Depressive disorder detection report generation system based on retrieval enhancement generation technology and large model fine tuning

Abstract

The invention discloses a depressive disorder detection report generation system based on a retrieval enhancement generation technology and large model fine tuning, which is applied to the technical field of depressive disorder detection, and comprises a data generation module used for rewriting social media data into psychological counseling scene dialogue data by utilizing prompt guidance GPT-4 based on a psychological counseling dialogue prompt project; the AI psychological counselor construction module is used for finely adjusting the large language model and constructing an AI psychological counselor based on the psychological counseling scene dialogue data and the clinical interview data; and the AI psychologist construction module is used for performing similarity matching on dialogue contents of the user and an AI psychologist and a DSM-V judgment standard by utilizing a retrieval enhancement generation technology, guiding GPT-4 by utilizing a prompt, generating a depressive disorder detection report according to a detection report prompt project, performing fine adjustment based on a large language model, and constructing an AI psychologist. The depressive disorder detection and detection report generation precision is effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of depression disorder detection, and in particular to a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning. Background Art

[0002] In recent years, depression has become increasingly serious, affecting the physical and mental health of more and more people. Due to factors such as tight medical resources and limited detection methods, automatic depression detection technology is needed to provide assistance for timely and accurate detection. The emergence of large models has brought about excellent text generation capabilities, so depression detection based on large models has become a hot topic of research.

[0003] Existing automatic depression detection techniques can use deep learning to learn high-dimensional representations from initial features, thereby capturing more discriminative depression-related features. Therefore, exploring the depression detection capabilities of different network architectures and designing learning strategies have become the mainstream of automatic depression detection (ADD) methods in recent years. However, these methods are often limited by insufficient data in depression detection scenarios, resulting in limited generalization capabilities of the models. In addition, they usually model depression detection as a classification or regression task, which lacks further interpretability of the results. Large language models (LLMs) perform well in processing and understanding large amounts of text data, enabling them to achieve high accuracy and versatility in various language-related tasks, and are widely used in healthcare, role-playing and other fields. At the same time, LLMs have also promoted the development of automatic depression detection (ADD) methods. By providing explanatory information for samples, LLMs enrich the paradigm of ADD methods. However, they still face challenges in depression detection scenarios due to lack of expertise and the emergence of hallucinations. In addition, existing methods usually rely only on the patient's descriptive information, so there is a phenomenon of patient concealment or incomplete expression, and the importance of family members in the patient's depression detection process is not considered.

[0004] Therefore, how to provide a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning that can effectively improve the insufficiency of depression disorder detection data samples, improve the accuracy of depression disorder detection, and further explain the depression disorder detection results, while introducing relevant professional knowledge in the detection process, and considering patient and family consultation data is an urgent problem that technicians in this field need to solve. Summary of the invention

[0005] In view of this, the present invention proposes a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning.

[0006] In order to achieve the above object, the present invention adopts the following technical solution: A depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning, comprising: Data generation module: Based on the psychological counseling dialogue prompt project, it uses prompts to guide GPT-4 to rewrite social media data into psychological counseling scenario dialogue data; AI Psychological Counselor Building Module: used to fine-tune large language models and build AI psychological counselors based on psychological counseling scenario dialogue data and clinical interview data; AI psychologist building module: used to use retrieval-enhanced generation technology to match the conversation content between users and AI psychologists with the DSM-V judgment criteria, use prompts to guide GPT-4, generate a depression disorder detection report based on the detection report prompt project, and fine-tune based on a large language model to build an AI psychologist.

[0007] Optionally, based on the psychological counseling dialogue prompt project, use prompts to guide GPT-4 to rewrite social media data into psychological counseling scenario dialogue data, specifically: The prompt guides GPT-4 to gradually analyze the experience and emotions in the social media data according to the rules set in the dialogue reconstruction prompt, and rewrite the social media data into a dialogue between the user and the AI ​​psychologist; among them, the dialogue reconstruction prompt is as follows: ; in, Regenerate prompts for the conversation; For the described dialogue generation task, large language models are required to follow rules Social media data Rewrite as dialogue; rules ,include: , used to control the format of the beginning of the conversation, , used to require different roles to share content according to role allocation and define the style of the conversation background, , used to limit the length and number of rounds of dialogue; For the corresponding problem.

