Mental health human-computer interaction assessment method and system, electronic device and storage medium

Abstract

This invention relates to the field of mental health technology, and more particularly to a human-computer interactive assessment method, system, electronic device, and storage medium for mental health. This invention embeds scale items into a natural human-computer dialogue process and analyzes user responses item by item. After all scale assessments are completed, an emotion-triggered task is triggered uniformly, separating standardized scale assessment from the exploration of deep psychological mechanisms in the process, while integrating the results. The scale assessment results reflect symptom severity, while psychological characteristic information reflects cognitive patterns, emotion regulation strategies, and intrinsic awareness. This avoids interference from task execution on scale scoring and overcomes the limitations of the traditional single total score dimension of scales. Ultimately, by integrating both, a multi-dimensional mental health assessment result is generated, achieving a dual-track assessment of "symptom severity" and "psychological characteristic patterns." This provides a multi-dimensional assessment framework for a refined understanding of an individual's psychological functional state, thereby providing richer targeted information for subsequent interventions.

Description

Technical Field

[0001] This invention relates to the field of mental health technology, and in particular to a method, system, electronic device, and storage medium for human-computer interaction assessment of mental health. Background Technology

[0002] Currently, the standard practice for digital psychological assessment is to directly port standardized scales such as PHQ-9 and GAD-7 to online platforms, collecting users' subjective self-assessment data in a fixed question-and-answer format. However, this electronic scale model has revealed multiple limitations in practical applications. On the one hand, the mechanical check-and-answer format is not only prone to social expectation bias but also fails to address the deeper psychological mechanisms behind symptoms, such as cognitive patterns and emotion regulation strategies. More importantly, the scale items use standardized medical terminology, while users are accustomed to using colloquial expressions such as "I'm so annoyed" and "I don't want to do anything" in natural conversation. There is a significant semantic gap between the two, making it difficult for existing methods to automatically understand and translate users' true expressions.

[0003] To address these limitations, recent years have seen research attempts to apply Natural Language Processing (NLP) and Large Language Models (LLM) to mental health screening. However, most of these approaches are still in the exploratory stage and suffer from issues such as insufficient validation of screening functions and inconsistent evaluation mechanisms. More critically, existing approaches generally lack dynamic quantitative assessment methods for deeper psychological indicators such as emotion regulation strategies and interoceptive awareness. This results in an inability to comprehensively reflect an individual's true psychological state, and a significant discrepancy remains between assessment results and clinical reality. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, system, electronic device and storage medium for human-computer interaction assessment of mental health, so as to solve the above-mentioned technical problem.

[0005] The technical solution of this invention to solve the above-mentioned technical problems is as follows: A human-computer interactive assessment method for mental health, comprising: acquiring interaction data generated by a user's responses to items on a mental health scale; for each piece of interaction data, performing natural language understanding processing on the interaction data to obtain the natural language parsing result corresponding to the interaction data; matching the target scale item corresponding to each piece of interaction data according to the natural language parsing result corresponding to each piece of interaction data, and determining the score corresponding to each target scale item to obtain the scale assessment result of the user; after completing the assessment of the mental health scale items, outputting an emotional triggering task to the user, and acquiring the task interaction data generated by the user during the execution of the emotional triggering task; extracting the user's psychological characteristic information based on the task interaction data; and generating the user's mental health assessment result based on the scale assessment result and the psychological characteristic information.

[0006] The beneficial effects of this invention are as follows: By embedding scale items into a natural human-computer dialogue process and analyzing user responses item by item, and triggering an emotion-triggered task uniformly after all scale assessments are completed, this invention separates standardized scale evaluation from the exploration of deep psychological mechanisms in terms of process but integrates them in terms of results. The scale assessment results reflect the severity of symptoms, while psychological characteristic information reflects cognitive patterns, emotion regulation, and interoceptive awareness. This avoids interference from task execution on scale scoring and overcomes the limitations of the traditional single total score dimension of scales. Ultimately, by integrating the two, a multi-dimensional mental health assessment result is generated, enabling the assessment to reflect both the severity of symptoms and reveal individual psychological characteristic patterns, significantly improving the comprehensiveness and clinical reference value of the assessment.

[0007] Based on the above technical solution, the present invention can be further improved as follows.

[0008] Furthermore, for each piece of interactive data, natural language understanding processing is performed on the interactive data to obtain the natural language parsing result corresponding to the interactive data, including: inputting the interactive data and the retrieval enhancement context into the natural language understanding model, and constraining the natural language understanding model to output the natural language parsing result corresponding to the interactive data according to the preset data structure fields through prompt words; wherein, the retrieval enhancement context includes at least one of the following: tag knowledge base, symptom definition, scale item description and similar historical cases.

[0009] Furthermore, the natural language understanding model is trained in the following manner: A training corpus is acquired, comprising at least one of the following: psychological interview corpus, PHQ-9 questionnaire corpus, GAD-7 questionnaire corpus, open-ended response corpus based on emotion-triggered tasks, cognitive behavioral feature questionnaire corpus, emotion regulation strategy questionnaire corpus, and interoceptive awareness questionnaire corpus; each piece of data in the training corpus is manually annotated to obtain structured annotation data corresponding to each piece of data; using the data in the training corpus as input and the structured annotation data corresponding to the data as the target output, a pre-set large language model is adjusted and trained to obtain the natural language understanding model.

[0010] Further, the step of matching the target scale item corresponding to each piece of interaction data according to the natural language parsing result corresponding to each piece of interaction data, and determining the score corresponding to each target scale item to obtain the scale evaluation result of the user, includes: for each piece of interaction data, matching the target scale item corresponding to the interaction data according to the symptom information in the natural language parsing result corresponding to the interaction data through a pre-trained emotion symptom mapping model; for each piece of interaction data, determining the score of the target scale item corresponding to the interaction data according to the frequency information in the natural language parsing result corresponding to the interaction data and a preset frequency score mapping relationship; maintaining an independent state for each scale item, and controlling the writing of the score of the target scale item corresponding to each piece of interaction data according to the current state of each scale item; after the score of each scale item is written, accumulating the scores of each scale item to obtain a total score, and generating a screening level according to the total score; obtaining the scale evaluation result of the user according to the scores of each scale item, the total score and the screening level.

[0011] Furthermore, for each piece of interaction data, based on the symptom information in the natural language parsing result corresponding to the interaction data, a pre-trained emotion symptom mapping model is used to match the target scale item corresponding to the interaction data. This includes: inputting the symptom information in the natural language parsing result corresponding to the interaction data into the pre-trained emotion symptom mapping model, which includes an encoding layer, a matching layer, a classification layer, and a rule correction layer; encoding the symptom information in the natural language parsing result corresponding to the interaction data into a symptom vector through the encoding layer; calculating the similarity between the symptom vector and a preset standard symptom label vector and a historical labeled case vector through the matching layer, and determining a candidate symptom set based on the obtained similarity calculation results; inputting the symptom vector, the current interview context vector, and the vectors corresponding to each candidate symptom in the candidate symptom set into a multi-label classification network through the classification layer to generate a probability value corresponding to each candidate symptom in the candidate symptom set; and determining the matching symptom label based on the probability value corresponding to each candidate symptom through the rule correction layer, and determining the target scale item corresponding to the symptom label based on the preset symptom scale item mapping relationship.

