Semantic analysis-based teenager psychological problem AI dynamic evaluation method

Abstract

The invention discloses a semantic analysis-based teenager psychological problem AI dynamic assessment method, and relates to the technical field of artificial intelligence psychological assessment. The method comprises the following steps: acquiring semantic data such as dialogue text content, emotion keyword information and semantic coherence information of a patient in different scenes, dividing a stable and change evaluation stage according to the change trend of the number of emotion keywords along with dialogue rounds, and screening dialogue scenes of key semantic types; in a key scene, a generative driving evaluation index is obtained by combining semantic data similarity conditions, round information and semantic coherence information, and a dynamic evaluation node possibility is obtained by combining a dialogue text in a change evaluation stage, a round corresponding to the semantic coherence information and repeated expression data; screening evaluation nodes, then constructing an evaluation chain tree structure, and determining psychological state dynamic evaluation schemes of different patients. The method can dynamically capture the psychological state change of the patient.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence psychological assessment technology, specifically to an AI dynamic assessment method for adolescent psychological problems based on semantic analysis. Background Technology

[0002] With the deepening application of artificial intelligence technology in the healthcare field, psychological state assessment is gradually evolving from traditional manual interviews towards intelligent and automated methods. Traditional psychological assessments primarily rely on professional physicians conducting face-to-face interactions and completing questionnaires. This approach is not only heavily influenced by the physician's subjective experience but also time-consuming, making it difficult to meet the rapid assessment needs of large populations. Furthermore, psychological states are inherently dynamic; the content of conversations and emotional expressions in different scenarios change over time and during communication. Traditional assessment methods struggle to capture these dynamic changes, limiting the timeliness and accuracy of the assessment results.

[0003] Currently, most existing semantic analysis-based psychological assessment techniques employ static analysis methods, relying on extracting keywords and emotional features from dialogue texts for single-dimensional assessment. This neglects the diversity of dialogue scenarios and the dynamic nature of the assessment process. For example, some methods only analyze dialogue content within specific scenarios, failing to differentiate the impact of different scenarios on psychological state expression, resulting in assessment results lacking universality. Other methods, while considering dialogue turn information, fail to effectively integrate the changing trends of emotional keywords and semantic coherence information, making it difficult to accurately identify key nodes in the assessment process and resulting in insufficient targeting of the assessment plan.

[0004] Furthermore, existing technologies often fail to effectively distinguish between stable and variable assessment phases when processing semantic data, and they also fail to filter out dialogue scenarios with key semantic types, resulting in a large amount of redundant data interfering with the assessment results. Simultaneously, when constructing assessment models, fixed assessment indicators are frequently used, lacking mechanisms for dynamic adjustment based on individual patient differences and the dialogue process, making it difficult to adapt to the complex changes in the psychological states of different patients. For example, when patients exhibit fluctuating emotions or incoherent semantic expression during dialogue, existing methods struggle to accurately capture these dynamic characteristics, easily leading to assessment bias.

[0005] With increasing societal pressures, the number of people requiring mental health assessments is constantly growing, leading to higher demands for assessment efficiency and accuracy. The limitations of traditional assessment methods and existing intelligent assessment technologies make it crucial to address how to achieve dynamic and accurate assessments of the mental states of different patients in diverse scenarios. Especially in complex dialogue scenarios, effectively extracting key semantic information, dividing assessment stages, identifying dynamic assessment nodes, and constructing scientifically sound assessment models are significant challenges for improving the level of mental health assessment. Summary of the Invention

[0006] The purpose of this invention is to provide an AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis, so as to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, this invention provides an AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis, the method comprising:

[0008] Semantic data of each patient in different dimensions is obtained in different dialogue scenarios; the semantic data includes dialogue text content, emotional keyword information, and semantic coherence information.

[0009] Based on the trend of the number of emotional keywords changing with the number of dialogue rounds in each dialogue scenario, we divide the dialogue into a stable evaluation phase and a change evaluation phase; based on the difference in the number of emotional keywords in the stable evaluation phase and the change evaluation phase in each dialogue scenario, we select dialogue scenarios with key semantic types.

[0010] In each key semantic type dialogue scenario, based on the similarity of the semantic data of each patient in the stable assessment phase and each patient in the change assessment phase, combined with round information and the patient's semantic coherence information, the generative driving assessment index of each patient in each key semantic type in the stable assessment phase is obtained.

[0011] Based on the dialogue text content and semantic coherence information of each patient during the change assessment phase, as well as the repetitive expression data, and combined with the generative-driven assessment indicators, the probability of dynamic assessment nodes for each patient under each key semantic type during the stable assessment phase is obtained.

[0012] Based on the probability of the dynamic assessment nodes, each patient under each key semantic type in the stable assessment phase is selected as an assessment node. Combining the semantic data of each patient under the same key semantic type in the change assessment phase, an assessment chain tree structure is constructed. Based on the assessment chain tree structure, a dynamic assessment scheme for the psychological state of different patients is determined.

[0013] Optionally, the dialogue text content includes the time of occurrence of each dialogue action; the semantic coherence information includes the semantic duration of each dialogue action and the time interval from the semantic interruption to the time of occurrence of each dialogue action, referred to as the semantic decision duration of each dialogue action; the semantic data also includes the number of repetitions.

[0014] Optionally, the method for obtaining the generative-driven assessment indicators for each patient under each key semantic type during the stable assessment phase specifically includes:

[0015] In any dialogue scenario with a key semantic type, patients with the semantic data present in the stable assessment phase are all designated as patients to be screened, and patients with the semantic data present in the changing assessment phase are all designated as patients to be assessed; any patient to be screened is recorded as a selected patient to be screened, and any patient to be assessed is recorded as a selected patient to be assessed.

[0016] Based on the timing of each dialogue, semantic duration, and similarity index between the selected patients and the selected patients being evaluated, the degree of dialogue influence of each dialogue behavior of the selected patients relative to the selected patients being evaluated is obtained.

[0017] The semantic similarity between the semantic data of each dialogue behavior of the selected patients to be screened during the stable assessment phase and the semantic data of each dialogue behavior of the selected patients to be assessed during the variable assessment phase is calculated to obtain the semantic similarity factor of the selected patients to be assessed for each dialogue behavior.

[0018] Using the degree of influence of the dialogue as a weight, the semantic similarity factor corresponding to the selected patient under each dialogue behavior is weighted and averaged to obtain the psychological assessment power of the selected patient to be screened relative to the selected patient under evaluation. The mean of the psychological assessment power of the selected patient to be screened relative to all patients under evaluation is used as the generative driving assessment index of the selected patient to be screened.

