Intelligent state parallel evaluation method and system based on high-standard verification data

Abstract

This invention relates to the field of intelligent data assessment technology, specifically disclosing an intelligent parallel state assessment method and system based on high-standard verification data. The invention collects first-response information from students' original assessment questions via a psychological testing terminal, along with psychological warning levels confirmed by manual evaluation. The first-response information includes numerical information features, textual information features, and answer behavior features. A joint judgment mechanism for multiple abnormal behaviors is constructed using the answer behavior features to identify abnormal student answer behaviors. A multi-abnormality screening strategy is developed based on the original answer information set to obtain third-response information. Using this third-response information as input, a non-linear mapping relationship is learned from the original psychological scale questions to the student's psychological state. Then, a dual-path parallel intelligent psychological decision-making mechanism is executed to determine the student's psychological state. This invention can achieve higher assessment credibility, lower misjudgment risk, and stronger management and collaboration capabilities in real educational scenarios.

Description

Technical Field

[0001] This invention relates to the field of intelligent data evaluation technology, and more specifically, to an intelligent state parallel evaluation method and system based on high-standard verification data. Background Technology

[0002] With the continuous advancement of mental health education in primary and secondary schools, regular screening of students' mental health has gradually become an important component of school daily management and education governance. Currently, in practical applications, student mental health screening mainly relies on standardized psychological scale assessment methods, such as the Symptom Checklist-90-R (SCR) and mental health tests. Students are organized to complete the questionnaires in a centralized or online manner to conduct a preliminary assessment of their mental health. However, the aforementioned existing technical solutions are essentially linear assessment models based on static rules. Their assessment results are highly dependent on the fixed scoring logic and preset threshold divisions of the scale items, making it difficult to reflect the complexity and dynamism of students' mental health. They also lack analysis of the correlation between student mental health samples, and for students with serious mental health problems, there is a high probability of insensitive identification and missed detection. In practical applications, students may be affected by factors such as temporary emotional fluctuations, social expectation biases, or perfunctory responses, leading to a significant deviation between the answers and their true, long-term mental health status. This reduces the accuracy and reliability of the screening results, making it difficult to meet the current educational context's practical needs for mental health screening that emphasizes accuracy, safety, and feasibility.

[0003] Therefore, it is necessary to provide an intelligent state parallel evaluation method and system based on high-standard verification data to solve the above-mentioned technical problems. In order to solve the above problems, a technical solution is provided. Summary of the Invention

[0004] To overcome the aforementioned shortcomings of existing technologies, this invention provides an intelligent parallel state assessment method and system based on high-standard verification data. This method addresses the problem that existing mental health screening in primary and secondary schools mainly relies on standardized psychological scales, employs static rules and linear scoring logic, and is therefore unable to reflect the complexity and dynamic changes in students' mental states.

[0005] To achieve the above objectives, the present invention provides the following technical solution: A parallel intelligent state evaluation method based on high-standard verification data, the specific steps of which are as follows: The system collects students' initial test responses and psychological warning levels confirmed by manual evaluation through a psychological testing terminal. The initial response information is preprocessed to form an initial response information set. The initial response information includes numerical information features, textual information features, and response behavior features. A joint judgment mechanism for multiple abnormal behaviors is constructed by identifying the characteristics of answering behavior, and an abnormal answering risk score is output in real time to identify students' abnormal answering behavior. Students with abnormal answering behavior are marked and removed. A multi-layered anomaly screening strategy was developed based on the original question answer information set to obtain third-party question answer information; Using the third-party answer information as input and the psychological warning level as the real label, a random forest model is constructed to calculate the first importance score. By constructing multiple decision trees to synthesize the first importance score, the non-linear mapping relationship from the original psychological scale items to the student's psychological state is learned. Based on the first answer information, a second importance score is output. The first psychological warning level is determined based on the first importance score, and the second psychological expectation level is determined based on the second importance score. The student's psychological state is determined by a dual-path parallel intelligent psychological decision-making mechanism that uses the first and second psychological warning levels.

[0006] As a further aspect of the present invention, the answering behavior features include the duration of answering a single question; the textual information features include the question text content, the question keyword set, and the question semantic vector representation.

[0007] As a further aspect of the present invention, a joint judgment mechanism for multiple abnormal behaviors is constructed based on the characteristics of answering behavior, and an abnormal answering risk score is output in real time to identify abnormal student answering behavior. Student information exhibiting abnormal answering behavior is then marked and removed. The specific steps are as follows: During the assessment and answering process, the answering time for each question is recorded in real time to form a corresponding answering time sequence, which is then stored in association with the question number to construct a dataset of students' answering behavior characteristics. Based on the time series of responses, calculate the characteristic indicators of response time, including the mean response time for a single question and the standard deviation of the response time for a single question; Based on the answer duration characteristic index, construct an index for judging abnormal answering behavior; Using abnormal answering behavior indicators as input, a joint judgment mechanism for multiple abnormal behaviors is constructed to calculate the abnormal answering risk score in real time and quantify the degree of abnormal answering behavior of students during the answering process. The abnormal answer risk score calculated in real time is compared with the preset risk threshold. When the abnormal answer risk score is lower than the preset risk threshold, the student's answering behavior is determined to be normal; when the abnormal answer risk score is higher than or equal to the preset risk threshold, the student's answering behavior is determined to be abnormal. For students whose answering behavior is deemed abnormal, abnormal answering information is generated in their corresponding assessment data. The abnormal answering information includes an abnormality type identifier and risk score results. The assessment data of students with abnormal answering behavior are then removed.

[0008] As a further aspect of the present invention, the indicators for judging abnormal answering behavior include indicators for random answering and indicators for sudden abnormality.