[0012] Furthermore, the emotion-triggered task includes at least one of a positive recall task, a negative recall task, a future imagination task, and an attention task; the positive recall task is used to assess a user's ability to process positive emotions; the negative recall task is used to assess a user's cognitive processing bias toward negative information; the future imagination task is used to assess a user's cognitive processing bias toward uncertain future situations; and the attention task is used to assess a user's attentional control ability.

[0013] Furthermore, the psychological characteristics information includes cognitive behavioral characteristics, emotion regulation strategies, and interoceptive awareness.

[0014] To address the aforementioned technical problems, the present invention also provides a human-computer interactive assessment system for mental health, comprising: The data acquisition module is used to acquire the interaction data generated by users answering questions on the mental health scale item by item; The data parsing module is used to perform natural language understanding processing on each piece of interactive data to obtain the natural language parsing result corresponding to the interactive data. The scale assessment module is used to match the target scale item corresponding to each interaction data according to the natural language parsing result corresponding to each interaction data, and determine the score corresponding to each target scale item, so as to obtain the scale assessment result of the user. The task interaction module is used to output emotional triggering tasks to users after they have completed the psychological health scale items assessment, and to obtain the task interaction data generated by the users during the execution of the emotional triggering tasks. The information extraction module is used to extract the user's psychological characteristic information based on the task interaction data; The results generation module is used to generate the user's mental health assessment results based on the scale assessment results and the psychological characteristic information.

[0015] To address the aforementioned technical problems, the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the aforementioned human-computer interaction assessment method for mental health.

[0016] To address the aforementioned technical problems, the present invention also provides a non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the aforementioned human-computer interaction assessment method for mental health. Attached Figure Description

[0017] Figure 1 This is a flowchart of the human-computer interaction assessment method for mental health of the present invention; Figure 2 This is a flowchart of the active recall task according to Embodiment 1 of the present invention; Figure 3 This is a flowchart of the negative memory task according to Embodiment 1 of the present invention; Figure 4 This is a flowchart of a future imagination task according to Embodiment 1 of the present invention; Figure 5 This is a flowchart of the attention task according to Embodiment 1 of the present invention; Figure 6 This is a schematic diagram of the human-computer interaction assessment system for mental health of the present invention; Figure 7 This is a schematic diagram of the electronic device of the present invention. Detailed Implementation

[0018] The principles and features of the present invention are described below. The examples given are only for explaining the present invention and are not intended to limit the scope of the present invention.

[0019] Example 1 like Figure 1 As shown, this embodiment provides a human-computer interaction assessment method for mental health, including: S101. Collect the interaction data generated by the user's answers to the items of the mental health scale one by one; S102. For each piece of interactive data, perform natural language understanding processing on the interactive data to obtain the natural language parsing result corresponding to the interactive data. S103. Match the target scale item corresponding to each interaction data according to the natural language parsing result corresponding to each interaction data, and determine the score corresponding to each target scale item to obtain the user's scale evaluation result. S104. After completing the psychological health scale item assessment, output the emotional trigger task to the user and obtain the task interaction data generated by the user in the process of performing the emotional trigger task. S105. Extract users' psychological characteristic information based on task interaction data; S106. Based on the scale assessment results and psychological characteristic information, generate the user's mental health assessment results.

[0020] Specifically, the system interacts with users through a front-end interface such as a WeChat mini-program or app, following a questionnaire-based process. First, standardized questions related to each item of the mental health scale, such as PHQ-9 and GAD-7, are posed to the user. The user answers in natural language, generating interactive data. For each piece of interactive data, the system performs natural language understanding processing, outputting structured symptom and frequency information. Based on the parsing results, it matches the corresponding target scale item and determines the score.

[0021] In this embodiment, symptom information refers to symptom tags identified from the user's natural language responses that are related to items on the mental health scale, including but not limited to worry about loss of control, excessive worry, irritability, loss of interest, low mood, fatigue, and sleep problems. Symptom information is used to subsequently match corresponding target scale items. Frequency information refers to the frequency of occurrence of the symptoms described by the user, used for subsequent scale scoring. Frequency information includes explicit frequency categories (e.g., none at all, a few days, more than half the days, almost every day, etc.) or vague frequency descriptions (e.g., frequently, quite a few, rather often, always like this, etc.). When frequency information is vague, a specific frequency category is obtained through follow-up questions.

[0022] After all items on the mental health scale have been assessed, an emotional trigger task is output to the user, and task interaction data of the user during the task execution process is obtained.

[0023] In this embodiment, task interaction data refers to multimodal interaction data generated by users during the execution of emotion-triggered tasks, including but not limited to: the user's natural language expression in the free recall or free imagination stage, the user's answers to questions related to emotion regulation strategies and intrinsic awareness in the retrospective interview stage, and behavioral data such as reaction time and accuracy in attention tasks.

[0024] The system extracts users' cognitive and behavioral characteristics, emotion regulation strategies, and intrinsic awareness from task interaction data. Finally, it integrates the scale assessment results and psychological characteristic information to generate a multidimensional mental health assessment result for the user.

[0025] Optionally, in the embodiments, for each piece of interactive data, natural language understanding processing is performed on the interactive data to obtain the natural language parsing result corresponding to the interactive data, including: inputting the interactive data and the retrieval enhancement context into the natural language understanding model, and constraining the natural language understanding model to output the natural language parsing result corresponding to the interactive data according to the preset data structure fields through prompt words; wherein, the retrieval enhancement context includes at least one of the following: tag knowledge base, symptom definition, scale item description and similar historical cases.

[0026] Natural Language Understanding (NLU) models are used to convert users' spoken natural language responses into structured, labeled data. NLU models do not output any diagnostic conclusions, but only structured understanding results that can be directly used in subsequent steps.

[0027] Specifically, during the inference phase, the NLU model receives interaction data generated by the user's responses to scale items. This interaction data, along with retrieval-enhanced context, is then input into the natural language understanding model. The role of retrieval enhancement is to incorporate relevant knowledge as context, along with the user input, to form the model's input sequence before or during inference. This helps the model better understand and judge the user's expressions, providing a clear basis for its judgments.

[0028] The natural language understanding model is constrained by prompt words to output structured data containing symptom and frequency information according to predefined data structure fields, without outputting arbitrary diagnostic conclusions. The predefined data structure fields include, but are not limited to: emotion category, symptom label, mapped scale item, frequency hint, need follow-up, risk flag, and evidence.