[0008] Optionally, fine-tune the large language model based on the psychological counseling scenario dialogue data and clinical interview data to build an AI psychological counselor, specifically: Calculate the cross entropy loss as follows: ; in, The first dataset is a fusion dataset of psychological counseling scene dialogue data and clinical interview data. In the conversation Cross entropy loss for round interactions; are model parameters; The first dataset is a fusion dataset of psychological counseling scene dialogue data and clinical interview data. In the conversation Wheel’s User Content; The first dataset is a fusion dataset of psychological counseling scene dialogue data and clinical interview data. In the conversation Model reply of round; A series of markers Composition The length of the target sequence; For a given input and the previous token in the sequence Target marker under the condition The conditional probability of Calculate the total cross entropy loss of the fused dataset as follows: ; in, is the total cross entropy loss of the fused dataset; is the fusion dataset; is the number of dialogues; is the number of interactions; Optimizing the objective function as follows: ; in, Parameters learned by the AI ​​psychologist.

[0009] Optionally, the retrieval enhancement generation technology is used to perform similarity matching between the conversation content between the user and the AI ​​psychologist and the DSM-V judgment criteria, specifically: Separate the symptom descriptions of different specific depression subtypes under the DSM-V criteria into different documents ;

[0010] For a given user content and a document , using the embedding network encoding into a high-dimensional feature vector, as follows: ; ; in, They are The high-dimensional feature vector of , is the dimension of the embedding space; To embed in the network; The first dataset is a fusion dataset of psychological counseling scene dialogue data and clinical interview data. User Content in a conversation; ; For the documents that describe symptoms of specific depression subtypes; Calculate user content using cosine similarity and each document The semantic similarity between them is as follows: ; Select the document index with the highest similarity score, as follows: ; in, The document with the highest similarity score is indexed.

[0011] Optionally, use prompts to guide GPT-4 to generate a depression disorder test report based on the test report prompt project, specifically: Based on the document with the highest similarity score and the corresponding dialogue input, severity standard and detection standard documents, the prompt is used to guide GPT-4 to generate a depression disorder detection report according to the thought chain steps set in the detection report generation prompt; among which, the detection report generation prompt is as follows: ; in, Prompt for test report generation; For the described detection report generation task, a large language model is required to generate a depressive disorder detection report based on the document with the highest similarity score, user content, and corresponding DSM-V judgment label according to the thought chain steps; The thought chain steps include: Step 1-2 guides the large language model to combine the dialogue input and test standard documents , gradually analyze the depressive symptoms that appear; Step 3 requires comparing the symptoms analyzed in Steps 1-2 with the provided test standard documents and severity criteria Step 4 instructs the large language model to predict the severity of depression based on the analysis information of steps 1-3; For the corresponding problem.

[0012] Optional: the content of the conversation between the user and the AI ​​psychologist, including: the content of the conversation between the patient and the patient's family and the AI ​​psychologist.

[0013] Optionally, fine-tune the large language model to build an AI psychologist, specifically: Based on the depression disorder detection report generated for each conversation, generate labels as follows: ; ; in, For labels; Instructions for removing documents with the highest similarity scores from detection report generation prompts and severity criteria and detection criteria documents; Enter for the conversation; for Corresponding depressive disorder test report; Generate tasks for the described detection reports; It is the step of thinking chain; For the corresponding question; Calculate the cross entropy loss for the depression detection report generation task as follows: in, Cross-entropy loss for the report generation task for depression detection; for Length; To give a hint in a given command , Dialogue Input And the sequence markers Under the condition of The conditional probability of are model parameters; Calculate the overall loss as follows: ; in, for overall loss; is the number of dialogues; Optimizing the objective function ,as follows: ; in, Parameters learned by the AI ​​psychologist.