[0019] Optionally, the method for obtaining the degree of dialogue influence of each dialogue behavior of the selected patient to be screened relative to the selected patient to be evaluated specifically includes:

[0020] Record any dialogue action in any order as the target dialogue action;

[0021] Based on the semantic similarity between the semantic data of the selected patients to be screened in each dimension and the semantic data of the selected patients to be evaluated in each dimension, the degree of semantic association between the selected patients to be screened and the selected patients to be evaluated is determined.

[0022] The ratio between the time of dialogue occurrence of the selected patients being evaluated and the selected patients to be screened under the target dialogue behavior is used as the first correlation coefficient; the ratio between the degree of semantic correlation and the semantic duration of the selected patients being evaluated under the target dialogue behavior is used as the second correlation coefficient.

[0023] The product of the first correlation coefficient and the second correlation coefficient is calculated to obtain the degree of dialogue influence of the selected patients to be screened relative to the selected patients to be evaluated under the target dialogue behavior.

[0024] Optionally, the step of obtaining the probability of dynamic assessment nodes for each patient under each key semantic type in the stable assessment phase, based on the dialogue text content and semantic coherence information corresponding to each patient during the change assessment phase, as well as the repetitive expression data, combined with the generative-driven assessment indicators, specifically includes:

[0025] The dialogue decision-making ability index of the selected patients to be screened was obtained by analyzing the semantic duration and semantic decision-making duration of each dialogue behavior and the number of repeated expressions.

[0026] The normalized value of the product of the generative-driven assessment index and the dialogue decision-making index of the selected patients to be screened is used as the probability of the dynamic assessment node of the selected patients to be screened.

[0027] Optionally, the step of obtaining the dialogue decision-making ability index of the selected patient based on the semantic duration and semantic decision-making duration of each dialogue behavior and the number of repetitions of the selected patient to be screened specifically includes:

[0028] The characteristic ratio of semantic duration to semantic decision duration for each dialogue behavior of the selected patients to be screened was calculated. The product of the mean of the characteristic ratios corresponding to all dialogue behaviors of the selected patients to be screened during the stable assessment phase and the number of repeated expressions of the selected patients to be screened during the stable assessment phase was used as the dialogue decision-making ability index of the selected patients to be screened.

[0029] Optionally, the step of selecting each patient under each key semantic type in the stable assessment phase as an assessment node based on the probability of the dynamic assessment node, and constructing an assessment chain tree structure by combining the semantic data of each patient under the same key semantic type in the change assessment phase, specifically includes:

[0030] Patients whose probability of being a dynamic assessment node is greater than a preset assessment threshold are respectively used as assessment nodes in each tree of the assessment chain tree structure.

[0031] Obtain the dialogue decision-making ability index of each evaluated patient, and multiply the degree of dialogue influence between each evaluated patient and the screening patients corresponding to the evaluation node with the dialogue decision-making ability index of the evaluated patient as the compliance index of each evaluated patient relative to the evaluation node.

[0032] For any assessment node, a tree structure is constructed according to the compliance index of each assessed patient relative to that assessment node in descending order. The compliance index of the assessed patients corresponding to nodes at the same level in the tree structure is the same.

[0033] The tree structure of all evaluation nodes constitutes the evaluation chain tree structure.

[0034] Optionally, the step of determining a dynamic assessment scheme for the psychological state of different patients based on the assessment chain tree structure specifically includes:

[0035] In evaluating the chain tree structure, the layer containing the first common child node of different evaluation nodes is taken as the first-stage target layer; after the first-stage target layer in the evaluation chain tree structure, the layer containing the largest number of child nodes is taken as the second-stage target layer.

[0036] For patients between the assessment node and the first-stage target layer, a first-type preset assessment strategy is used; for patients between the first-stage target layer and the second-stage target layer, a second-type preset assessment strategy is used; and for patients between the second-stage target layer and the bottom layer, a third-type preset assessment strategy is used.

[0037] Optionally, the step of dividing the dialogue scenario into a stable evaluation phase and a change evaluation phase based on the changing trend of the number of emotional keywords with each dialogue round specifically includes:

[0038] For any dialogue scenario, obtain the keyword quantity sequence consisting of the number of emotional keywords in each round, calculate the first difference value of the keyword quantity sequence, and select the rounds with positive first difference values ​​as the growth rounds.

[0039] Arrange the growth cycles in ascending order of their corresponding first-order difference values ​​to obtain a difference sequence. Calculate the second-order difference value of the difference sequence. Take the growth cycle corresponding to the maximum value of the second-order difference value as the characteristic cycle. The stage before the characteristic cycle is the stability evaluation stage, and the stage after the characteristic cycle is the change evaluation stage.

[0040] Optionally, the step of selecting dialogue scenarios with key semantic types based on the difference in the number of emotion keywords in the stable evaluation phase and the change evaluation phase in each dialogue scenario specifically includes:

[0041] For any dialogue scenario, obtain the first mean of the number of emotional keywords during the stable evaluation phase and the second mean of the number of emotional keywords during the change evaluation phase; normalize the difference between the second mean and the first mean to obtain the semantic difference coefficient; if the semantic difference coefficient is greater than the preset difference threshold, then the dialogue scenario is regarded as a dialogue scenario with key semantic type.

[0042] Compared with the prior art, the technical solution of the present invention has at least the following beneficial effects:

[0043] In this embodiment of the invention, numerous problems existing in current psychological assessment techniques are effectively addressed through multi-dimensional semantic data processing and a phased assessment strategy. Firstly, this method enables refined processing of dialogue content across different scenarios. By dividing the assessment into stable and variable phases and combining the changing trends in the number of emotional keywords, it filters dialogue scenarios with key semantic types, avoiding the drawbacks of traditional methods that apply a blanket approach to all scenarios, thus making the assessment process more targeted. This precise selection of scenarios eliminates interference from irrelevant scenes, allowing assessment resources to focus on dialogue content that has a more significant impact on psychological state, thereby better reflecting the patient's true psychological expression characteristics.

[0044] In dialogue scenarios involving key semantic types, this method combines multiple factors, including the patient's semantic coherence, turn information, and repetitive expression data, to construct generative-driven assessment indicators and dynamic assessment node probabilities, enabling dynamic tracking of the assessment process. Compared to static assessment methods, this dynamic assessment mechanism can capture the psychological changes of patients at different stages of the dialogue. For example, based on the assessment formed in the stable assessment stage, adjustments can be made to the semantic data in the changing assessment stage to promptly detect fluctuations and shifts in the patient's psychological state, making the assessment results more reflective of real-time changes in psychological state.

[0045] By constructing an assessment chain tree structure, this method can integrate and correlate semantic data from different patients in key semantic types of scenarios, forming a systematic assessment plan. This structure not only reflects the psychological characteristics of individual patients at different assessment stages but also reveals the commonalities and individual differences in the psychological states of a group by comparing and analyzing the assessment chains of different patients. For example, in dialogue scenarios with the same key semantic types, different patients exhibit differences in the distribution of emotional keywords and the degree of semantic coherence. The assessment chain tree structure can clearly present these differences, providing a more detailed basis for personalized psychological assessment.