[0029] For example, given the input "Current entry is GAD-2, user says: I'm so annoyed lately, my mind is racing, I'm always worried something will happen", the model's target output is structured data: "Emotion = Anxiety, Symptoms = Worrying about losing control, Irritability, Mapping entry = GAD-2, GAD-6, Frequency cue = Unclear, Need to ask further = true, Risk indicator = false".

[0030] Furthermore, after the model generates structured data, the structured data is validated. Validation includes at least one of the following: format validation (such as JSON format validation), label validity validation (checking whether the label is in the preset set), symptom scale item mapping validation (checking whether the mapping relationship conforms to expert rules), frequency field validation, and risk word validation (checking whether there are self-harm risk words that the model has not labeled as risk).

[0031] When the verification fails, perform at least one of the following operations: automatically correct the structured data, re-invoke the natural language understanding model, or trigger manual review.

[0032] Optionally, in this embodiment, the natural language understanding model is trained in the following manner: A training corpus is acquired, comprising at least one of the following: psychological interview (MINI interview) corpus, PHQ-9 scale question-and-answer corpus, GAD-7 scale question-and-answer corpus, open-ended response corpus based on emotion-triggered tasks, cognitive behavioral feature scale question-and-answer corpus, emotion regulation strategy scale question-and-answer corpus, and interoceptive awareness scale question-and-answer corpus; each corpus in the training corpus is manually annotated to obtain structured annotation data corresponding to each corpus; the corpus in the training corpus is used as input, and the structured annotation data corresponding to the corpus is used as the target output to adjust and train a preset large language model to obtain the natural language understanding model.

[0033] Specifically, when manually annotating each corpus entry in the training corpus, psychological professionals annotate according to unified annotation standards to obtain structured annotated data for each corpus entry. Annotation fields include: user's original words, current interview stage, current scale item, emotion category, symptom label, candidate scale item, frequency clues, whether follow-up questions are needed, risk markers, and evidence fragments.

[0034] In this embodiment, all subjects were independently labeled by psychiatrists according to the DSM-5 diagnostic criteria, and data quality was ensured through consistency checks. The labeling results served as the gold standard for training and validating the multimodal model.

[0035] The input includes task description, current interview context, user's original expressions, current scale items, and a set of optional labels. The output is a fixed JSON structure. Using the corpus from the training corpus as input and the corresponding structured labeled data as the target output, a pre-defined large language model is fine-tuned and trained to obtain a natural language understanding model.

[0036] In some embodiments, the large language model employs an LLaMA Chinese augmentation model, using the QLoRA parameter fine-tuning method to update only a small number of adapter parameters, reducing training costs while preserving the original language capabilities of the base model. The training objective is to enable the model to learn the ability to extract structured information in psychological interview scenarios, rather than learning direct diagnosis.

[0037] Optionally, in this embodiment, the target scale item corresponding to each piece of interactive data is matched according to the natural language parsing result corresponding to each piece of interactive data, and the score corresponding to each target scale item is determined to obtain the user's scale evaluation result. This includes: for each piece of interactive data, matching the target scale item corresponding to the interactive data according to the symptom information in the natural language parsing result corresponding to the interactive data through a pre-trained emotion symptom mapping model; for each piece of interactive data, determining the score of the target scale item corresponding to the interactive data according to the frequency information in the natural language parsing result corresponding to the interactive data and the preset frequency score mapping relationship; maintaining an independent state for each scale item, and controlling the writing of the score of the target scale item corresponding to each piece of interactive data according to the current state of each scale item; after the score of each scale item is written, accumulating the scores of each scale item to obtain a total score, and generating a screening level according to the total score; and obtaining the user's scale evaluation result according to the scores of each scale item, the total score, and the screening level.

[0038] Optionally, in this embodiment, for each piece of interaction data, based on the symptom information in the natural language parsing result corresponding to the interaction data, a pre-trained emotion symptom mapping model is used to match the target scale item corresponding to the interaction data. This includes: inputting the symptom information in the natural language parsing result corresponding to the interaction data into the pre-trained emotion symptom mapping model, which includes an encoding layer, a matching layer, a classification layer, and a rule correction layer; encoding the symptom information in the natural language parsing result corresponding to the interaction data into symptom vectors through the encoding layer; calculating the similarity between the symptom vectors and preset standard symptom label vectors and historical labeled case vectors through the matching layer, and determining a candidate symptom set based on the obtained similarity calculation results; inputting the symptom vectors, the current interview context vector, and the vectors corresponding to each candidate symptom in the candidate symptom set into a multi-label classification network through the classification layer to generate probability values ​​corresponding to each candidate symptom in the candidate symptom set; and determining the matched symptom label based on the probability values ​​corresponding to each candidate symptom through the rule correction layer, and determining the target scale item corresponding to the symptom label based on the preset symptom scale item mapping relationship.

[0039] The emotion symptom mapping model employs a hybrid network structure consisting of an encoding layer, a matching layer, a classification layer, and a rule correction layer. Specifically, the encoding layer uses the Embedding module of BERT (Bidirectional Encoder Representations from Transformers) as a text encoder to encode symptom information into vector representations.

[0040] In the matching layer, the symptom vector is compared with the preset standard symptom label vector and the historical labeled case vector, respectively, and the candidate symptom set is determined based on the similarity calculation results. In this embodiment, after calculating the similarity with the preset standard symptom label vector and the historical labeled case vector, the final similarity is calculated by weighting, and then multiple candidate symptoms are determined according to the similarity from high to low.

[0041] In the classification layer, the symptom vector, the current interview context vector, and the vectors corresponding to each candidate symptom in the candidate symptom set are input into the multi-label classification network to generate a probability value for each candidate symptom in the candidate symptom set. The probability value is used to indicate the confidence level of the candidate symptom being matched.

[0042] The multi-label classification network uses a pre-trained Chinese language model as the text encoder. User input is segmented and then fed into the encoder. The BERT model adds a special classification token [CLS] (Classification Token) at the beginning of each input sequence, and the encoder output vector corresponding to this token serves as the semantic representation of the entire sentence. This is then passed through fully connected layers, Dropout layers, and a Sigmoid multi-label output layer to obtain the independent probabilities of each symptom label. Each output node corresponds to a standard symptom label or scale item related to a symptom, such as worry and loss of control, excessive worry, irritability, loss of interest, low mood, fatigue, sleep problems, etc. Since the same user expression may contain multiple symptoms simultaneously, a Sigmoid independent output is used.

[0043] In the rule correction layer, matching symptom labels are determined based on the probability values ​​corresponding to each candidate symptom, and target scale entries corresponding to the symptom labels are determined based on a preset symptom scale entry mapping relationship. Specifically, the classification results are filtered, corrected, and sorted by confidence level according to the expert-preset symptom scale entry mapping table. When the probability value (confidence level) is lower than a preset threshold, a follow-up question flag is output to trigger a follow-up question to the user to obtain supplementary information.