[0009] As a further aspect of the present invention, an abnormal answering behavior judgment index is used as input to construct a multi-abnormal behavior joint judgment mechanism, which calculates the abnormal answering risk score in real time. The multi-abnormal behavior joint judgment mechanism is as follows: Extract the random answering abnormal indicators and the sudden abnormal indicators, and compare the random answering abnormal indicators with the first abnormal indicator. If the random answering abnormal indicator is greater than or equal to the first abnormal indicator, it is determined that the student has abnormal answering behavior; if the random answering abnormal indicator is less than the first abnormal indicator, an abnormal answering risk assessment is triggered. The sudden abnormality indicator is compared with the first abnormality indicator. If the sudden abnormality indicator is greater than or equal to the first abnormality indicator, the student is judged to have abnormal answering behavior; if the sudden abnormality indicator is less than the first abnormality indicator, the abnormal answering risk assessment is triggered. The abnormal answer risk assessment is specifically conducted by weighting the abnormal indicators of random answering and sudden abnormality to obtain an abnormal answer risk score.

[0010] As a further aspect of the present invention, a multi-layered anomaly screening strategy is formulated based on the original question-answering information set to obtain the third question-answering information, specifically as follows: By performing the first round of anomaly screening on the numerical information features in the first answer information, the boundary compression of extreme values ​​is achieved, and the second answer information is output by integrating the textual information features and answer behavior features. The second round of filtering is performed by extracting numerical and textual information features from the second answer information. After identifying and eliminating similar types of questions, the third answer information is output.

[0011] As a further aspect of the present invention, the boundary compression of extreme values ​​is achieved by performing a first-level anomaly screening on numerical information features. Specifically, this involves extracting numerical information features from the first answer information, performing tail reduction processing on each numerical information feature, and pulling extreme values ​​that exceed the 5% and 95% quantiles back to the quantile boundaries.

[0012] As a further aspect of the present invention, a second round of filtering is performed by extracting numerical and textual information features from the second answer information. After identifying and eliminating similar types of questions, the third answer information is output. The specific steps are as follows: The text similarity between different questions is calculated based on textual information features. When the text similarity of a question is higher than a preset similarity threshold, the corresponding question is determined to be a candidate duplicate question. Extract the numerical information features of candidate duplicate questions, calculate the difference, and compare the difference with a preset difference. If the difference is greater than or equal to the preset difference, the candidate duplicate questions are not similar questions; if the difference is less than the preset difference, the candidate duplicate questions are similar questions. For questions of similar types, keep one of them and remove the rest.

[0013] As a further aspect of the present invention, a dual-path parallel intelligent psychological decision-making mechanism is used to determine the student's psychological state through a first psychological warning level and a second psychological warning level. The specific steps are as follows: If the first psychological warning level and the second psychological warning level are the same, then the corresponding level judgment rule shall be applied directly. If the first and second psychological warning levels are different, the higher level shall be applied.

[0014] A parallel intelligent state assessment system based on high-standard verification data includes a psychological test data acquisition and processing module, a question-answering behavior anomaly detection module, a multiple anomaly screening module, a state mapping modeling and evaluation module, and an intelligent psychological state decision-making module. The psychological assessment data acquisition and processing module is used to collect the first answer information of students' original assessment questions and the psychological warning level confirmed by manual evaluation through the psychological assessment terminal, and to preprocess the first answer information to form the original answer information set; The question-answering behavior anomaly detection module is used to construct a joint judgment mechanism for multiple abnormal behaviors based on question-answering behavior characteristics, output abnormal question-answering risk scores in real time, identify abnormal question-answering behaviors of students, and mark and remove students with abnormal question-answering behaviors. The multiple anomaly filtering module is used to formulate multiple anomaly filtering strategies based on the original answer information set to obtain third answer information; The state mapping modeling and evaluation module is used to construct a random forest model to calculate the first importance score by taking the third answer information as input and the psychological warning level as the real label. By constructing multiple decision trees to synthesize the first importance score, it learns the nonlinear mapping relationship from the original psychological scale items to the student's psychological state. The intelligent psychological state decision-making module is used to output a second importance score based on the first answer information, determine the corresponding first psychological warning level based on the first importance score, and then determine the corresponding second psychological expectation level based on the second importance score. The student's psychological state is determined by a dual-path parallel intelligent psychological decision-making mechanism through the first psychological warning level and the second psychological warning level.

[0015] The technical effects and advantages of this invention, a parallel intelligent state assessment method and system based on high-standard verification data, are as follows: This invention effectively overcomes the reliance of traditional scale assessments on fixed thresholds and single-path judgments by introducing a parallel intelligent state assessment method based on high-standard verification data. By constructing multiple anomaly screening mechanisms and a highly reliable answer information set based on the original answer data, psychological assessment is no longer directly based on potentially biased original answers, but is instead prioritized on verified, denoised, high-quality data, fundamentally improving the accuracy and stability of psychological state assessment. By simultaneously constructing a first assessment path based on high-standard verification data and a second assessment path based on original answer information, a dual-path parallel analysis of students' psychological states is achieved. On the one hand, the first assessment path focuses on uncovering the nonlinear correlation between psychological scale items and psychological states, identifying long-term, implicit, or structural psychological risks, and compensating for the shortcomings of traditional linear scoring models in recognizing complex psychological patterns. On the other hand, the second assessment path can sensitively capture immediate abnormal changes in students during the assessment process, effectively addressing short-term biases caused by temporary emotional fluctuations or perfunctory answers, making the assessment results closer to the students' true psychological states.