[0014] It can be seen from the above technical solutions that compared with the prior art, the present invention proposes a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning. First, based on the rich experience and feeling information in the social media depression disorder data set, the psychological counseling dialogue prompt project is used to rewrite social media posts into psychological counseling scenario dialogue data, and combined with clinical interview data, the large language model is fine-tuned to build an AI psychological counselor, which effectively improves the problem of insufficient samples of depression disorder detection data, so that the fine-tuned large language model has the ability to meet the psychological counseling scenario. At the same time, the AI ​​psychological counselor constructed by the present invention is suitable for dual consultation of patients and their families. On the basis of the original patient consultation data, the patient family consultation data is introduced for subsequent depression disorder detection and test report generation, which further improves the subsequent depression disorder detection and test report generation accuracy. Secondly, the retrieval enhancement generation technology is used to match the conversation content between the user and the AI ​​psychologist with the DSM-V judgment criteria to provide professional detection knowledge for the detection report prompt project. Finally, based on the data with the highest similarity score, the prompt is used to guide GPT-4 to generate a depression disorder detection report according to the detection report prompt project, and fine-tune it based on the large language model to build an AI psychologist, which introduces relevant professional knowledge in the detection process and can further explain the results of depression disorder detection, including: binary classification, severity estimation, symptom and standard consistency, and subtype classification. At the same time, the fine-tuning of the large language model by the above-mentioned invention also further improves the accuracy of depression disorder detection and detection report generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0016] Figure 1 It is a schematic diagram of the processing flow of the system of the present invention.

[0017] Figure 2 A flowchart for constructing an AI psychological counselor of the present invention.

[0018] Figure 3 A schematic diagram of a flow chart for constructing the AI ​​psychologist of the present invention. DETAILED DESCRIPTION

[0019] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0020] Embodiment 1: Embodiment 1 of the present invention discloses a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning, such as Figure 1 As shown, including: Data generation module: used to guide GPT-4 to rewrite social media data into psychological counseling scenario dialogue data based on the psychological counseling dialogue prompt engineering. There are many challenges in using large language models (LLMs) to build AI psychological counselors. On the one hand, existing LLMs often fail to meet the dialogue needs in psychological counseling scenarios, and often generate repetitive or context-irrelevant responses. On the other hand, although instruction fine-tuning can enhance the performance of LLM in specific scenarios, psychological counseling data is difficult to obtain, and public datasets usually contain only dozens or hundreds of samples, which is far from enough for effective training of LLM. Compared with publicly available interview datasets, depression-related social media data is richer and provides rich experiences and emotions. Based on this, the present invention develops a psychological counseling dialogue prompt engineering, which is used to rewrite social media data into dialogues between users and AI psychological counselors, generate multi-round dialogue content between users and AI psychological counselors, and the content is derived from the rich and real experience and feelings in social media posts, thereby avoiding fictional content.

[0021] Based on the psychological counseling dialogue prompt project, prompts are used to guide GPT-4 to rewrite social media data into psychological counseling scenario dialogue data, specifically: The prompt guides GPT-4 to gradually analyze the experience and emotions in the social media data according to the rules set in the dialogue reconstruction prompt, and rewrite the social media data into a dialogue between the user and the AI ​​psychologist; among them, the dialogue reconstruction prompt is as follows: ; in, Regenerate prompts for the conversation; For the described dialogue generation task, large language models are required to follow rules to convert social media data Rewrite as dialogue; rules ,include: , used to control the format of the beginning of the conversation, , used to require different roles to share content according to role allocation and define the style of the conversation background, , used to limit the length and number of rounds of dialogue; For the corresponding questions, you are required to analyze and rewrite step by step.