[0046] This method, when processing semantic data, fully considers the repetitive expressions and semantic coherence of patients, effectively identifying implicit characteristics such as psychological conflict and emotional repression that may exist in the patient's dialogue. Traditional assessment methods often overlook these details, while this method, through in-depth data mining, can more comprehensively understand the patient's psychological state and reduce assessment bias caused by information omissions. Furthermore, the introduction of generative assessment indicators means that assessments no longer rely on fixed templates or scales, but rather dynamically generate assessment criteria based on the patient's actual semantic expression, making it more adaptable to the language habits and expression styles of different patients.

[0047] This method achieves dynamic adjustment of the assessment scheme by selecting assessment nodes and combining semantic data from the changing assessment phase. During the assessment process, as the dialogue deepens and new semantic data is generated, the assessment chain tree structure is continuously updated, ensuring that the assessment scheme remains synchronized with changes in the patient's psychological state. This dynamic adjustment mechanism effectively addresses the uncertainties and complexities commonly encountered in psychological state assessment, making the assessment results more timely and accurate, and providing a more scientific and flexible technical approach for psychological state assessment. Attached Figure Description

[0048] Figure 1 A schematic diagram illustrating the working principle of the AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis, provided in an embodiment of the present invention.

[0049] Figure 2 A flowchart of a generative-driven evaluation metric acquisition method provided in an embodiment of the present invention;

[0050] Figure 3 A flowchart for calculating generative-driven evaluation metrics provided in an embodiment of the present invention;

[0051] Figure 4 A flowchart for calculating the degree of influence of a dialogue provided in an embodiment of the present invention. Detailed Implementation

[0052] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0053] Please see Figures 1-4 This invention provides an AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis. The method includes the following steps:

[0054] Step 1: Obtain semantic data for each patient in different dimensions under different dialogue scenarios; the semantic data includes dialogue text content, emotional keyword information, and semantic coherence information.

[0055] In practical applications, dialogue acquisition devices record patients' conversations in various preset scenarios (such as daily communication scenarios, stress test scenarios, and recall scenarios), from which semantic data is extracted. The dialogue text content consists of the patient's actual language expressions in each scenario; emotional keyword information consists of emotion-related words identified from the dialogue text, such as "happy," "anxious," and "angry"; semantic coherence information is used to reflect the continuity and coherence of semantics in the dialogue.

[0056] Step 2: Based on the trend of the number of emotional keywords changing with the number of dialogue rounds in each dialogue scenario, divide the dialogue into a stable evaluation phase and a change evaluation phase; based on the difference in the number of emotional keywords in the stable evaluation phase and the change evaluation phase in each dialogue scenario, select dialogue scenarios with key semantic types.

[0057] For each dialogue scenario, the number of emotional keywords appearing in each round of dialogue was counted, forming a sequence that changed with each round. By analyzing the changing patterns of this sequence, the dialogue process was divided into a stable assessment phase and a change assessment phase. Then, the differences in the number of emotional keywords between the two phases were compared, and scenarios with significant differences were selected as dialogue scenarios of key semantic types. These scenarios can more effectively reflect changes in the patient's psychological state.

[0058] Step 3: In each key semantic type dialogue scenario, based on the similarity of the semantic data of each patient in the stable assessment phase and the changing assessment phase, combined with round information and the patient's semantic coherence information, the generative-driven assessment index of each patient in each key semantic type in the stable assessment phase is obtained.

[0059] For the selected key semantic type dialogue scenarios, the semantic data of patients in the stable assessment stage and the changing assessment stage were analyzed separately. By calculating the similarity of the semantic data of patients in different stages, and combining the dialogue turn information and semantic coherence information, a generative-driven assessment index was constructed. This index is used to measure the driving ability of patients in the stable assessment stage to assess the psychological state of patients in the changing assessment stage.

[0060] Step 4: Based on the dialogue text content and semantic coherence information corresponding to each patient in the change assessment phase, as well as the repetitive expression data, and combined with the generative-driven assessment indicators, obtain the probability of dynamic assessment nodes for each patient under each key semantic type in the stable assessment phase.

[0061] Collect the dialogue text content, semantic coherence information corresponding to the rounds, and repetitive expression data of patients in the change assessment phase. Combine this data with the generative driven assessment index obtained in step 3, and obtain the dynamic assessment node probability through a specific calculation method. This probability reflects the suitability of patients in the stable assessment phase as assessment nodes.

[0062] Step 5: Based on the probability of the dynamic assessment node, select each patient under each key semantic type in the stable assessment phase as an assessment node, combine the semantic data of each patient under the same key semantic type in the change assessment phase, construct an assessment chain tree structure, and determine the dynamic assessment scheme for the psychological state of different patients based on the assessment chain tree structure.

[0063] Based on the probability of dynamic assessment nodes, suitable assessment nodes are selected, and then, combined with the semantic data of patients in the changing assessment stage, an assessment chain tree structure is constructed. By analyzing the hierarchical relationships and correlation characteristics of each node in this structure, dynamic assessment schemes for the psychological state of different patients are formulated.

[0064] Example 1:

[0065] The dialogue text content covers the exact moment each dialogue action occurs. The moment of dialogue occurrence refers to the specific time when the patient utters a verbal expression during the dialogue, automatically recorded by the dialogue acquisition system. The recording precision can be adjusted according to actual needs, such as to the second or millisecond level. In multi-turn dialogues, the start and end times of each turn are also recorded synchronously to clearly present the temporal sequence of the dialogue. This time information reflects the rhythm of the dialogue, such as the time interval between consecutive dialogues and the duration of a single dialogue. The distribution of dialogue occurrence times varies among different patients in different scenarios; some patients may initiate multiple dialogues in a short period, while others may respond only after a longer interval.

[0066] Semantic coherence information includes the semantic duration of each conversational action and the semantic decision-making duration within each conversational action. Semantic duration refers to the length of time during which the patient's expressed semantics remain coherent within a single conversational action. For example, if a patient continuously expresses their feelings and experiences regarding the topic of "work stress," the time from the start of discussing this topic until the topic shifts or the conversation is interrupted is the semantic duration of this conversational action. If the patient pauses briefly during their expression but does not change the topic, the pause time is still included in the semantic duration; if the topic is changed midway, the semantic duration ends at the moment of topic change.

[0067] Semantic decision duration refers to the time interval between the last semantic interruption of a conversation and the occurrence of the current conversation. The last semantic interruption may be due to the patient ceasing to express themselves, the topic being interrupted, or the conversation ending naturally. For example, if a patient stops expressing themselves after discussing "family relationships" and then resumes discussing "health status" after a period of time, the time between stopping discussing "family relationships" and starting to discuss "health status" is the semantic decision duration of the current conversation about "health status." This duration reflects the time it takes for the patient to reorganize their language and initiate a new conversation after a break. Patients in different psychological states may exhibit different characteristics in semantic decision duration; some patients can quickly resume the conversation, while others need a longer period of thought before continuing to express themselves.