[0044] In this embodiment, the emotion symptom mapping model is trained in the following way: Construct a manually labeled dataset. Each sample includes the user's original expression, context, emotion category, symptom label, corresponding scale item, and whether follow-up questions are needed.

[0045] During the training phase, multi-hot label vectors were constructed using manually annotated psychological interview corpora, and supervised training was conducted using the binary cross-entropy loss function. This enabled the multi-label classification network to identify the probabilities of multiple symptom labels from user input text. This loss function directly affects the raw scores output by the model, internally converting them into probability values ​​using the sigmoid function before calculating the cross-entropy loss between the model and the true labels. This merging of sigmoid and cross-entropy calculations aims to improve the numerical stability of the training.

[0046] A contrastive learning model is used to train the semantic matching model, bringing semantically similar user expressions closer to the correct symptom labels and distancing irrelevant or easily confused labels. For example, "can't stop thinking" is brought closer to "worried about losing control" and further away from "loss of interest". For polysemous expressions such as "tired", "annoyed", and "can't hold on", ambiguous samples are set in the training set to make the model output a lower confidence level, and follow-up questions are triggered by subsequent decision trees.

[0047] In this embodiment, a decision tree engine is used for dialogue management and state control. The decision tree engine dynamically determines the next interaction action based on the structured labels output by the NLU model and the current evaluation progress. Specifically, the decision tree engine executes the following decision logic: Frequency Information Judgment: When the system detects that the frequency information of the current scale item is unclear (such as the user's vague descriptions like "frequently," "quite often," "relatively frequently," or "always like this"), it generates follow-up questions to obtain the user's clear frequency category. For example, it might ask, "In the past two weeks, has this situation been more or less the same on several days, more than half the days, or almost every day?" Item switching: After the current scale item has been scored, the system automatically switches to the next unassessed scale item.

[0048] Risk Response: When a risk indicator is detected, a safety assessment or manual intervention process is triggered.

[0049] Optionally, in the embodiments, the scorer maintains an independent state for each scale entry and controls the writing of the score of the target scale entry corresponding to each interactive data according to the current state of each scale entry. This includes: the scorer is implemented by combining a deterministic rule engine, a session state machine and database records, and is essentially an interpretable computation module that runs according to the PHQ-9 and GAD-7 standard scoring rules.

[0050] Specifically, a rating scale table is first created in the database, with independent fields set for the 9 items of PHQ-9 and the 7 items of GAD-7. Each item includes the following fields: item number, item name, whether it has been hit, frequency category, score, evidence text, source round, model confidence, rating status, and update time.

[0051] For example, for the GAD-2 entry, the following can be recorded: item_code=GAD-2, frequency_category=nearly_every_day, score=3, evidence_text="almost every day", source_turn_id=12, score_status=completed.

[0052] The rating scale has a pre-set frequency normalization rule table and score mapping table, which uniformly maps the frequency of symptoms experienced by users over the past two weeks to standard scale scores. The mapping rules are shown in Table 1. Table 1 Mapping Rules Table When the input frequency category belongs to the aforementioned explicit category and the confidence level reaches the preset threshold, the scorer calls the rule function to complete the single-item scoring, that is, it queries the corresponding score according to the frequency_score_map and writes it into the score field of that item. When the frequency information is unclear or the confidence level is lower than the preset threshold, scoring is not performed temporarily, and a follow-up question is output to obtain the explicit frequency information.

[0053] The scorer maintains an independent state for each item, including states such as not started, mentioned, requires follow-up inquiry, frequency confirmed, completed, and risk pending. The scorer controls the writing of individual scores for each item based on its current state. The specific logic for state control is as follows: When an item is in the "Not Started" state, it means that the item has not yet been touched; when an item is in the "Mentioned" state, it means that the user has mentioned the relevant symptoms, but the rating has not yet been completed; when an item is in the "Follow-up Questions Required" state, it means that further frequency information needs to be obtained; when an item is in the "Frequency Confirmed" state, it means that the frequency information has been confirmed and is awaiting rating; when an item is in the "Completed" state, it means that the item has been successfully rated, but this does not overwrite the item's existing score, and is only updated and retained when the user explicitly corrects it; when an item is in the "Risk Pending Processing" state, it means that a risk rule has been triggered and special handling is required.

[0054] The scorer determines whether to allow writing a score based on the current status, preventing the same entry from being scored repeatedly or incorrectly overwritten. After scoring each entry, the scores for each entry are summed to obtain a total score, and a screening level is generated based on the total score.

[0055] Specifically, the PHQ-9 total score is obtained by summing the scores of 9 items, with a value range of 0-27. According to the preset stratification rules: 0-4 points are normal or very mild, 5-9 points are mild, 10-14 points are moderate, 15-19 points are moderate to severe, and 20-27 points are severe.

[0056] The GAD-7 total score is obtained by summing the scores of 7 items, with a value range of 0-21. According to the preset stratification rules: 0-4 points are normal or very mild, 5-9 points are mild, 10-14 points are moderate, and 15-21 points are severe.

[0057] The scorer also performs risk assessment. When the score of a preset risk item is greater than zero or a risk flag is detected, a risk status is triggered, pausing the current assessment process and switching to a safety assessment or manual intervention process.

[0058] In this embodiment, the preset risk item is PHQ-9 item 9 (item involving self-harm or suicidal ideation). Specifically, for PHQ-9 item 9 involving self-harm or suicidal ideation, as long as the score of the item is greater than 0, or the NLU model outputs the risk flag risk_flag=true, a risk state is triggered, the ordinary scale process is paused, and the process enters the safety assessment or manual intervention process.

[0059] Optionally, in the embodiments, the emotion-triggered task includes at least one of a positive recall task, a negative recall task, a future imagination task, and an attention task; the positive recall task is a task used to assess the user's ability to process positive emotions; the negative recall task is a task used to assess the user's cognitive processing bias towards negative information; the future imagination task is a task used to assess the user's cognitive processing bias towards uncertain future situations; and the attention task is a task used to assess the user's attention control ability.

[0060] This embodiment designed four emotion-triggered tasks based on the RDoC research framework. The positive recall task is based on the positive valence system, the negative recall and future imagination tasks on the negative valence system, and the attention task on the cognitive system. The aim is to activate specific cognitive processes, thereby capturing the user's authentic reactions in a natural state. Traditional scales ask "How do you think you are?", while emotion-triggered tasks measure "How you would actually be?", providing data for subsequent assessment and the development of personalized intervention plans.

[0061] In this embodiment, after completing the assessment of all items on the PHQ-9 and GAD-7 scales, the system uniformly triggers the aforementioned emotion triggering task.

[0062] Specifically, the positive recall task is used to assess a user's ability to process positive emotions. The task aims to assess the cognitive-behavioral characteristics and positive emotion processing deficits in depressed patients, including "not knowing how to retain happiness" (strategic deficit) and "not feeling bodily pleasure signals" (internal perception deficit). The task process is as follows: Figure 2 As shown.