[0022] Dialogue regeneration prompt , as shown in Table 1. Table 1 Dialogue regeneration prompts

[0023]

[0024] AI Psychological Counselor Building Module: It is used to fine-tune large language models based on psychological counseling scenario dialogue data and clinical interview data, and build AI psychological counselors, such as Figure 2 shown.

[0025] Based on the psychological counseling scenario dialogue data and clinical interview data, fine-tune the large language model and build an AI psychological counselor. Specifically: Calculate the cross entropy loss as follows: ; in, The first dataset is a fusion dataset of psychological counseling scene dialogue data and clinical interview data. In the conversation Cross entropy loss for round interactions; are model parameters; The first dataset is a fusion dataset of psychological counseling scene dialogue data and clinical interview data. In the conversation Wheel’s User Content; The first dataset is a fusion dataset of psychological counseling scene dialogue data and clinical interview data. In the conversation Model reply of round; A series of markers Composition The length of the target sequence; For a given input and the previous token in the sequence Target marker under the condition The conditional probability of Calculate the total cross entropy loss of the fused dataset as follows: ; in, is the total cross entropy loss of the fused dataset; To fuse the dataset, , For psychological counseling scene dialogue data, For clinical interview data; is the number of dialogues; is the number of interactions; Optimizing the objective function ,as follows: ; in, Parameters learned by the AI ​​psychologist.

[0026] AI psychologist building module: It is used to use retrieval enhancement generation technology to match the conversation content between users and AI psychologists with the DSM-V judgment criteria, use prompts to guide GPT-4, generate depression disorder detection reports based on the detection report prompt engineering, and fine-tune based on large language models to build AI psychologists, such as Figure 3 In order to use the knowledge in DSM-V to generate professional auxiliary test reports, this paper introduces the retrieval enhancement generation (RAG) technology. The depression disorder detection report containing the conversation can be used as high-quality data for instruction fine-tuning to build an AI psychologist. Since the detection criteria of DSM-V are mainly based on the patient's symptoms. Therefore, the present invention uses RAG technology to match the user's conversation content with the detection criteria of each depression subtype in DSM-V for similarity. The most similar semantic detection criteria are provided to GPT-4 to generate a report.

[0027] The search enhancement generation technology is used to match the conversation content between the user and the AI ​​psychologist with the DSM-V judgment criteria. Specifically: Separate the symptom descriptions of different specific depression subtypes under the DSM-V criteria into different documents ; For a given user content and a document , using the embedding network encoding into a high-dimensional feature vector, as follows: ; ; in, , They are , The high-dimensional feature vector of , is the dimension of the embedding space; To embed in the network; The first dataset is a fusion dataset of psychological counseling scene dialogue data and clinical interview data. User Content in a conversation; ; For the documents that describe symptoms of specific depression subtypes; Calculate user content using cosine similarity and each document The semantic similarity between them is as follows: ; Select the document index with the highest similarity score, as follows: ; in, The document with the highest similarity score is indexed.

[0028] Optional: the content of the conversation between the user and the AI ​​psychologist, including: the content of the conversation between the patient and the patient's family and the AI ​​psychologist.

[0029] Use prompts to guide GPT-4 and generate a depression disorder test report based on the test report prompt project, specifically: Based on the document with the highest similarity score and the corresponding dialogue input, severity standard and detection standard documents, the prompt is used to guide GPT-4 to generate a depression disorder detection report according to the thought chain steps set in the detection report generation prompt; among which, the detection report generation prompt is as follows: ; in, Prompt for test report generation; For the described detection report generation task, a large language model is required to generate a depressive disorder detection report based on the document with the highest similarity score, user content, and corresponding DSM-V judgment label according to the thought chain steps; The thought chain steps guide the large language model to break down complex tasks into smaller, more manageable steps, thereby producing more accurate and coherent output, including: Steps 1-2 guide the large language model to combine the dialogue input and test standard documents , gradually analyze the depressive symptoms that appear; Step 3 requires comparing the symptoms analyzed in Steps 1-2 with the provided test standard documents and severity criteria Step 4 instructs the large language model to predict the severity of depression based on the analysis information of steps 1-3; For the corresponding problem, LLM is required to output reports in a fixed JSON format.