[0068] Semantic data also includes the number of repetitions. The number of repetitions refers to the number of times a patient repeats the same or highly similar content during a conversation. The same or highly similar content includes both completely identical statements and content with the same core meaning but slightly different expressions. For example, a patient repeatedly saying "I feel very anxious," or first mentioning "I haven't been able to sleep lately," and then saying "I've been having insomnia for the past few nights," are both examples of repetitions. The scope of repetitions can be within a single conversational scenario or across multiple related scenarios; the specific statistical scope is determined based on the assessment needs. When counting the number of repetitions, normal emphatic repetitions need to be excluded, and only repetitive expressions without obvious emphatic intent should be counted. This can be done by analyzing the conversation text using text comparison tools to identify and count the repeated content segments.

[0069] During the actual data collection process, the dialogue text is converted into text format using audio-to-text technology, while simultaneously linking it to the corresponding dialogue timestamps. Semantic duration and semantic decision duration are determined by analyzing the timestamps and semantic coherence of the dialogue text. Natural language processing techniques are used to identify the beginning and end of semantic topics, thereby calculating the corresponding durations. The number of repeated expressions is determined by comparing each sentence of the converted dialogue text using a text similarity algorithm, marking repeated content and accumulating the number of repetitions. This data is stored in a database, forming a structured semantic dataset. Each patient's semantic data for each dialogue scenario has a unique identifier, facilitating subsequent querying and analysis.

[0070] Semantic data from different patients exhibit diverse characteristics. For example, in stress testing scenarios, some patients show more frequent dialogues, shorter semantic duration, longer semantic decision-making time, and more repetitions; while in everyday communication scenarios, dialogues are more evenly distributed, semantic duration is longer, semantic decision-making time is shorter, and repetitions are fewer. These differences provide foundational data for subsequent assessment phase segmentation, key scenario selection, and assessment indicator construction. Comprehensive analysis of this multi-dimensional semantic data allows for a more complete understanding of patients' language behavior patterns in different scenarios, thus providing information support for the dynamic assessment of their psychological state.

[0071] During data collection, it is crucial to avoid data errors caused by equipment malfunctions or environmental interference. For example, ensure clear audio recording and minimize the impact of background noise on speech-to-text conversion. Simultaneously, regularly verify the collected semantic data, checking the accuracy of dialogue occurrence times, the reasonableness of semantic duration and decision-making time, and the precision of repeated expression counts to guarantee data quality. Furthermore, individual patient differences must be considered. Patients of different ages, genders, and educational backgrounds exhibit varying language expression habits, which should be appropriately addressed during data collection and analysis to ensure that the semantic data accurately reflects the patient's actual conversational state.

[0072] Example 2:

[0073] The acquisition of generative-driven assessment indicators for each patient under each key semantic type during the stabilization assessment phase needs to be carried out in a specific key semantic type dialogue scenario.

[0074] The scope of patients to be screened and patients to be evaluated is clearly defined. All patients with semantic data present during the stable evaluation phase are included in the patient to be screened cohort, and all patients with semantic data present during the change evaluation phase are included in the patient to be evaluated cohort. In practice, one patient is randomly selected from the patient to be screened as the selected patient to be screened, and one patient is randomly selected from the patients to be evaluated as the selected patient to be evaluated. By analyzing the semantic data of these two groups, generative-driven evaluation indicators are derived step by step.

[0075] The analysis first requires determining the degree of influence of the selected patient to be screened on each dialogue behavior of the selected patient to be assessed. This process involves considering the timing of each dialogue, the duration of semantic continuity, and similarity indicators between the two patients. The timing of the dialogue refers to the specific point in time during each dialogue; the duration of semantic continuity is the length of time during which semantics remain coherent; and the similarity indicators are derived by comparing the overlap and consistency between the two in dimensions such as dialogue text content, emotional keywords, and semantic coherence. Through comprehensive analysis of this information, the degree of influence of the selected patient to be screened on each dialogue behavior of the selected patient to be assessed can be quantified.

[0076] The semantic similarity between selected patients to be screened and selected patients to be evaluated is calculated to obtain the semantic similarity factor corresponding to the selected patient to be evaluated for each dialogue behavior. Specifically, the semantic data of each dialogue behavior of the selected patients to be screened during the stable evaluation phase is compared one by one with the semantic data of each dialogue behavior of the selected patients to be evaluated during the variable evaluation phase. The comparison includes the overlap of words in the dialogue text, the degree of matching of emotional keywords, the similarity of semantic duration and semantic decision duration, etc. Through the comprehensive processing of these comparison results, the semantic similarity factor corresponding to each dialogue behavior is obtained. This factor is used to reflect the degree of semantic association between the two in a specific dialogue behavior.

[0077] After obtaining the degree of dialogue influence and semantic similarity factor, the degree of dialogue influence is used as a weight to calculate a weighted average of the semantic similarity factor for each dialogue behavior, thus obtaining the psychological assessment power of the selected patients to be screened relative to the selected patients being assessed. In this calculation process, the magnitude of the weight directly affects the final average result; the greater the degree of dialogue influence, the higher the proportion of its corresponding semantic similarity factor in the calculation. This weighting method can more accurately reflect the differences in the role of different dialogue behaviors in the assessment process.

[0078] The average psychological assessment ability of the selected patients relative to all assessed patients is calculated, and the resulting mean is the generative-driven assessment index for the selected patients. This means that the generative-driven assessment index for each selected patient needs to be calculated by comparing it with all assessed patients in the same key semantic type dialogue scenario. By comparing with multiple assessed patients, the influence of individual differences on the index results can be reduced, making the generative-driven assessment index more representative.

[0079] Throughout the process, the comparison of dialogue text content utilizes natural language processing techniques for word segmentation and feature extraction, followed by calculation of the correlation between text vectors. Emotional keyword matching is performed using a pre-defined emotional lexicon, statistically analyzing the number and percentage of successfully matched keywords. The comparison of semantic coherence primarily focuses on the numerical differences in semantic duration and semantic decision duration. All these processing steps are automated using AI algorithms, preserving original data and intermediate results for subsequent verification and traceability.

[0080] While the calculation process for generative-driven assessment metrics remains consistent across dialogue scenarios with varying key semantic types, the input semantic data differs due to variations in the scenarios themselves, leading to discrepancies in the final metrics results. For instance, in stress testing scenarios, patients' emotional keywords are more intense, resulting in greater fluctuations in semantic duration and semantic decision-making time. Consequently, the calculated generative-driven assessment metrics may exhibit different distribution characteristics compared to results from everyday communication scenarios. This variation in metrics due to scenario differences precisely reflects the impact of different scenarios on the assessment of patients' psychological states, providing a basis for subsequent dynamic assessment node selection.