[0063] After the task begins, the system first plays a pre-set piece of music and presents the user with several cue words to choose from, helping the user enter a suitable emotional state for recall. This stage lasts approximately 15 seconds. Next, the system presents standardized instructions, informing the user to recall positive events that occurred in the past and setting a time window for recall. This stage lasts approximately 30 seconds. After completing the instructions, the user enters a free recall stage, having one minute to independently recall and describe the positive events corresponding to the selected cue words. After free recall, the system enters an open-ended retrospective interview stage, lasting approximately 2 minutes. This stage guides the user to describe the emotion regulation strategies used during the recall process (e.g., whether they tried to maintain a positive mood, whether rumination or avoidance occurred), and changes in bodily sensations (e.g., whether they felt relaxed, changes in heart rate, etc.). Finally, the system enters a structured five-category questionnaire stage, lasting approximately 1.5 minutes, using a standardized questionnaire to further quantify and assess the user's emotion regulation and intrapersonal awareness levels during the positive recall process. The total duration of the entire positive recall task is approximately 5-6 minutes.

[0064] Optionally, in the embodiments, psychological characteristic information includes cognitive behavioral characteristics, emotion regulation strategies, and interoceptive awareness.

[0065] This task assesses individuals at three levels: At the cognitive-behavioral level, it focuses on identifying depression-related characteristics such as overgeneralized autobiographical memories (i.e., voluntary overgeneralization of memories), anhedonia, and deficits in maintaining positive emotions; at the emotion regulation strategy level, it measures maladaptive strategies, adaptive strategies, and taste ability. Maladaptive strategies include rumination, avoidance, repression, and worry; adaptive strategies include acceptance, cognitive reappraisal, and problem-solving; and taste ability encompasses taste capacity and taste behavior / beliefs. At the intrinsic awareness level, referencing the theoretical framework of the eight-dimensional MAIA (Multidimensional Assessment of Interoceptive Awareness) scale, it selects the two most closely related dimensions, noticing and emotional awareness, for measurement, while supplementing the taste / maintenance dimension to more comprehensively assess an individual's ability to perceive bodily signals and maintain emotions during positive emotional experiences.

[0066] The negative recall task is designed to assess a user's cognitive processing bias towards negative information. The task aims to evaluate the cognitive-behavioral characteristics and deficits in negative emotion processing in depressed patients. Figure 3 As shown, after the task begins, the system first plays a pre-set piece of negative emotion-triggered music, lasting about one minute, to help the user enter an emotional state suitable for recalling negative events. After the music ends, the system presents the user with several negative cue words to choose from, lasting about 15 seconds, allowing the user to select the word that best evokes their negative memories. Next, the system presents standardized instructions, informing the user that they need to recall negative events related to the cue word that occurred in the past, and sets a time window for recall, lasting about 30 seconds. After completing the instructions, the user enters a free recall phase, having 1.5 minutes to independently recall and describe the specific details and feelings associated with the selected cue word's negative event. After free recall, the system enters an AI interview interaction phase, lasting about 1.5 to 2 minutes, conducting a point-by-point scoring interview based on 15 pre-set items to capture the user's cognitive processing characteristics during the negative recall process. Afterwards, the system executes an optional adaptive open-ended follow-up questioning phase, lasting about 30 seconds, to delve deeper into aspects of the user's answers that are insufficient or ambiguous, in order to obtain more complete evaluation data. The total duration of the entire negative memory task is approximately 5 to 6 minutes.

[0067] This task assesses depression-related characteristics at three levels: First, at the cognitive-behavioral level, it focuses on identifying rumination, involuntary (intrusive) memory bias (highly specific), mood-consistent memory (biased towards negative memories), and attentional, interpretive, and memory biases towards negative information. Second, at the emotion regulation strategy level, it measures maladaptive strategies (including rumination, avoidance, and repression), adaptive strategies (including acceptance, cognitive reappraisal, and problem-solving), and taste. Third, at the intrinsic awareness level, it uses the MAIA eight-dimensional scale framework to measure five dimensions: awareness, lack of worry, emotional awareness, self-regulation, and trust.

[0068] The future imagination task is designed to assess a user's cognitive processing biases regarding uncertain future situations. Its goal is to evaluate the cognitive-behavioral characteristics of anxiety patients and their deficits in processing negative emotions. For example... Figure 4As shown, after the task begins, the system first presents users with several standardized scenario descriptions to choose from, such as speeches, defenses, and interviews. This phase lasts approximately 15 seconds, during which users select the scenario type that most triggers their anxiety. Next, the system presents standardized instructions, requiring users to imagine themselves in the chosen scenario and to vividly imagine the details, atmosphere, and their own feelings. This phase lasts approximately 30 seconds. After completing the instructions, users enter a free imagination phase, having one minute to independently imagine and describe the specific process of the scenario and their own psychological experience. After the free imagination phase, the system enters an AI interview interaction phase, lasting approximately 2 to 2.5 minutes. This phase involves a point-by-point scoring interview based on 18 preset items to capture the user's anxiety-related cognitive characteristics during the future imagination process. Following this, the system executes an optional adaptive open-ended follow-up questioning phase, lasting approximately 30 seconds, to delve deeper into aspects of the user's answers that are insufficient or ambiguous, in order to obtain more complete evaluation data. The total duration of the entire future imagination task is approximately 4.5 to 5.5 minutes.

[0069] This task assesses anxiety at three levels: at the cognitive-behavioral level, it focuses on identifying anxiety-related characteristics such as rumination, threatening attentional bias, catastrophic interpretation bias, intolerance to uncertainty, and worry maintenance patterns; at the emotion regulation strategy level, it measures maladaptive strategies (including rumination, avoidance, repression, and worry), adaptive strategies (including acceptance, cognitive reappraisal, and problem-solving), and taste ability; at the intrinsic awareness level, it selects three dimensions—non-worry, emotional awareness, and trust—for measurement, referring to the theoretical framework of the MAIA eight-dimensional scale.

[0070] The attention task is designed to assess a user’s attention control ability. It employs a dual-task design with arrow and emotion Flanker, aiming to measure attention control and interference inhibition ability under cold cognition (arrow) conditions and hot cognition (emotion) conditions, respectively. Its cognitive positioning belongs to the attention and inhibition control subconstructs of the cognitive system in the RDoC framework. The total duration is about 5 minutes and it is compatible with computer (keyboard) or tablet (touchscreen) devices.