[0030] Test report generation prompt , as shown in Table 2.

[0031] Table 2 Tips for generating test reports

[0032]

[0033] Fine-tune the large language model to build an AI psychologist, specifically: Based on the depression disorder detection report generated for each conversation, generate labels as follows: ; ; in, For labels; Instructions for removing documents with the highest similarity scores from detection report generation prompts and severity criteria and detection criteria documents; Enter for the conversation; for Corresponding depressive disorder test report; Generate tasks for the described detection reports; It is the step of thinking chain; For the corresponding question; Calculate the cross entropy loss for the depression detection report generation task as follows: ; in, Cross-entropy loss for the report generation task for depression detection; for Length; To give a hint in a given command , Dialogue Input And the sequence markers Under the condition of The conditional probability of are model parameters; Calculate the overall loss as follows: ; in, for overall loss; is the number of dialogues; Optimizing the objective function ,as follows: ; in, Parameters learned by the AI ​​psychologist.

[0034] Finally, the LLM-based AI psychologist of the present invention is able to generate a professional test report containing four attributes: binary classification, severity estimation, symptom-criteria consistency, and subtype classification, thereby overcoming the limitations of traditional automatic depression detection (ADD) methods.

[0035] The embodiment of the present invention discloses a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning. First, based on the rich experience and feeling information in the social media depression disorder data set, the psychological counseling dialogue prompt project is used to rewrite social media posts into psychological counseling scenario dialogue data, and combined with clinical interview data, the large language model is fine-tuned to build an AI psychological counselor, which effectively improves the problem of insufficient samples of depression disorder detection data, so that the fine-tuned large language model has the ability to meet the psychological counseling scenario. At the same time, the AI ​​psychological counselor constructed by the present invention is suitable for dual consultation of patients and their families. On the basis of the original patient consultation data, the patient family consultation data is introduced for subsequent depression disorder detection and test report generation, which further improves the subsequent depression disorder detection and test report generation accuracy. Secondly, the retrieval enhancement generation technology is used to match the conversation content between the user and the AI ​​psychologist with the DSM-V judgment criteria to provide professional detection knowledge for the detection report prompt project. Finally, based on the data with the highest similarity score, the prompt is used to guide GPT-4 to generate a depression disorder detection report according to the detection report prompt project, and fine-tune it based on the large language model to build an AI psychologist, which introduces relevant professional knowledge in the detection process and can further explain the results of depression disorder detection, including: binary classification, severity estimation, symptom and standard consistency, and subtype classification. At the same time, the fine-tuning of the large language model by the above-mentioned invention also further improves the accuracy of depression disorder detection and detection report generation.

[0036] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0037] The above description of the disclosed embodiments enables one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning, characterized in that: include: Data generation module: Based on the psychological counseling dialogue prompt project, it uses prompts to guide GPT-4 to rewrite social media data into psychological counseling scenario dialogue data; AI Psychological Consultant Construction Module: used to fine-tune the large language model and construct an AI psychological consultant based on the psychological consultation scenario dialogue data and clinical interview data; AI psychologist building module: used to use retrieval enhancement generation technology to match the conversation content between the user and the AI ​​psychologist with the DSM-V judgment criteria, use the prompts to guide GPT-4, generate a depressive disorder detection report according to the detection report prompt project, and fine-tune based on a large language model to build an AI psychologist.

2. According to claim 1, a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning is characterized in that: Based on the psychological counseling dialogue prompt project, prompts are used to guide GPT-4 to rewrite social media data into psychological counseling scenario dialogue data, specifically: The prompt guides GPT-4 to gradually analyze the experience and emotions in the social media data according to the rules set in the dialogue regeneration prompt, and rewrite the social media data into a dialogue between the user and the AI ​​psychologist; wherein the dialogue regeneration prompt is as follows: ； in, Regenerate prompts for the conversation; For the described dialogue generation task, large language models are required to follow rules The social media data Rewrite as dialogue; rules ,include: , used to control the format of the beginning of the conversation, , used to require different roles to share content according to role allocation and define the style of the conversation background, , used to limit the length and number of rounds of dialogue; For the corresponding problem.