[0016] This invention introduces a dual-path parallel intelligent psychological decision-making mechanism, combined with a psychological early warning level consistency judgment and a "highest priority" handling strategy. When assessment results are consistent, it can quickly output clear conclusions; when assessment results are inconsistent, it prioritizes a more conservative and safer risk assessment method. This significantly reduces the probability of missed diagnoses for students with serious psychological problems, improves the sensitivity and safety of psychological risk screening, and meets the practical need for "strictness over leniency" in campus mental health management regarding risk prevention and control. By introducing an importance scoring mechanism for psychological scale items, the psychological assessment process is transformed from a "black box result output" to an "interpretable assessment decision-making process." This provides psychological teachers, homeroom teachers, and management departments with clear evidence of risk sources, enhancing the pertinence and feasibility of subsequent psychological interventions and educational governance measures. Compared to existing psychological screening schemes that rely on static scale rules, this invention can achieve higher assessment credibility, lower misjudgment risk, and stronger management collaboration capabilities in real educational scenarios, demonstrating significant practical application value and promotional significance. Attached Figure Description

[0017] Figure 1 A flowchart of an intelligent state parallel evaluation method based on high-standard verification data provided in an embodiment of the present invention; Figure 2 This is a system block diagram of an intelligent state parallel evaluation system based on high-standard verification data, provided for an embodiment of the present invention. Detailed Implementation

[0018] The technical solutions of this invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described technical solutions are only a part of this invention, and not all of it. All other technical solutions obtained by those skilled in the art based on the technical solutions of this invention without inventive effort are within the scope of protection of this invention.

[0019] like Figure 1 The diagram shown is a flowchart of an intelligent state parallel evaluation method based on high-standard verification data provided by an embodiment of the present invention. Figure 1 The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S5 are detailed below: Step S1: Collect the first answer information of the student's original test questions and the psychological warning level confirmed by manual evaluation through the psychological assessment terminal. Preprocess the first answer information to form the original answer information set; the first answer information includes numerical information features, textual information features and answer behavior features; Step S2: Construct a joint judgment mechanism for multiple abnormal behaviors based on the characteristics of answering behavior, output abnormal answering risk scores in real time, and use them to identify students' abnormal answering behavior. Students with abnormal answering behavior are marked and removed. Step S3: Based on the original answer information set, formulate a multi-anomaly filtering strategy to obtain the third answer information; Step S4: Using the third answer information as input and the psychological warning level as the real label, a random forest model is constructed to calculate the first importance score. By constructing multiple decision trees to synthesize the first importance score, the nonlinear mapping relationship from the original psychological scale items to the student's psychological state is learned. Step S5: Output the second importance score based on the first answer information, determine the corresponding first psychological warning level based on the first importance score, and then determine the corresponding second psychological expectation level based on the second importance score. The student's psychological state is determined by executing a dual-path parallel intelligent psychological decision-making mechanism through the first psychological warning level and the second psychological warning level.

[0020] Preferably, the answering behavior features include the duration of answering a single question; the textual information features include the question text content, the set of question keywords, and the semantic vector representation of the question.

[0021] Preferably, a joint judgment mechanism for multiple abnormal behaviors is constructed based on the characteristics of answering behavior, and an abnormal answering risk score is output in real time to identify students' abnormal answering behavior. Students exhibiting abnormal answering behavior are marked and removed. The specific steps are as follows: During the assessment and answering process, the answering time for each question is recorded in real time to form a corresponding answering time sequence, which is then stored in association with the question number to construct a dataset of students' answering behavior characteristics. Based on the time series of responses, calculate the characteristic indicators of response time, including the mean response time for a single question and the standard deviation of the response time for a single question; Based on the answer duration characteristic index, construct an index for judging abnormal answering behavior; Using abnormal answering behavior indicators as input, a joint judgment mechanism for multiple abnormal behaviors is constructed to calculate the abnormal answering risk score in real time and quantify the degree of abnormal answering behavior of students during the answering process. The abnormal answer risk score calculated in real time is compared with the preset risk threshold. When the abnormal answer risk score is lower than the preset risk threshold, the student's answering behavior is determined to be normal; when the abnormal answer risk score is higher than or equal to the preset risk threshold, the student's answering behavior is determined to be abnormal. For students whose answering behavior is deemed abnormal, abnormal answering information is generated in their corresponding assessment data. The abnormal answering information includes an abnormality type identifier and risk score results. The assessment data of students with abnormal answering behavior are then removed.

[0022] In one embodiment of the present invention, during an online psychological assessment, students sequentially complete an assessment task containing multiple psychological scale questions through a psychological assessment terminal. After the student begins answering the questions, the process of answering each question is monitored in real time, and the time taken to answer each question from the presentation of the question to the submission of the answer is automatically recorded. The time sequence of the answers is then formed according to the question number and stored as a dataset of the student's answering behavior characteristics.

[0023] During the question-answering process, answer duration characteristics are calculated in real time based on the answer time series, including the average answer duration and standard deviation of the answer duration for each completed question. Statistical analysis revealed that Student A's answer duration for multiple consecutive questions was concentrated between 1 and 2 seconds, with a standard deviation close to 0. The overall answering rhythm was highly consistent, significantly deviating from the normal answering distribution characteristics of similar students.

[0024] Based on the above-mentioned answer duration characteristics, an abnormal random answer indicator is constructed to assess whether the rhythm of answer duration exhibits a mechanical consistency. Meanwhile, for the situation where Student B's answer duration for the two questions was 25 seconds and 1 second respectively, a sudden abnormality indicator is constructed based on the changes in adjacent answer durations to reflect the behavior of drastic changes in answer duration within a short period of time.