[0081] In practical applications, the numerical range of generative-driven evaluation metrics may vary depending on the data processing method. However, regardless of the numerical value, its core function is to be used for horizontal comparison between different patients to be screened, as well as vertical comparison of the same patient in different key semantic type dialogue scenarios.

[0082] Example 3:

[0083] To determine the impact of each dialogue action of the selected patient to be screened relative to the selected patient to be evaluated, analysis should begin with individual dialogue actions. Any dialogue action in any sequence is designated as the target dialogue action. This action can be any verbal interaction during the dialogue, whether initiated by the patient to be screened or the patient to be evaluated. As long as it falls within the scope of the analysis as a valid dialogue action within the key semantic type of the dialogue scenario, it is included in the analysis.

[0084] Before analyzing the target dialogue behavior, it is necessary to determine the degree of semantic association between the selected patients to be screened and the selected patients to be evaluated. This degree of association is derived based on the semantic similarity of the two in each dimension of semantic data. The semantic data in each dimension includes dialogue text content, emotional keyword information, and semantic coherence information. For dialogue text content, the text is segmented to extract core words and thematic information, and the overlap between the two in vocabulary usage and thematic expression is compared. For emotional keyword information, the matching of emotional keywords used by the two in terms of category and quantity is statistically analyzed according to a pre-defined emotion classification system. For semantic coherence information, the similarity between the two in terms of semantic duration and semantic decision duration is compared. The similarity of these dimensions is then comprehensively processed to obtain the degree of semantic association, which reflects the overall semantic closeness between the two.

[0085] After determining the degree of semantic relevance, the first and second correlation coefficients are calculated. The first correlation coefficient is the ratio between the time of dialogue occurrence of the selected patient being evaluated and the time of dialogue occurrence of the selected patient to be screened under the same target dialogue behavior. The time of dialogue occurrence is the specific point in time when the dialogue behavior occurs. If the time of dialogue occurrence of the selected patient being evaluated under the target dialogue behavior is T1, and the time of dialogue occurrence of the selected patient to be screened under the same target dialogue behavior is T2, then the first correlation coefficient is the ratio of T1 to T2. When the time of dialogue occurrence of the two patients is relatively close, the ratio approaches 1; if the time difference is large, the ratio will deviate from 1, and the specific degree of deviation depends on the size of the time difference.

[0086] The second correlation coefficient is the ratio between the degree of semantic relevance and the semantic duration of the selected patient in the target dialogue behavior. Semantic duration is the length of time the selected patient maintains semantic coherence during the target dialogue behavior. If the degree of semantic relevance is R and the semantic duration is D, then the second correlation coefficient is the ratio of R to D. This ratio combines the strength of semantic relevance and the duration of the dialogue; the higher the degree of semantic relevance and the shorter the semantic duration, the larger the ratio; conversely, the smaller the ratio.

[0087] Finally, the product of the first correlation coefficient and the second correlation coefficient is calculated, and the result is the degree of influence of the selected patients to be screened on the dialogue behavior relative to the selected patients to be evaluated.

[0088] In practice, the above steps need to be followed for each target dialogue behavior. For example, in a dialogue scenario about "childhood experiences," the dialogue between selected patient A and selected patient B in the 5th round of dialogue (i.e., the target dialogue behavior) occurs at 10:05:23 and 10:05:25 respectively. The semantic association degree is calculated to be 0.7, and the semantic duration of patient B in this round of dialogue is 45 seconds. Then the first association coefficient is 25 / 23 (this is only a simplified example of the time point values), and the second association coefficient is 0.7 / 45. The product of the two is the degree of influence of patient A on patient B in this round of dialogue.

[0089] For multi-round dialogues, the degree of influence needs to be calculated for each round to form a sequence of influence encompassing all target dialogue behaviors. This sequence reflects the changes in the influence of the selected patients on the selected patients being assessed throughout the entire dialogue process. Some rounds show a higher degree of influence, while others show a lower degree. These changes are related to factors such as the depth of the dialogue content and emotional fluctuations.

[0090] During the calculation process, all data processing is automated by the AI ​​system. The system reads the stored semantic data, compares and calculates it according to a preset process, and stores the results under the corresponding database entries. The calculation of the impact of dialogues between different patients is independent and does not interfere with each other, ensuring the independence and accuracy of each result. The dialogue impact obtained in this way can provide a quantitative weighting basis for the subsequent calculation of generative-driven assessment indicators, making the calculation of assessment indicators more closely reflect the interaction in actual dialogues.

[0091] The impact of different target dialogue behaviors on the same patient will exhibit different numerical distributions. In some target dialogue behaviors, the impact is higher due to their proximity in time and strong semantic connection; while in others, the impact may be lower due to longer time intervals or weaker semantic connection. These differences directly affect the weighted calculation of subsequent semantic similarity factors, enabling the final evaluation indicators to more accurately reflect the actual role of different dialogue behaviors in the evaluation process.

[0092] Example 4:

[0093] Based on the dialogue text content and semantic coherence information corresponding to each patient during the change assessment phase, as well as the repetitive expression data, and combined with generative-driven assessment indicators, the probability of dynamic assessment nodes for each patient under each key semantic type during the stable assessment phase is obtained, as follows:

[0094] The dialogue decision-making ability index for selected patients is derived by analyzing the semantic duration, semantic decision-making duration, and number of repetitions in each dialogue. In practice, semantic duration refers to the time a patient maintains semantic coherence within a single dialogue. For example, if a patient discusses "work stress" and the conversation lasts for two minutes from the start to the next topic, this is the semantic duration of that dialogue. Semantic decision-making duration refers to the interval between the interruption of the previous dialogue and the start of the current one. For instance, if a patient pauses for 30 seconds after discussing "family relationships" before starting to discuss "health issues," this 30-second pause is the semantic decision-making duration of the current dialogue. The number of repetitions is the number of times a patient repeatedly mentions similar content during the stable assessment phase. For example, when discussing "sleep quality," if a patient repeatedly mentions phrases like "can't sleep at night" or "easily wakes up in the early morning," these will be counted as repetitions.

[0095] When calculating the dialogue decision-making ability index, it is first necessary to determine the feature ratio of each dialogue behavior, that is, the relative relationship between semantic duration and semantic decision-making duration. For example, if the semantic duration in a dialogue is 120 seconds and the semantic decision-making duration is 60 seconds, the feature ratio is the numerical ratio of the two. The feature ratios of all dialogue behaviors within the stable assessment phase are averaged to obtain the patient's overall feature level in this phase. Then, this average is multiplied by the number of repetitions to obtain the dialogue decision-making ability index. For example, if a patient's average feature ratio is 2.5 and the number of repetitions is 3, the product of the two is the patient's dialogue decision-making ability index.