[0071] like Figure 5As shown, the entire task is executed sequentially, starting with the arrows and then moving to the emotions. At the start of the task, the system first enters the Flanker arrow module, presenting instructions to the user that they need to react to the central target arrow and ignore the distracting arrows on either side; this phase lasts approximately 30 seconds. This is followed by a practice phase, consisting of 8 trials with feedback, helping the user familiarize themselves with the task rules, lasting approximately 30 seconds. After completing the practice, the formal testing phase begins, consisting of 24 trials, lasting approximately 1 minute. The system records the user's reaction time and accuracy under consistent conditions (distracting stimuli point in the same direction as the target stimulus) and inconsistent conditions (distracting stimuli point in opposite directions to the target stimulus). The difference between these two conditions is the Flanker effect, a behavioral indicator measuring attentional control and distraction inhibition abilities.

[0072] After the arrow Flanker module completes, the system displays a transition prompt for approximately 5 seconds. It then proceeds to the emotion Flanker module, which replaces the arrows with emotion faces, including three types: happy, fear / angry, and neutral. In each trial, a 3-stimulus array structure is used, with three emotion faces arranged horizontally. The central face is the target stimulus, and the faces on either side are interference stimuli.

[0073] Based on the relationship between the emotional types of the central face and the two side faces, the trial was divided into three conditions: consistency condition (the central face and the two side faces have the same emotion, e.g., the central face is happy, and the two side faces are also happy), inconsistency condition (the central face and the two side faces have different emotions, e.g., the central face is happy, and the two side faces are fear / anger), and neutral control condition (the two side faces are neutral). Participants were required to judge the emotional category of the central face, that is, to distinguish between happiness and fear / anger, ignoring the interference from the side faces.

[0074] This module also presents instructions sequentially (approximately 30 seconds), a practice phase (8 trials with feedback, approximately 30 seconds), and a formal test (24 trials, including 8 trials for a consistent condition, 8 trials for an inconsistent condition, and 8 trials for a neutral control, approximately 1.5 minutes). In the emotion Flanker module, each trial proceeds as follows: a fixation point is presented for 500 to 1000 milliseconds, followed by a stimulus presentation for up to 2000 milliseconds or until the subject presses a button, with a trial interval of 500 milliseconds; an accuracy rate of 70% or higher is required to pass the practice phase. The total duration of the entire dual Flanker task is approximately 4.5 to 5 minutes.

[0075] By comprehensively scoring the reaction time and accuracy data of the Arrow Flanker and Emotion Flanker tasks, the system identifies and evaluates the following dimensions: the interference inhibition ability of anxiety patients, i.e., assessing their ability to suppress irrelevant information and maintain goal-oriented attention under interference conditions; and the attentional control efficiency of depression patients, i.e., assessing the efficiency of their overall attentional resource allocation and the speed of processing conflicting information. Based on this, further index calculations, NIH comprehensive scores, and emotion-specific index assessments are performed, and the results are interpreted in conjunction with clinical reference standards.

[0076] In some embodiments, psychological characteristic information of users is extracted based on task interaction data. Specifically, this includes: for free recall texts and retrospective interview texts in positive recall tasks, negative recall tasks, and future imagination tasks, psychological characteristic information in three dimensions—cognitive behavioral features, emotion regulation strategies, and interoceptive awareness—is extracted through word segmentation, psycholinguistic dictionary statistics, and semantic analysis using a fine-tuned large language model. Cognitive behavioral features are identified by analyzing the specificity of the recalled content, the density of emotional words, and semantic bias. Emotion regulation strategies are identified by semantic matching and frequency statistics of strategy-related keywords in the retrospective interview text. These keywords include, but are not limited to, rumination, avoidance, repression, acceptance, cognitive reappraisal, and problem-solving. Interoceptive awareness is scored based on semantic analysis of bodily perception-related expressions, referencing the MAIA dimensional framework. These bodily perception-related expressions include, but are not limited to, chest tightness, stomach tension, and rapid heartbeat. For attention tasks, the level of attention control is scored based on the user's reaction time and accuracy in the Flanker task.

[0077] This method organically combines the RDoC research framework with the DSM-5 clinical diagnostic criteria. RDoC is an innovative research framework proposed by the National Institute of Mental Health in the United States. It aims to advance the mechanistic research and precision treatment of mental disorders by integrating multidimensional data (such as genetics, neural circuits, and behavior) beyond traditional diagnostic classifications. The RDoC framework is constructed from six domains: Negative Valence Systems, Positive Valence Systems, Cognitive Systems, Systems for Social Processes, Arousal / Regulatory Systems, and Sensorimotor Systems.

[0078] Positive valence systems refer to the neurobehavioral systems within the RDoC framework that are primarily responsible for responding to positively motivating situations or rewarding stimuli, such as reward seeking, consumption behavior, and reward / habit learning. Negative valence systems refer to the neurobehavioral systems that are primarily responsible for responding to aversive situations or threatening stimuli, such as fear, anxiety, and loss. Cognitive systems refer to the functional areas within the RDoC framework responsible for cognitive processes such as attention, perception, memory, language, and cognitive control.

[0079] This embodiment systematically categorizes the 9 items of PHQ-9 and the 7 items of GAD-7 according to the RDoC framework, as shown in Table 2: Table 2 Item Classification Table The above classification references the research findings of Gunzler et al., who mapped PHQ-9 and GAD-7 to the RDoC framework.

[0080] Based on the above RDoC dimensions, specific emotional triggering tasks were designed. Specifically, the positive valence system corresponds to the positive recall task, the negative valence system corresponds to the negative recall task and the future imagination task, and the cognitive system corresponds to the attention task. After completing the PHQ-9 and GAD-7 scales, the triggering tasks and responses began.

[0081] The scale items included in the arousal / regulation system and sensorimotor system are relatively specific and clear, and users can usually answer them truthfully. Therefore, no emotional triggering tasks are set, and the assessment is completed only by asking questions through the scale items.

[0082] The core output of this method in the human-computer interaction process is to generate a structured, multi-dimensional mental health profile for each user, specifically including the following quantitative indicators: Anxiety risk level assessment: e.g., normal, subclinical, high risk; Depression risk level assessment: such as normal, subclinical, high risk; Cognitive and behavioral characteristics: such as overgeneralization of autobiographical memory, rumination, attentional bias, etc. Emotion regulation strategies: Distribution of maladaptive and adaptive strategies; Intrasensory awareness: Scores and explanations for each dimension.

[0083] The final output is a comprehensive report, including risk level assessment, scores and explanations for each dimension, and personalized intervention recommendations. The mental health profile also includes evidence chain information, which records the source interaction data for the target scale items, the scoring criteria, and the sources from which psychological characteristic information was extracted. This ensures that every assessment result has source evidence and a status record, making it interpretable, traceable, and verifiable.

[0084] Example 2 like Figure 6 As shown, this embodiment provides a human-computer interaction assessment system 200 for mental health, including: Data acquisition module 201 is used to acquire the interaction data generated by the user's answers to the items of the mental health scale one by one; The data parsing module 202 is used to perform natural language understanding processing on each piece of interactive data to obtain the natural language parsing result corresponding to the interactive data. The scale assessment module 203 is used to match the target scale item corresponding to each interaction data according to the natural language parsing result corresponding to each interaction data, and determine the score corresponding to each target scale item in order to obtain the user's scale assessment result. The task interaction module 204 is used to output emotional triggering tasks to users after they have completed the psychological health scale item assessment, and to obtain the task interaction data generated by users during the execution of the emotional triggering tasks. Information extraction module 205 is used to extract users' psychological characteristic information based on task interaction data; The results generation module 206 is used to generate the user's mental health assessment results based on the scale assessment results and psychological characteristic information.

