: Establish a sensitive vocabulary list, perform matching filtering on the data, and remove or replace the conversation data containing sensitive words; S16: Manually proofread the conversation data processed by the above steps, and construct a psychological conversation data set with the proofread conversation data.

4. The psychological consultation system based on AI dialogue system is characterized by: It includes a user input acquisition module, a dialogue task generation module, a dialogue scene generation module, and a psychological dialogue model construction module; The dialogue task generation module is used to generate dialogue tasks for user psychological consultation; The user input acquisition module is used to acquire user input content; The dialogue scene generation module is used to understand user input in real time through natural language processing technology, and generate a dialogue scene according to the user input content; The psychological dialogue model building module is used to build a psychological dialogue model for user psychological consultation, and generate an optimal consultation strategy according to the dialogue scenario through the psychological dialogue model.

5. The psychological consultation system based on the AI ​​dialogue system according to claim 4 is characterized in that: The psychological dialogue model construction module includes a data acquisition module, a data preprocessing module, a dialogue model pre-training module, a dialogue model fine-tuning training module, and a dialogue model reinforcement training module; The data collection module is used to collect conversation data of users from different regions, age groups, genders, and cultural backgrounds; The data preprocessing module is used to preprocess the collected user dialogue data and construct a psychological dialogue data set using the preprocessed dialogue data; The dialogue model pre-training module is used to obtain a pre-trained psychological dialogue model by performing incremental pre-training on the psychological dialogue dataset using Llama3-8B as a basic model; The dialogue model fine-tuning training module is used to fine-tune the pre-trained psychological dialogue model based on the psychological dialogue dataset using LoRA technology to obtain a fine-tuned psychological dialogue model; The dialogue model reinforcement training module is used to perform reinforcement learning training on the constructed psychological dialogue model in a simulation environment; The dialogue model intensive training module includes a simulation environment construction submodule, a direct strategy optimization submodule, and a simulation strategy optimization submodule; The simulation environment construction submodule is used to construct a simulation environment for simulating the interaction process between the user and the system; The direct strategy optimization submodule is used to perform direct strategy optimization on the consultation strategy in the simulation environment using the DPO technology, so that the model can obtain the best strategy in each dialogue scenario in the simulation environment; The simulation strategy optimization submodule uses SimPO technology to perform simulation strategy optimization by simulating user feedback and environmental changes.

6. The psychological consultation system based on the AI ​​dialogue system according to claim 4 is characterized in that: The system also includes an artificial feedback reinforcement learning module, which optimizes the decision-making and response capabilities of the psychological dialogue model by introducing artificial feedback, including a user feedback information acquisition submodule, a user feedback information processing submodule, and an enhanced training submodule; The user feedback information acquisition submodule is used to acquire feedback information including user satisfaction scores, suggestions, and opinions on system responses; The user feedback information processing submodule is used to process and analyze the collected feedback information and construct a feedback data set; The reinforcement training submodule is used to use the feedback data set for model training to obtain an iterative psychological dialogue model.

7. The psychological consultation system based on the AI ​​dialogue system according to claim 5, characterized in that: The data preprocessing module includes a sampling evaluation submodule, a rule filtering submodule, a data deduplication submodule, a data evaluation submodule, a sensitive word filtering submodule, and a manual proofreading and screening submodule; The sampling evaluation submodule is used to randomly extract a certain proportion of the original data in the psychological dialogue data set, and manually evaluate the quality of the extracted data to remove noise and irrelevant content in the extracted data; The rule filtering submodule is used to predefine data filtering rules and clean the data according to the predefined rules to remove obvious noise and irrelevant content; The data deduplication submodule is used to perform deduplication processing on the data using a text similarity algorithm to eliminate duplicate conversation data in the data set; The data evaluation submodule is used to score the acquired dialogue data using a pre-trained large language model, set a scoring threshold, and retain data within the scoring threshold based on the scoring result; The sensitive word filtering submodule is used to establish a sensitive word list, perform matching filtering on the data, and remove or replace the conversation data containing sensitive words; The manual proofreading and screening submodule is used to manually proofread the dialogue data and construct a psychological dialogue data set with the proofread dialogue data.

8. The psychological consultation system based on the AI ​​dialogue system according to claim 4 is characterized in that: The system also includes a memory, a processor, and an electronic program stored in the memory and capable of running on the processor, wherein the processor can implement the steps of the method described in any one of claims 1 to 3 when running the electronic program.

Images