[0025] Subsequently, using indicators of arbitrary answering and sudden abnormalities as inputs, a multi-abnormal behavior joint judgment mechanism is implemented to comprehensively analyze students' current answering behavior and calculate abnormal answering risk scores in real time. For students with a highly consistent answering pace, their abnormal answering risk scores continuously increase with the progress of answering; for students with significant abrupt changes in answering time, their risk scores rise rapidly at the point of abnormality, thus dynamically reflecting the severity of abnormal answering behavior.

[0026] The abnormal answer risk score calculated in real time is compared with a preset risk threshold. When a student's abnormal answer risk score is lower than the risk threshold, their answering behavior is continuously judged to be in a normal state; when a student's abnormal answer risk score reaches or exceeds the risk threshold, the student is immediately judged to have abnormal answering behavior, and corresponding abnormal answer annotation information is generated in their assessment data. The annotation information includes an abnormality type identifier and abnormal answer risk score result.

[0027] After the assessment, based on the abnormal answer annotation information, the assessment data of students identified as exhibiting abnormal answering behavior were removed, preventing them from participating in subsequent psychological state modeling and assessment calculations. This method avoids interference from abnormal behaviors such as arbitrary or emotional answers on the psychological state assessment results, improving the authenticity and reliability of the psychological assessment data.

[0028] Preferably, the indicators for judging abnormal answering behavior include indicators for random answering and indicators for sudden abnormal answering.

[0029] The formula for calculating the abnormal index of random answers is: In the formula: To avoid arbitrary and abnormal indicators, The standard deviation of the time taken to answer a single question. This represents the average time taken to answer a single question.

[0030] The formula for calculating the sudden abnormality index is: In the formula: This is a sudden abnormal indicator. The number of questions Let be the time taken to answer the (i-1)th question. Let be the time taken to answer the i-th question. For indicator functions, This is the threshold for sudden changes.

[0031] In one embodiment of the present invention, during an online psychological assessment consisting of 40 psychological scale questions, the time taken to answer each question is recorded sequentially after the student begins answering the questions. And form a time sequence of answers according to the order of the questions. During the assessment, the average answering time for each question is calculated in real time based on the answering time series of the currently completed questions. and the standard deviation of the time spent answering a single question Based on this, an abnormal index for arbitrary responses is calculated. ,in .

[0032] In the actual assessment, student C's answer time for the first 20 questions was concentrated between 1.2 and 1.5 seconds, resulting in... The value is small and Keep it at a low level, making Approaching 0, thus making the random answering abnormal indicator If the value is close to 1, it indicates that the student's answering rhythm exhibits a clear mechanical consistency, which is consistent with the typical characteristics of random answering abnormal behavior.

[0033] Meanwhile, for the same answer time series, further calculations of sudden anomaly indicators were performed. Specifically, this included the proportion of change in answer time between two adjacent questions. Perform calculations one by one, and when the change ratio exceeds the preset sudden change threshold... At that time, the indicator function at the corresponding position will be... Set to 1. For a certain student, the time spent answering questions 15 and 16 suddenly dropped from 22 seconds to 2 seconds, making the change rate significantly greater than the threshold. This location is counted in the number of sudden anomalies.

[0034] The number of sudden anomalies in all adjacent questions is counted and divided by the total number of questions minus one to obtain the sudden anomaly index. When a student experiences significant abrupt changes in answering time in multiple adjacent questions, their sudden anomaly index value increases significantly, reflecting emotional fluctuations or irrational answering behavior during the answering process.

[0035] By simultaneously calculating both random and sudden abnormal indicators, student answering behavior can be characterized from two dimensions: "consistency of overall answering rhythm" and "abrupt changes in local answering duration." This provides a stable and quantifiable input basis for subsequent joint judgment mechanisms of multiple abnormal behaviors and effectively supports the real-time calculation of abnormal answering risk scores.

[0036] Preferably, the abnormal answering behavior judgment index is used as input to construct a multi-abnormal behavior joint judgment mechanism, and the abnormal answering risk score is calculated in real time. The multi-abnormal behavior joint judgment mechanism is as follows: Extract the random answering abnormal indicators and the sudden abnormal indicators, and compare the random answering abnormal indicators with the first abnormal indicator. If the random answering abnormal indicator is greater than or equal to the first abnormal indicator, it is determined that the student has abnormal answering behavior; if the random answering abnormal indicator is less than the first abnormal indicator, an abnormal answering risk assessment is triggered. The sudden abnormality indicator is compared with the first abnormality indicator. If the sudden abnormality indicator is greater than or equal to the first abnormality indicator, the student is judged to have abnormal answering behavior; if the sudden abnormality indicator is less than the first abnormality indicator, the abnormal answering risk assessment is triggered. The abnormal answer risk assessment is specifically conducted by weighting the abnormal indicators of random answering and sudden abnormality to obtain an abnormal answer risk score.

[0037] In one embodiment of the present invention, during an online psychological assessment consisting of 50 psychological scale questions, while the student is answering, the system continuously calculates both random answering abnormality indicators and sudden abnormality indicators, and uses these two types of abnormal indicators as input to a multi-abnormal behavior joint judgment mechanism. After the assessment begins, when the student completes the first few questions, the current random answering abnormality indicator value is obtained in real time and compared with a preset first abnormality indicator for judgment.

[0038] In actual operation, during the first 15 questions, student D's abnormal index of random answering continued to rise, and after answering the 16th question, it reached and exceeded the first abnormal index threshold. Based on this, it was directly determined that the student had abnormal answering behavior, and without entering the subsequent risk assessment calculation, abnormal answering label information was immediately generated for the student's assessment data, indicating that his answering behavior had obvious characteristics of random answering.