[0085] Optionally, in an embodiment, the data parsing module 202 includes: The data parsing unit is used to input interactive data and retrieval enhancement context into the natural language understanding model, and to output the natural language parsing results corresponding to the interactive data according to the preset data structure fields by constraining the natural language understanding model through prompt words; wherein, the retrieval enhancement context includes at least one of the following: tag knowledge base, symptom definition, scale item description and similar historical cases.

[0086] Optionally, in this embodiment, the natural language understanding model is trained in the following manner: A training corpus is acquired, comprising at least one of the following: psychological interview corpus, PHQ-9 questionnaire corpus, GAD-7 questionnaire corpus, open-ended response corpus based on emotion-triggered tasks, cognitive behavioral feature questionnaire corpus, emotion regulation strategy questionnaire corpus, and interoceptive awareness questionnaire corpus; each piece of corpus in the training corpus is manually annotated to obtain structured annotation data corresponding to each piece of corpus; the pre-defined large language model is adjusted and trained using the corpus in the training corpus as input and the structured annotation data corresponding to the corpus as the target output to obtain the natural language understanding model.

[0087] Optionally, in an embodiment, the scale assessment module 203 includes: The matching unit is used to match the target scale item corresponding to each piece of interaction data with the symptom information in the natural language parsing result corresponding to the interaction data through a pre-trained emotion symptom mapping model. The scoring unit is used to determine the score of the target scale item corresponding to each piece of interactive data based on the frequency information in the natural language parsing results of the interactive data and the preset frequency score mapping relationship. The writing unit is used to maintain an independent state for each scale item and control the writing of the score of the target scale item corresponding to each interactive data according to the current state of each scale item. The rating determination unit is used to accumulate the scores of each scale item after the scores of each scale item have been written to obtain the total score, and generate the screening rating based on the total score. The assessment results generation unit is used to obtain the user's scale assessment results based on the scores of each scale item, the total score, and the screening level.

[0088] Optionally, in an embodiment, the matching unit includes: The input subunit is used to input the symptom information in the natural language parsing results corresponding to the interaction data into the pre-trained emotion symptom mapping model. The pre-trained emotion symptom mapping model includes an encoding layer, a matching layer, a classification layer, and a rule correction layer. The encoding subunit is used to encode symptom information from the natural language parsing results corresponding to the interactive data into symptom vectors through the encoding layer; The symptom candidate subunit is used to calculate the similarity between the symptom vector and the preset standard symptom label vector and the historical labeled case vector through the matching layer, and to determine the candidate symptom set based on the obtained similarity calculation results. The probability calculation subunit is used to input the symptom vector, the current interview context vector, and the vectors corresponding to each candidate symptom in the candidate symptom set into the multi-label classification network through the classification layer, and generate the probability value corresponding to each candidate symptom in the candidate symptom set. The target determination subunit is used to determine the matching symptom label based on the probability value corresponding to each candidate symptom through the rule correction layer, and to determine the target scale item corresponding to the symptom label based on the preset symptom scale item mapping relationship.

[0089] Optionally, in the embodiments, the emotion-triggered task includes at least one of a positive recall task, a negative recall task, a future imagination task, and an attention task; the positive recall task is a task used to assess the user's ability to process positive emotions; the negative recall task is a task used to assess the user's cognitive processing bias towards negative information; the future imagination task is a task used to assess the user's cognitive processing bias towards uncertain future situations; and the attention task is a task used to assess the user's attention control ability.

[0090] Optionally, in the embodiments, psychological characteristic information includes cognitive behavioral characteristics, emotion regulation strategies, and interoceptive awareness.

[0091] In some embodiments, the mental health human-computer interaction assessment system 200 of the present invention can be implemented in a combination of hardware and software. As an example, the mental health human-computer interaction assessment system 200 of the present invention can be a processor in the form of a hardware decoding processor, which is programmed to execute the mental health human-computer interaction assessment method of the present invention. For example, the processor in the form of a hardware decoding processor can be one or more application-specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field-programmable gate arrays (FPGAs), or other electronic components.

[0092] The modules described in the embodiments of this invention can be implemented in software or hardware. The names of the modules are not, in some cases, limiting the scope of the module itself.

[0093] Example 3 like Figure 7 As shown, this embodiment provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, it implements the human-computer interaction assessment method for mental health as described in Embodiment 1.

[0094] In other words, an electronic device according to an embodiment of the present invention may include, but is not limited to: a processor and a memory; the memory is used to store computer programs; the processor is used to execute the human-computer interaction assessment method for mental health shown in any embodiment of the present invention by calling the computer programs.

[0095] In one alternative embodiment, an electronic device is provided. Figure 7The illustrated electronic device 300 includes a processor 301 and a memory 303. The processor 301 and the memory 303 are connected, for example, via a bus 302. Optionally, the electronic device 300 may further include a transceiver 304, which can be used for data interaction between the electronic device and other electronic devices, such as sending and / or receiving data. It should be noted that in practical applications, the transceiver 304 is not limited to one type, and the structure of the electronic device 300 does not constitute a limitation on the embodiments of the present invention.

[0096] Processor 301 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this invention. Processor 301 may also be a combination that implements computational functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0097] Bus 302 may include a path for transmitting information between the aforementioned components. Bus 302 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 302 can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 7 The bus 302 is represented by only one thick line, but this does not mean that there is only one bus or one type of bus.

[0098] The memory 303 may be a ROM (Read Only Memory) or other type of static storage device capable of storing static information and instructions, RAM (Random Access Memory) or other type of dynamic storage device capable of storing information and instructions, or an EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited thereto.

[0099] The memory 303 is used to store application code (computer program) for executing the present invention, and its execution is controlled by the processor 301. The processor 301 is used to execute the application code stored in the memory 303 to implement the content shown in the foregoing method embodiments.

[0100] Among them, electronic devices can also be terminal devices, which can be any device that can install applications, including at least one of smartphones, tablets, laptops, desktop computers, smart speakers, smartwatches, smart TVs, and smart in-vehicle devices.

[0101] It should be noted that, Figure 7 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of use of the embodiments of the present invention.

[0102] Example 4 This embodiment provides a non-transitory computer-readable storage medium that stores computer instructions for causing a computer to execute a human-computer interaction assessment method for mental health as described in Embodiment 1.

[0103] Alternatively, the computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a compact disc read-only memory (CD-ROM), magnetic tape, a floppy disk, and an optical data storage device, etc.