[0096] After obtaining the generative-driven assessment index and the dialogue decision-making ability index, the two are multiplied and normalized. The resulting value represents the patient's dynamic assessment node probability. The generative-driven assessment index reflects the patient's driving ability to assess other patients, while the dialogue decision-making ability index reflects the patient's decision-making characteristics and expressive stability in their own dialogue. The combination of the two comprehensively reflects the patient's suitability as an assessment node. Normalization transforms the product result to a specific range, such as between 0 and 1, making the dynamic assessment node probabilities of different patients comparable.

[0097] In a specific scenario, such as a dialogue scenario focusing on the key semantic type of "stress coping," a patient was selected who exhibited five dialogue behaviors during the stabilization assessment phase. The first dialogue had a semantic duration of 90 seconds and a semantic decision duration of 30 seconds, with a feature ratio of 3. The second dialogue had a semantic duration of 60 seconds and a semantic decision duration of 40 seconds, with a feature ratio of 1.5. The third dialogue had a semantic duration of 150 seconds and a semantic decision duration of 50 seconds, with a feature ratio of 3. The fourth dialogue had a semantic duration of 120 seconds and a semantic decision duration of 60 seconds, with a feature ratio of 2. The fifth dialogue had a semantic duration of 80 seconds and a semantic decision duration of 20 seconds, with a feature ratio of 4. The average of these five feature ratios is (3 + 1.5 + 3 + 2 + 4) ÷ 5 = 2.7. If the patient repeated these behaviors four times during the stabilization assessment phase, the dialogue decision-making ability index would be 2.7 × 4 = 10.8. Assuming the patient's generative-driven assessment index is 0.6, multiplying the two yields 6.48. After normalization, if the result is 0.7, then the probability of the patient's dynamic assessment node is 0.7. In this embodiment, the calculation process of patient A's dialogue decision-making ability index (2.7×4=10.8) specifically corresponds to the semantic coherence information rounds and repeated expression data in the change assessment stage; the semantic continuation duration (e.g., 90 seconds for the first dialogue, 60 seconds for the second) and semantic decision duration (e.g., 30 seconds for the first dialogue, 40 seconds for the second) are derived from the round statistics corresponding to the semantic coherence information in the change assessment stage; the number of repeated expressions (4 times) is directly derived from the repeated expression data statistics in the change assessment stage; the mean feature ratio (2.7) is calculated by the ratio of semantic continuation duration to semantic decision duration, reflecting the coherence characteristics of the dialogue text content. The subsequent calculation of the probability of the dynamic assessment node (0.7) in combination with the generative-driven assessment index (0.6) is precisely the specific implementation method of combining the change assessment stage data with the generative-driven assessment index.

[0098] The likelihood of a dynamic assessment node varies among patients due to differences in generative-driven assessment metrics and dialogue decision-making ability metrics. Some patients may have high generative-driven assessment metrics but low dialogue decision-making ability metrics, resulting in a moderate final likelihood value; others may have high metrics in both areas, leading to a high likelihood of a dynamic assessment node. These numerical differences provide a clear basis for subsequent screening of assessment nodes, facilitating the selection of the most suitable assessment node from patients in the stable assessment phase.

[0099] This calculation method remains consistent across multiple rounds of dialogue and various scenarios, but the specific values ​​will vary depending on the patient's dialogue performance. For example, in an "interpersonal relationship" dialogue scenario, a patient with a generally longer semantic continuation time, shorter semantic decision-making time, higher mean feature ratio, and fewer repetitions may have a moderate level of dialogue decision-making ability. However, if their generative-driven evaluation index is high, their final dynamic evaluation node probability may still be high. This approach allows for a comprehensive assessment of the patient's performance in different key semantic types of scenarios, providing foundational data for constructing an evaluation chain tree structure.

[0100] Throughout the process, all data calculations are based on the actual collected semantic data and are completed automatically by the AI ​​system without human intervention. The system records intermediate results for each calculation, including feature ratios, means, and products, for subsequent traceability and verification. The calculation logic remains consistent across dialogue scenarios with different key semantic types, ensuring the comparability of results across different scenarios and providing coherent information support for the development of dynamic psychological state assessment schemes.

[0101] Example 5:

[0102] Based on the probability of dynamic assessment nodes, each patient under each key semantic type in the stable assessment phase is selected as an assessment node. Combining the semantic data of each patient under the same key semantic type in the changing assessment phase, an assessment chain tree structure is constructed, as follows:

[0103] Patients whose likelihood of being identified as a dynamic assessment node exceeds a preset assessment threshold are designated as assessment nodes in each tree of the assessment chain structure. The assessment threshold is set based on the characteristics of the dialogue scenario and key semantic types; the threshold may differ across scenarios. For example, in dialogue scenarios with significant emotional fluctuations, the assessment threshold may be set relatively low to include more patients with potential assessment value; while in scenarios with more stable emotions, the assessment threshold may be set higher to ensure that the selected assessment nodes are more representative.

[0104] The dialogue decision-making ability index of each assessed patient was obtained. The product of the degree of dialogue influence between each assessed patient and the corresponding screened patients at the assessment node and the patient's dialogue decision-making ability index was used as the patient's compliance index relative to that assessment node. The dialogue decision-making ability index reflects the assessed patient's decision-making characteristics and expression stability in dialogue, while the degree of dialogue influence reflects the strength of the assessment node's influence on the assessed patient. The product of the two can comprehensively reflect the closeness of the association between the assessed patient and the assessment node, as well as the patient's own dialogue characteristics.

[0105] For any given assessment node, a tree structure is constructed based on the compliance index of each assessed patient relative to that assessment node, in descending order. Nodes at the same level within the tree structure correspond to patients with the same compliance index. During construction, the patient with the highest compliance index is first placed as the first-level child node of the assessment node. If multiple patients have the same highest compliance index, they are placed at the same level. Next, the patient with the second highest compliance index is placed as the next-level child node, and so on, until all assessed patients are included in the tree structure.

[0106] The tree structure of all assessment nodes constitutes an assessment chain tree structure. Each assessment node corresponds to an independent tree, and there may be shared child nodes between trees. That is, a patient being assessed may belong to the child nodes of multiple assessment nodes at the same time, depending on the compliance index between the patient being assessed and different assessment nodes.

[0107] Based on the assessment chain tree structure, a dynamic assessment plan for the psychological state of different patients is determined as follows: In the assessment chain tree structure, the layer containing the first common child node of different assessment nodes is designated as the first-stage target layer; after the first-stage target layer, the layer with the most child nodes is designated as the second-stage target layer; for patients between the assessment node and the first-stage target layer, a first-type preset assessment strategy is used; for patients between the first-stage target layer and the second-stage target layer, a second-type preset assessment strategy is used; and for patients between the second-stage target layer and the bottom layer, a third-type preset assessment strategy is used.