[0039] For another student, whose random answering anomaly index remained below the first anomaly threshold, no direct anomaly was determined. Instead, the anomaly risk assessment process was continued. At this point, the student's sudden anomaly index was further extracted and compared with the second anomaly index. When the sudden anomaly index rapidly increased during the assessment and reached the second anomaly threshold, the student was also directly determined to have exhibited abnormal answering behavior, and the corresponding anomaly type was recorded as a sudden anomaly.

[0040] In the assessment of some students, both the random answering and sudden abnormality indicators did not reach the corresponding abnormality judgment thresholds, but both indicators showed a certain degree of deviation. To address this, an abnormal answering risk assessment mechanism was activated. Based on a weighted calculation of the random answering and sudden abnormality indicators, an abnormal answering risk score was obtained. This risk score is used to comprehensively quantify the degree of potential abnormal answering behavior by students during the test.

[0041] As the quiz progresses, the risk score for abnormal answers is updated in real time. When the risk score gradually approaches or exceeds a preset risk threshold, the student's assessment data is marked as a high-risk abnormal answer state. When the risk score remains within a safe range, the student's answering behavior is continuously judged as normal. Through this multi-abnormal behavior joint judgment mechanism, rapid identification of obvious abnormal behavior and progressive risk assessment of latent abnormal behavior are achieved, effectively improving the accuracy and stability of abnormal answer identification.

[0042] Preferably, a multi-layered anomaly filtering strategy is developed based on the original answer information set to obtain the third answer information, specifically as follows: By performing the first round of anomaly screening on the numerical information features in the first answer information, the boundary compression of extreme values ​​is achieved, and the second answer information is output by integrating the textual information features and answer behavior features. The second round of filtering is performed by extracting numerical and textual information features from the second answer information. After identifying and eliminating similar types of questions, the third answer information is output.

[0043] Preferably, by performing a first-level anomaly screening on numerical information features, boundary compression of extreme values ​​is achieved. Specifically, this involves extracting numerical information features from the first answer information, performing tail-shrinking processing on each numerical information feature, and pulling extreme values ​​exceeding the 5% and 95% quantiles back to the quantile boundaries. The calculation formula is as follows: In the formula: This represents the i-th numerical information feature after tail reduction processing. For the i-th numerical information feature, The 5th percentile boundary. The 95th percentile boundary.

[0044] Preferably, the second answer information is filtered by extracting numerical and textual information features to identify and eliminate similar types of questions before outputting the third answer information. The specific steps are as follows: The text similarity between different questions is calculated based on textual information features. When the text similarity of a question is higher than a preset similarity threshold, the corresponding question is determined to be a candidate duplicate question. Extract the numerical information features of candidate duplicate questions, calculate the difference, and compare the difference with a preset difference. If the difference is greater than or equal to the preset difference, the candidate duplicate questions are not similar questions; if the difference is less than the preset difference, the candidate duplicate questions are similar questions. For questions of similar types, keep one of them and remove the rest.

[0045] In one embodiment of the present invention, during a mental health assessment conducted on first-year high school students, the students' initial response information was collected through a psychological assessment terminal. This initial response information included scores for each item on the psychological scale, the text content of the items, the set of keywords for each item, and the time taken to answer each item. After the assessment, the initial response information of all students was compiled into an original response information set. Based on this original response information set, a multiple anomaly screening strategy was initiated to obtain third response information for subsequent modeling and analysis.

[0046] In the first stage of anomaly screening, numerical information features are extracted from the initial answer data, such as question scores, scale scores corresponding to the questions, and answer time for each question. For each type of numerical information feature, its 5th percentile and 95th percentile boundaries in the entire student sample are calculated. When a student's numerical feature value for a particular question is lower than... When compressing this value to When the value is higher than When that happens, the value is compressed to... When the value falls between the two extremes, the original value remains unchanged. This tail-shrinking process pulls extreme outliers back within the statistical distribution boundary, thereby reducing the impact of individual extreme responses on the overall data distribution. After completing the boundary compression of numerical information features, the processed numerical information features are integrated with the corresponding textual information features and response behavior features to form the second response information.

[0047] In the second anomaly screening stage, textual and numerical information features are further extracted from the second set of answer information to identify and eliminate similar questions. First, based on the question text content and its semantic vector representation, the text similarity between questions from different psychological scales is calculated. When the text similarity between any two questions exceeds a preset similarity threshold, the two questions are marked as candidate duplicate questions. Subsequently, numerical information features of the candidate duplicate questions are further extracted, such as the average score or score distribution characteristics of the corresponding questions, and the numerical difference between the two questions is calculated.

[0048] When the numerical difference is greater than or equal to a preset difference, the candidate duplicate question is determined to have significant differences in the answer results and is not considered a similar question type, therefore it is retained. When the numerical difference is less than the preset difference, the candidate duplicate question is determined to be highly similar in both semantics and answer results, and is considered a similar question type. For questions determined to be similar, only one question is retained, and the remaining similar questions are removed from the dataset.

[0049] Through the collaborative processing of the first and second layers of anomaly screening, the third answer information is finally output. This third answer information has eliminated the interference of extreme outliers in its numerical distribution and avoided the repetitive influence of similar question types in its question structure, providing a highly reliable and stable input data foundation for subsequent psychological state modeling and importance analysis.