3. According to claim 1, a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning is characterized in that: Based on the psychological counseling scenario dialogue data and clinical interview data, fine-tune the large language model and build an AI psychological counselor, specifically: Calculate the cross entropy loss as follows: ； in, The first In the conversation Cross entropy loss for round interactions; are model parameters; The first In the conversation User Content of Wheels; The first In the conversation Model reply of round; A series of markers Composition The length of the target sequence; For a given input and the previous token in the sequence Target marker under the condition The conditional probability of The total cross entropy loss of the fused data set is calculated as follows: ； in, is the total cross entropy loss of the fused data set; is the fused data set; is the number of dialogues; is the number of interactions; Optimizing the objective function as follows: ； in, Parameters learned by the AI ​​psychologist.

4. According to claim 1, a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning is characterized in that: The search enhancement generation technology is used to match the conversation content between the user and the AI ​​psychologist with the DSM-V judgment criteria for similarity, specifically: Separate the symptom descriptions of different specific depression subtypes under the DSM-V criteria into different documents ; For a given user content and a document , using the embedding network encoding into a high-dimensional feature vector, as follows: ； ； in, They are The high-dimensional feature vector of , is the dimension of the embedding space; To embed in the network; The first User Content in a conversation; ; For the documents that describe symptoms of specific depression subtypes; Calculate user content using cosine similarity and each document The semantic similarity between them is as follows: ； Select the document index with the highest similarity score, as follows: ； in, The document with the highest similarity score is indexed.

5. According to claim 1, a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning is characterized in that: The prompt is used to guide GPT-4 to generate a depressive disorder test report according to the test report prompt project, specifically: Based on the document with the highest similarity score and the corresponding dialogue input, severity standard and detection standard document, the prompt is used to guide GPT-4 to generate a depressive disorder detection report according to the thought chain steps set in the detection report generation prompt; wherein the detection report generation prompt is as follows: ； in, Prompt for test report generation; For the described detection report generation task, a large language model is required to generate a depressive disorder detection report based on the document with the highest similarity score, user content, and corresponding DSM-V judgment label according to the thought chain steps; The thought chain steps include: Step 1-2 guides the large language model to combine the dialogue input and test standard documents , gradually analyze the depressive symptoms that appear; Step 3 requires comparing the symptoms analyzed in Steps 1-2 with the provided test standard documents and severity criteria Step 4 instructs the large language model to predict the severity of depression based on the analysis information of steps 1-3; For the corresponding problem.

6. According to claim 1, a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning is characterized in that: The content of the conversation between the user and the AI ​​psychological counselor includes: the content of the conversation between the patient and the patient's family and the AI ​​psychological counselor.

7. According to claim 1, a depression disorder detection report generation system based on retrieval enhancement generation technology and large model fine-tuning is characterized in that: Fine-tune the large language model to build an AI psychologist, specifically: Based on the depression disorder detection report generated for each conversation, generate labels as follows: ； ； in, For labels; Instructions for removing documents with the highest similarity scores from detection report generation prompts and severity criteria and detection criteria documents; Enter for the conversation; for Corresponding depressive disorder test report; Generate tasks for the described detection reports; It is the step of thinking chain; For the corresponding question; Calculate the cross entropy loss for the depression detection report generation task as follows: in, Cross-entropy loss for the report generation task for depression detection; for Length; To give a hint in a given command , Dialogue Input And the sequence markers Under the condition of The conditional probability of are model parameters; Calculate the overall loss as follows: ； in, for overall loss; is the number of dialogues; Optimizing the objective function ,as follows: ； in, Parameters learned by the AI ​​psychologist.

Images