[0108] The first type of pre-defined assessment strategy focuses on a detailed analysis of the patient's basic psychological characteristics, including emotional tendencies and semantic coherence stability in the dialogue text, capturing subtle changes in psychological state through frequent short-term assessments. The second type of pre-defined assessment strategy, building on the basic analysis, adds tracking of trends in the patient's psychological state, combining semantic data changes from multiple rounds of dialogue to form phased assessment conclusions. The third type of pre-defined assessment strategy places greater emphasis on assessing the long-term stability of psychological state, identifying potential patterns and regularities in psychological state by integrating semantic data over a longer time span.

[0109] In practical applications, such as dialogue scenarios focusing on the key semantic type of "family relationships," three assessment nodes were identified after screening. Within the tree structure corresponding to each assessment node, the compliance indicators of the assessed patients differ, and multi-layered sub-nodes are constructed through sorting. The first shared sub-node of different assessment nodes appears at level 3, thus level 3 is designated as the first-stage target level. Following the first-stage target level, level 5 has the most sub-nodes and is designated as the second-stage target level. For patients at assessment nodes up to level 3, the first type of assessment strategy is used, with assessments twice a week; for patients at levels 3 to 5, the second type of assessment strategy is used, with assessments once every two weeks; and for patients from level 5 to the bottom, the third type of assessment strategy is used, with assessments once a month. This hierarchical assessment scheme allows for targeted and dynamic assessments based on the patient's position within the assessment chain tree structure.

[0110] Example 6:

[0111] Based on the trend of the number of emotional keywords changing with the number of dialogue rounds in each dialogue scenario, the assessment is divided into a stable assessment phase and a change assessment phase, as detailed below:

[0112] For any dialogue scenario, first collect the number of emotional keywords appearing in each round of dialogue within that scenario. These numbers are arranged in chronological order to form a keyword count sequence. For example, in the "daily communication" scenario, 3 emotional keywords appear in round 1, 5 in round 2, 4 in round 3, and so on. These numbers form a sequence in chronological order. Then, calculate the first-order difference of this sequence, which is the number of emotional keywords in the later round minus the number of emotional keywords in the previous round. If the first-order difference of a round is positive, it indicates that the number of emotional keywords in that round has increased compared to the previous round, and such rounds are marked as increasing rounds.

[0113] Arrange all growth cycles in ascending order of their corresponding first-order difference values ​​to form a difference sequence. For example, if the first-order difference values ​​for growth cycles are 2, 1, and 3, arranging them in ascending order yields the difference sequence 1, 2, 3. Next, calculate the second-order difference value of this difference sequence, which is the difference between the previous and subsequent values. Find the maximum value among the second-order difference values; the growth cycle corresponding to this maximum value is determined as the characteristic cycle. All cycles before the characteristic cycle constitute the stability evaluation phase, and all cycles after the characteristic cycle constitute the change evaluation phase. For example, if the maximum value of the second-order difference value in the difference sequence appears in the 3rd growth cycle, then this growth cycle is the characteristic cycle, the cycles before it belong to the stability evaluation phase, and the cycles after it belong to the change evaluation phase.

[0114] Based on the difference in the number of emotion keywords during the stable evaluation phase and the change evaluation phase in each dialogue scenario, dialogue scenarios with key semantic types were selected as follows:

[0115] For any dialogue scenario, calculate the average number of emotional keywords across all rounds in the stable evaluation phase (the first mean); calculate the average number of emotional keywords across all rounds in the change evaluation phase (the second mean). Then calculate the difference between the second mean and the first mean, and normalize this difference to obtain the semantic difference coefficient. Normalization converts the difference into a relative value, making it unaffected by the original data volume. If the semantic difference coefficient is greater than a preset difference threshold, the dialogue scenario is identified as a dialogue scenario with a key semantic type.

[0116] In practice, the processing flow is consistent across different dialogue scenarios, but the specific data will exhibit different characteristics. For example, in a "stress test" scenario, the number of emotional keywords in the stable assessment phase may be relatively small and increase gradually, while the number increases significantly in the change assessment phase. The first-order difference value changes more greatly, and the feature rounds may appear in later rounds, resulting in a shorter stable assessment phase and a longer change assessment phase. Its first mean might be 4, the second mean might be 10, and the difference might be 6. If the semantic difference coefficient after normalization is greater than the set difference threshold of 0.5, then the scenario is classified as a key semantic type.

[0117] In the "interests and hobbies" scenario, the number of emotional keywords may change relatively smoothly at different stages, the first-order difference value is small, the feature rounds are not obvious, the difference between the first mean and the second mean is small, and the semantic difference coefficient may be lower than the difference threshold. Therefore, it is not listed as a key semantic type.

[0118] This approach allows for the selection of scenarios from various dialogue contexts where the number of emotional keywords varies significantly across different stages. These scenarios better reflect the changing characteristics of the patient's psychological state and are suitable as key semantic types for in-depth analysis, providing a focused scope for subsequent assessment indicator calculations and assessment plan development. During data processing, all calculations are automated by the AI ​​system, which records the results of each step, including the keyword quantity sequence, first-order difference value, second-order difference value, mean, difference, and semantic difference coefficient, for easy review and verification. The difference thresholds for different scenarios can be adjusted according to actual assessment needs to accommodate varying assessment precision and scenario characteristics.

[0119] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0120] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A dynamic AI assessment method for adolescent psychological problems based on semantic analysis, characterized in that, The method includes the following steps: Semantic data of each patient in different dimensions is obtained in different dialogue scenarios; the semantic data includes dialogue text content, emotional keyword information, and semantic coherence information. Based on the trend of the number of emotional keywords changing with the number of dialogue rounds in each dialogue scenario, we divide the dialogue into a stable evaluation phase and a change evaluation phase; based on the difference in the number of emotional keywords in the stable evaluation phase and the change evaluation phase in each dialogue scenario, we select dialogue scenarios with key semantic types. In each key semantic type dialogue scenario, based on the similarity of the semantic data of each patient in the stable assessment phase and each patient in the change assessment phase, combined with round information and the patient's semantic coherence information, the generative driving assessment index of each patient in each key semantic type in the stable assessment phase is obtained. Based on the dialogue text content and semantic coherence information of each patient during the change assessment phase, as well as the repetitive expression data, and combined with the generative-driven assessment indicators, the probability of dynamic assessment nodes for each patient under each key semantic type during the stable assessment phase is obtained. Based on the probability of the dynamic assessment nodes, each patient under each key semantic type in the stable assessment phase is selected as an assessment node. Combining the semantic data of each patient under the same key semantic type in the change assessment phase, an assessment chain tree structure is constructed. Based on the assessment chain tree structure, a dynamic assessment scheme for the psychological state of different patients is determined.

2. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 1, characterized in that, The dialogue text content includes the time when each dialogue action occurs; the semantic coherence information includes the semantic duration of each dialogue action and the time interval between the semantic interruption and the time when the dialogue occurs under each dialogue action, which is called the semantic decision duration of each dialogue action; the semantic data also includes the number of times the expression is repeated.

3. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 2, characterized in that, The specific methods for obtaining generative-driven assessment indicators for each patient under each key semantic type during the stable assessment phase include: In any dialogue scenario with a key semantic type, patients with the semantic data present in the stable assessment phase are all designated as patients to be screened, and patients with the semantic data present in the changing assessment phase are all designated as patients to be assessed; any patient to be screened is recorded as a selected patient to be screened, and any patient to be assessed is recorded as a selected patient to be assessed. Based on the timing of each dialogue, semantic duration, and similarity index between the selected patients and the selected patients being evaluated, the degree of dialogue influence of each dialogue behavior of the selected patients relative to the selected patients being evaluated is obtained. The semantic similarity between the semantic data of each dialogue behavior of the selected patients to be screened during the stable assessment phase and the semantic data of each dialogue behavior of the selected patients to be assessed during the variable assessment phase is calculated to obtain the semantic similarity factor of the selected patients to be assessed for each dialogue behavior. Using the degree of influence of the dialogue as a weight, the semantic similarity factor corresponding to the selected patient under each dialogue behavior is weighted and averaged to obtain the psychological assessment power of the selected patient to be screened relative to the selected patient under evaluation. The mean of the psychological assessment power of the selected patient to be screened relative to all patients under evaluation is used as the generative driving assessment index of the selected patient to be screened.

4. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 3, characterized in that, The method for obtaining the degree of dialogue influence of each dialogue behavior of the selected patient to be screened relative to the selected patient to be evaluated specifically includes: Record any dialogue action in any order as the target dialogue action; Based on the semantic similarity between the semantic data of the selected patients to be screened in each dimension and the semantic data of the selected patients to be evaluated in each dimension, the degree of semantic association between the selected patients to be screened and the selected patients to be evaluated is determined. The ratio between the time of dialogue occurrence of the selected patients being evaluated and the selected patients to be screened under the target dialogue behavior is used as the first correlation coefficient; the ratio between the degree of semantic correlation and the semantic duration of the selected patients being evaluated under the target dialogue behavior is used as the second correlation coefficient. The product of the first correlation coefficient and the second correlation coefficient is calculated to obtain the degree of dialogue influence of the selected patients to be screened relative to the selected patients to be evaluated under the target dialogue behavior.

5. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 3, characterized in that, The method involves using the dialogue text content and semantic coherence information of each patient during the change assessment phase, along with the corresponding rounds and repetitive expression data, and combining this with the generative-driven assessment indicators to obtain the probability of dynamic assessment nodes for each patient under each key semantic type during the stable assessment phase. Specifically, this includes: The dialogue decision-making ability index of the selected patients to be screened was obtained by analyzing the semantic duration and semantic decision-making duration of each dialogue behavior and the number of repeated expressions. The normalized value of the product of the generative-driven assessment index and the dialogue decision-making index of the selected patients to be screened is used as the probability of the dynamic assessment node of the selected patients to be screened.

6. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 5, characterized in that, The method for obtaining the dialogue decision-making ability index of the selected patients based on the semantic duration and semantic decision-making duration of each dialogue behavior and the number of repetitions specifically includes: The characteristic ratio of semantic duration to semantic decision duration for each dialogue behavior of the selected patients to be screened was calculated. The product of the mean of the characteristic ratios corresponding to all dialogue behaviors of the selected patients to be screened during the stable assessment phase and the number of repeated expressions of the selected patients to be screened during the stable assessment phase was used as the dialogue decision-making ability index of the selected patients to be screened.

7. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 5, characterized in that, The step of selecting each patient under each key semantic type in the stable assessment phase as an assessment node based on the probability of the dynamic assessment node, and constructing an assessment chain tree structure by combining the semantic data of each patient under the same key semantic type in the change assessment phase, specifically includes: Patients whose probability of being a dynamic assessment node is greater than a preset assessment threshold are respectively used as assessment nodes in each tree of the assessment chain tree structure. Obtain the dialogue decision-making ability index of each evaluated patient, and multiply the degree of dialogue influence between each evaluated patient and the screening patients corresponding to the evaluation node with the dialogue decision-making ability index of the evaluated patient as the compliance index of each evaluated patient relative to the evaluation node. For any assessment node, a tree structure is constructed according to the compliance index of each assessed patient relative to that assessment node in descending order. The compliance index of the assessed patients corresponding to nodes at the same level in the tree structure is the same. The tree structure of all evaluation nodes constitutes the evaluation chain tree structure.

8. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 7, characterized in that, The dynamic assessment scheme for determining the psychological state of different patients based on the assessment chain tree structure specifically includes: In evaluating the chain tree structure, the layer containing the first common child node of different evaluation nodes is taken as the first-stage target layer; after the first-stage target layer in the evaluation chain tree structure, the layer containing the largest number of child nodes is taken as the second-stage target layer. For patients between the assessment node and the first-stage target layer, a first-type preset assessment strategy is used; for patients between the first-stage target layer and the second-stage target layer, a second-type preset assessment strategy is used; and for patients between the second-stage target layer and the bottom layer, a third-type preset assessment strategy is used.

9. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 1, characterized in that, The method divides the evaluation into a stable assessment phase and a change assessment phase based on the changing trend of the number of emotional keywords in each dialogue scenario with the number of dialogue rounds. Specifically, it includes: For any dialogue scenario, obtain the keyword quantity sequence consisting of the number of emotional keywords in each round, calculate the first difference value of the keyword quantity sequence, and select the rounds with positive first difference values ​​as the growth rounds. Arrange the growth cycles in ascending order of their corresponding first-order difference values ​​to obtain a difference sequence. Calculate the second-order difference value of the difference sequence. Take the growth cycle corresponding to the maximum value of the second-order difference value as the characteristic cycle. The stage before the characteristic cycle is the stability evaluation stage, and the stage after the characteristic cycle is the change evaluation stage.

10. The AI-based dynamic assessment method for adolescent psychological problems based on semantic analysis according to claim 9, characterized in that, The process of selecting dialogue scenarios with key semantic types based on the difference in the number of emotion keywords during the stable evaluation phase and the change evaluation phase in each dialogue scenario specifically includes: For any dialogue scenario, obtain the first mean of the number of emotional keywords during the stable evaluation phase and the second mean of the number of emotional keywords during the change evaluation phase; normalize the difference between the second mean and the first mean to obtain the semantic difference coefficient; if the semantic difference coefficient is greater than the preset difference threshold, then the dialogue scenario is regarded as a dialogue scenario with key semantic type.

Images