[0050] Specifically, using the third-response information as input and the psychological warning level as the true label, a random forest model is constructed to calculate the first importance score. By constructing multiple decision trees to synthesize the first importance score, the non-linear mapping relationship from the original psychological scale items to the student's psychological state is learned. The specific steps are as follows: After screening for abnormal answering behaviors, the third set of answering information obtained from the screening is used as input data for the model. This third set of answering information includes the student's answers to questions on each psychological scale and their corresponding numerical feature representations, and feature vectors are constructed on a student-by-student basis. Simultaneously, the psychological warning level corresponding to each of the third set of answering information is obtained as the ground truth label. The psychological warning level is used to characterize the student's current psychological state and serves as the training objective for the supervised learning model.

[0051] Based on the third-response information and the psychological warning level, a random forest model is constructed as a psychological state assessment model. The random forest model consists of multiple independent decision trees. During model training, a portion of training samples are randomly sampled from the third-response information for each decision tree, and a subset of psychological scale item features are randomly selected as the input feature subset of the decision tree to reduce the correlation between features and improve the model's generalization ability.

[0052] In the construction of each decision tree, the psychological warning level is used as the target variable for node splitting. Decision nodes are recursively generated using splitting criteria based on information gain, Gini index, or entropy, thus forming a mapping path from the characteristics of psychological scale items to the psychological warning level. In this way, a single decision tree can depict the discriminative patterns of different combinations of psychological scale items on students' psychological states.

[0053] After the decision tree training is completed, the decrease in impurity resulting from the participation of each psychological scale item feature in node splitting within the decision tree is statistically analyzed. This decrease in impurity is used as the importance contribution value of that feature in the current decision tree. By summing the importance contribution values ​​of the same psychological scale item feature in all splitting nodes within a single decision tree, the feature importance result of that feature in the single decision tree is obtained.

[0054] Furthermore, the feature importance results of all decision trees in the random forest model are comprehensively processed. The importance contribution values ​​of the same psychological scale item features in multiple decision trees are weighted and summed or normalized to obtain the first importance score of each psychological scale item. The first importance score is used to quantify the contribution of each original psychological scale item in the determination of the psychological warning level.

[0055] By integrating the prediction results of multiple decision trees, the random forest model can comprehensively judge the information from the third response, thereby learning the nonlinear mapping relationship between the original psychological scale items and the student's psychological state. Compared with a single linear model, this approach can effectively characterize the complex interaction relationships between different psychological scale items, improve the accuracy and stability of psychological state assessment, and provide a reliable basis for subsequent psychological early warning and intervention.

[0056] Preferably, an intelligent psychological decision-making mechanism that uses a dual-path parallel approach, employing a first psychological warning level and a second psychological warning level, is used to determine the student's psychological state. The specific steps are as follows: If the first psychological warning level and the second psychological warning level are the same, then the corresponding level judgment rule shall be applied directly. If the first and second psychological warning levels are different, the higher level shall apply. For example, if the first psychological warning level is Level 1 and the second psychological warning level is Level 2, the higher level shall be applied, and the situation shall be classified as a Level 1 psychological warning.

[0057] It should be noted that the psychological warning levels include Level 1, Level 2, and Level 3. Level 1 indicates the presence of serious psychological problems and should be reported directly to school leaders and the district education bureau. Level 2 indicates the presence of psychological problems and requires notification of the school counselor and homeroom teacher for joint confirmation and counseling. Level 3 indicates no problems.

[0058] In one embodiment of the present invention, during a routine student mental health assessment conducted in a middle school, a first psychological warning level is output based on high-standard verification data, and a second psychological warning level is output based on the original answer information. After the assessment is completed, the first and second psychological warning levels are used as parallel inputs through dual paths to activate an intelligent psychological decision-making mechanism for a final determination of the student's mental state.

[0059] In actual operation, for a certain student, both the first and second psychological warning levels were judged as level three psychological warnings, indicating that the student did not show obvious psychological risks under both the high-standard verification perspective and the original answer perspective. Based on the judgment rule of consistent results of the two paths, the final psychological state judgment result was directly output according to the level three psychological warning level, and the result was recorded as "no psychological problems", without triggering the subsequent manual intervention process.

[0060] For another student, the first psychological warning level was determined to be Level 2 under the high-standard verification data path, and the second psychological warning level was also determined to be Level 2 under the original answer information path. Since the psychological warning levels were consistent in both paths, the processing procedure was directly executed according to the judgment rules for Level 2 psychological warning, automatically sending warning information to the school counselor and the class teacher of the student's class, prompting further interviews and targeted psychological counseling for the student.

[0061] In another real-world case, a student was assessed as having a Level 1 psychological warning based on the most important rating, but a Level 2 warning based on the second most important rating. Because the warning levels differed between the two approaches, the higher level was applied, resulting in the student being classified as having a Level 1 psychological warning. A serious psychological problem alert was automatically generated, and the relevant information was reported to the school administration and the district education bureau according to the pre-defined procedures. Simultaneously, the student was marked as a high-priority individual.

[0062] The above-mentioned dual-path parallel intelligent psychological decision-making mechanism can ensure the stability of assessment results while avoiding the omission of high-risk students due to misjudgment by a single path. It ensures that in the event of disagreement, a more conservative and safer psychological early warning and handling strategy is prioritized, thereby improving the timeliness and reliability of campus psychological risk prevention and control.