[0104] In an exemplary embodiment, a computer program product or computer program is also provided, which includes computer instructions stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the aforementioned human-computer interaction assessment method for mental health.

[0105] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof. These programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—and conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer. In cases involving a remote computer, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0106] It should be understood that the flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of methods and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0107] The computer-readable storage medium provided in this invention can be, but is not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EEPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0108] The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to perform the method shown in the above embodiments.

[0109] The above description is merely a preferred embodiment of the present invention and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this invention is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-disclosed concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this invention.

[0110] It should be noted that the terms "first," "second," etc., used in the specification and claims of this application are used to distinguish similar objects and represent a limitation on a specific order or sequence. Where appropriate, the order of use for similar objects can be interchanged so that the embodiments of this application described herein can be implemented in an order other than that shown or described.

[0111] Those skilled in the art will recognize that this invention can be implemented as a system, method, or computer program product. Therefore, this invention can be specifically implemented in the following forms: it can be entirely hardware, entirely software (including firmware, resident software, microcode, etc.), or a combination of hardware and software, generally referred to herein as a "circuit," "module," or "system." Furthermore, in some embodiments, this invention can also be implemented as a computer program product contained in one or more computer-readable media, which includes computer-readable program code.

[0112] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A human-computer interaction assessment method for mental health, characterized in that, include: Collect interaction data generated by users' responses to items on a mental health scale one by one; For each piece of interaction data, natural language understanding processing is performed on the interaction data to obtain the natural language parsing result corresponding to the interaction data; The target scale item corresponding to each interaction data is matched according to the natural language parsing result of each interaction data, and the score corresponding to each target scale item is determined to obtain the scale evaluation result of the user. After completing the psychological health scale items assessment, an emotional trigger task is output to the user, and task interaction data generated by the user in the process of performing the emotional trigger task is obtained; Extract the user's psychological characteristic information based on the task interaction data; Based on the scale assessment results and the psychological characteristic information, a mental health assessment result for the user is generated.

2. The human-computer interaction assessment method for mental health according to claim 1, characterized in that, For each piece of interaction data, natural language understanding processing is performed on the interaction data to obtain the corresponding natural language parsing result, including: The interaction data and the enhanced retrieval context are input into the natural language understanding model. The natural language understanding model is constrained by prompt words to output the natural language parsing result corresponding to the interaction data according to the preset data structure fields. The enhanced retrieval context includes at least one of the following: a tag knowledge base, symptom definitions, scale entry descriptions, and similar historical cases.

3. The human-computer interaction assessment method for mental health according to claim 2, characterized in that, The natural language understanding model was trained in the following way: Obtain a training corpus, which includes at least one of the following: psychological interview corpus, PHQ-9 scale question and answer corpus, GAD-7 scale question and answer corpus, open-ended response corpus based on emotion triggering task, cognitive behavioral feature scale question and answer corpus, emotion regulation strategy scale question and answer corpus, and interoceptive awareness scale question and answer corpus. Each corpus in the training corpus is manually annotated to obtain structured annotation data corresponding to each corpus; Using the corpus in the training corpus as input and the structured labeled data corresponding to the corpus as the target output, the preset large language model is adjusted and trained to obtain the natural language understanding model.

4. The human-computer interaction assessment method for mental health according to claim 1, characterized in that, The process of matching the target scale item corresponding to each interaction data point with the natural language parsing result and determining the score corresponding to each target scale item to obtain the user's scale evaluation result includes: For each piece of interaction data, based on the symptom information in the natural language parsing result corresponding to the interaction data, the target scale item corresponding to the interaction data is matched through a pre-trained emotion symptom mapping model. For each piece of interactive data, the score of the target scale item corresponding to the interactive data is determined based on the frequency information in the natural language parsing result corresponding to the interactive data and the preset frequency score mapping relationship. Maintain an independent state for each scale item, and control the writing of the score of the target scale item corresponding to each interactive data according to the current state of each scale item; After the scores for each scale item are entered, the scores for each scale item are summed to obtain the total score, and the screening level is generated based on the total score. The user's scale assessment result is obtained based on the scores of each scale item, the total score, and the screening level.

5. The human-computer interaction assessment method for mental health according to claim 4, characterized in that, For each piece of interaction data, based on the symptom information in the natural language parsing result corresponding to the interaction data, a pre-trained emotion symptom mapping model is used to match the target scale item corresponding to the interaction data, including: The symptom information in the natural language parsing results corresponding to the interaction data is input into a pre-trained emotion symptom mapping model, which includes an encoding layer, a matching layer, a classification layer, and a rule correction layer. The encoding layer encodes the symptom information in the natural language parsing results corresponding to the interactive data into a symptom vector. The matching layer calculates the similarity between the symptom vector and the preset standard symptom label vector and historical labeled case vector, respectively, and determines the candidate symptom set based on the obtained similarity calculation results. The classification layer inputs the symptom vector, the current interview context vector, and the vectors corresponding to each candidate symptom in the candidate symptom set into a multi-label classification network to generate probability values ​​corresponding to each candidate symptom in the candidate symptom set. Through the rule correction layer, matching symptom labels are determined based on the probability values ​​corresponding to each candidate symptom, and target scale entries corresponding to the symptom labels are determined based on the preset symptom scale entry mapping relationship.

6. The human-computer interaction assessment method for mental health according to claim 1, characterized in that, The emotional triggering task includes at least one of the following: positive recall task, negative recall task, future imagination task, and attention task; the positive recall task is used to assess a user's ability to process positive emotions; the negative recall task is used to assess a user's cognitive processing bias towards negative information; the future imagination task is used to assess a user's cognitive processing bias towards uncertain future situations; and the attention task is used to assess a user's ability to control attention.

7. The human-computer interaction assessment method for mental health according to claim 1, characterized in that, The psychological characteristics information includes cognitive-behavioral characteristics, emotion regulation strategies, and interoceptive awareness.

8. A human-computer interactive assessment system for mental health, characterized in that, include: The data acquisition module is used to acquire the interaction data generated by users answering questions on the mental health scale item by item; The data parsing module is used to perform natural language understanding processing on each piece of interactive data to obtain the natural language parsing result corresponding to the interactive data. The scale assessment module is used to match the target scale item corresponding to each interaction data according to the natural language parsing result corresponding to each interaction data, and determine the score corresponding to each target scale item, so as to obtain the scale assessment result of the user. The task interaction module is used to output emotional triggering tasks to users after they have completed the psychological health scale items assessment, and to obtain the task interaction data generated by the users during the execution of the emotional triggering tasks. The information extraction module is used to extract the user's psychological characteristic information based on the task interaction data; The results generation module is used to generate the user's mental health assessment results based on the scale assessment results and the psychological characteristic information.

9. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the human-computer interaction assessment method for mental health as described in any one of claims 1 to 7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to execute the human-computer interaction assessment method for mental health as described in any one of claims 1 to 7.

Images