[0063] A parallel intelligent state evaluation system based on high-standard verification data includes a psychological assessment data acquisition and processing module, a question-answering behavior anomaly detection module, a multiple anomaly screening module, a state mapping modeling and evaluation module, and an intelligent psychological state decision-making module. The psychological assessment data acquisition and processing module is connected to the question-answering behavior anomaly detection module, the question-answering behavior anomaly detection module is connected to the multiple anomaly screening module, the multiple anomaly screening module is connected to the state mapping modeling and evaluation module, and the state mapping modeling and evaluation module is connected to the intelligent psychological state decision-making module. The psychological assessment data acquisition and processing module is used to collect the first answer information of students' original assessment questions and the psychological warning level confirmed by manual evaluation through the psychological assessment terminal, and to preprocess the first answer information to form the original answer information set; The question-answering behavior anomaly detection module is used to construct a joint judgment mechanism for multiple abnormal behaviors based on question-answering behavior characteristics, output abnormal question-answering risk scores in real time, identify abnormal question-answering behaviors of students, and mark and remove students with abnormal question-answering behaviors. The multiple anomaly filtering module is used to formulate multiple anomaly filtering strategies based on the original answer information set to obtain third answer information; The state mapping modeling and evaluation module is used to construct a random forest model to calculate the first importance score by taking the third answer information as input and the psychological warning level as the real label. By constructing multiple decision trees to synthesize the first importance score, it learns the nonlinear mapping relationship from the original psychological scale items to the student's psychological state. The intelligent psychological state decision-making module is used to output a second importance score based on the first answer information, determine the corresponding first psychological warning level based on the first importance score, and then determine the corresponding second psychological expectation level based on the second importance score. The student's psychological state is determined by a dual-path parallel intelligent psychological decision-making mechanism through the first psychological warning level and the second psychological warning level.

[0064] like Figure 2 The diagram shown is a system block diagram of an intelligent state parallel evaluation system based on high-standard verification data according to an embodiment of the present invention, which can be used to execute... Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0065] Through the above embodiments, this invention, by introducing an intelligent parallel assessment method for psychological states based on high-standard verification data, effectively overcomes the reliance of traditional scale assessments on fixed thresholds and single-path judgments. By constructing multiple anomaly screening mechanisms and a highly reliable set of answer information based on the original answer data, psychological assessment is no longer directly based on potentially biased original answers, but is instead prioritized on verified, denoised, high-quality data, fundamentally improving the accuracy and stability of psychological state assessment. By simultaneously constructing a first assessment path based on high-standard verification data and a second assessment path based on original answer information, a dual-path parallel analysis of students' psychological states is achieved. On the one hand, the first assessment path focuses on uncovering the nonlinear correlation between psychological scale items and psychological states, identifying long-term, implicit, or structural psychological risks, and compensating for the shortcomings of traditional linear scoring models in recognizing complex psychological patterns. On the other hand, the second assessment path can sensitively capture immediate abnormal changes in students during the assessment process, effectively addressing short-term biases caused by temporary emotional fluctuations or perfunctory answers, making the assessment results closer to the students' true psychological states.

[0066] This invention introduces a dual-path parallel intelligent psychological decision-making mechanism, combined with a psychological early warning level consistency judgment and a "highest priority" handling strategy. When assessment results are consistent, it can quickly output clear conclusions; when assessment results are inconsistent, it prioritizes a more conservative and safer risk assessment method. This significantly reduces the probability of missed diagnoses for students with serious psychological problems, improves the sensitivity and safety of psychological risk screening, and meets the practical need for "strictness over leniency" in campus mental health management regarding risk prevention and control. By introducing an importance scoring mechanism for psychological scale items, the psychological assessment process is transformed from a "black box result output" to an "interpretable assessment decision-making process." This provides psychological teachers, homeroom teachers, and management departments with clear evidence of risk sources, enhancing the pertinence and feasibility of subsequent psychological interventions and educational governance measures. Compared to existing psychological screening schemes that rely on static scale rules, this invention can achieve higher assessment credibility, lower misjudgment risk, and stronger management collaboration capabilities in real educational scenarios, demonstrating significant practical application value and promotional significance.

[0067] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

[0068] Finally: The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A method for intelligent parallel state evaluation based on high-standard verification data, characterized in that, The specific steps are as follows: The system collects students' initial test responses and psychological warning levels confirmed by manual evaluation through a psychological testing terminal. The initial response information is preprocessed to form an initial response information set. The initial response information includes numerical information features, textual information features, and response behavior features. A joint judgment mechanism for multiple abnormal behaviors is constructed by identifying the characteristics of answering behavior, and an abnormal answering risk score is output in real time to identify students' abnormal answering behavior. Students with abnormal answering behavior are marked and removed. A multi-layered anomaly screening strategy was developed based on the original question answer information set to obtain third-party question answer information; Using the third-party answer information as input and the psychological warning level as the real label, a random forest model is constructed to calculate the first importance score. By constructing multiple decision trees to synthesize the first importance score, the non-linear mapping relationship from the original psychological scale items to the student's psychological state is learned. Based on the first answer information, a second importance score is output. The first psychological warning level is determined based on the first importance score, and the second psychological expectation level is determined based on the second importance score. The student's psychological state is determined by a dual-path parallel intelligent psychological decision-making mechanism that uses the first and second psychological warning levels.

2. The intelligent state parallel evaluation method based on high-standard verification data according to claim 1, characterized in that, Answering behavior characteristics include the duration of answering a single question; textual information characteristics include the question text content, the set of question keywords, and the semantic vector representation of the question.

3. The intelligent state parallel evaluation method based on high-standard verification data according to claim 1, characterized in that, A joint judgment mechanism for multiple abnormal behaviors is constructed based on the characteristics of answering behavior. An abnormal answering risk score is output in real time to identify students' abnormal answering behavior. Students exhibiting abnormal answering behavior are marked and removed. The specific steps are as follows: During the assessment and answering process, the answering time for each question is recorded in real time to form a corresponding answering time sequence, which is then stored in association with the question number to construct a dataset of students' answering behavior characteristics. Based on the time series of responses, calculate the response duration characteristic indicators, including the mean response time for a single question and the standard deviation of the response time for a single question; Based on the answer duration characteristic index, construct an index for judging abnormal answering behavior; Using abnormal answering behavior indicators as input, a joint judgment mechanism for multiple abnormal behaviors is constructed to calculate the abnormal answering risk score in real time and quantify the degree of abnormal answering behavior of students during the answering process. The abnormal answer risk score calculated in real time is compared with the preset risk threshold. When the abnormal answer risk score is lower than the preset risk threshold, the student's answer behavior is determined to be normal. When the risk score for abnormal answering is higher than or equal to the preset risk threshold, the student is deemed to have abnormal answering behavior. For students whose answering behavior is deemed abnormal, abnormal answering information is generated in their corresponding assessment data. The abnormal answering information includes an abnormality type identifier and risk score results. The assessment data of students with abnormal answering behavior are then removed.

4. The intelligent state parallel evaluation method based on high-standard verification data according to claim 3, characterized in that, The indicators for judging abnormal answering behavior include indicators for random answering and indicators for sudden abnormal answering.

5. The intelligent state parallel evaluation method based on high-standard verification data according to claim 3, characterized in that, Using abnormal answering behavior indicators as input, a joint judgment mechanism for multiple abnormal behaviors is constructed to calculate the risk score of abnormal answering behavior in real time. The specific details of the joint judgment mechanism for multiple abnormal behaviors are as follows: Extract the random answering abnormal indicators and the sudden abnormal indicators, compare the random answering abnormal indicators with the first abnormal indicator, and if the random answering abnormal indicator is greater than or equal to the first abnormal indicator, it is determined that the student has abnormal answering behavior. If the abnormal indicator for random answers is less than the first abnormal indicator, an abnormal answer risk assessment will be triggered. The sudden abnormality indicator is compared with the first abnormality indicator. If the sudden abnormality indicator is greater than or equal to the first abnormality indicator, the student is judged to have abnormal answering behavior; if the sudden abnormality indicator is less than the first abnormality indicator, the abnormal answering risk assessment is triggered. The abnormal answer risk assessment is specifically conducted by weighting the abnormal indicators of random answering and sudden abnormality to obtain an abnormal answer risk score.

6. The intelligent state parallel evaluation method based on high-standard verification data according to claim 1, characterized in that, A multi-layered anomaly filtering strategy is developed based on the original answer information set to obtain third-party answer information, specifically: By performing the first round of anomaly screening on the numerical information features in the first answer information, the boundary compression of extreme values ​​is achieved, and the second answer information is output by integrating the textual information features and answer behavior features. The second round of filtering is performed by extracting numerical and textual information features from the second answer information. After identifying and eliminating similar types of questions, the third answer information is output.

7. The intelligent state parallel evaluation method based on high-standard verification data according to claim 6, characterized in that, By performing the first round of anomaly screening on numerical information features, the boundary compression of extreme values ​​is achieved. Specifically, the numerical information features in the first answer information are extracted, and each numerical information feature is processed to reduce its tail, pulling the extreme values ​​that exceed the 5% and 95% quantiles back to the quantile boundaries.

8. The intelligent state parallel evaluation method based on high-standard verification data according to claim 6, characterized in that, The second round of filtering involves extracting numerical and textual information features from the second answer information to identify and eliminate similar types of questions before outputting the third answer information. The specific steps are as follows: The text similarity between different questions is calculated based on textual information features. When the text similarity of a question is higher than a preset similarity threshold, the corresponding question is determined to be a candidate duplicate question. Extract the numerical information features of candidate duplicate questions, calculate the difference, compare the difference with the preset difference, and if the difference is greater than or equal to the preset difference, the candidate duplicate questions are not similar types of questions. If the difference is less than the preset difference, the candidate duplicate question is a similar type question; For questions of similar types, keep one of them and remove the rest.

9. The intelligent state parallel evaluation method based on high-standard verification data according to claim 1, characterized in that, The student's psychological state is determined through a dual-path parallel intelligent psychological decision-making mechanism that uses a first and second psychological warning level. The specific steps are as follows: If the first psychological warning level and the second psychological warning level are the same, then the corresponding level judgment rule shall be applied directly. If the first and second psychological warning levels are different, the higher level shall be applied.

10. A parallel intelligent state evaluation system based on high-standard verification data, applied to the parallel intelligent state evaluation method based on high-standard verification data as described in any one of claims 1-9, characterized in that, It includes a psychological assessment data collection and processing module, a question-answering behavior anomaly detection module, a multiple anomaly screening module, a state mapping modeling and evaluation module, and an intelligent psychological state decision-making module; The psychological assessment data acquisition and processing module is used to collect the first answer information of students' original assessment questions and the psychological warning level confirmed by manual evaluation through the psychological assessment terminal, and to preprocess the first answer information to form the original answer information set; The question-answering behavior anomaly detection module is used to construct a joint judgment mechanism for multiple abnormal behaviors based on question-answering behavior characteristics, output abnormal question-answering risk scores in real time, identify abnormal question-answering behaviors of students, and mark and remove students with abnormal question-answering behaviors. The multiple anomaly filtering module is used to formulate multiple anomaly filtering strategies based on the original answer information set to obtain third answer information; The state mapping modeling and evaluation module is used to construct a random forest model to calculate the first importance score by taking the third answer information as input and the psychological warning level as the real label. By constructing multiple decision trees to synthesize the first importance score, it learns the nonlinear mapping relationship from the original psychological scale items to the student's psychological state. The intelligent psychological state decision-making module is used to output a second importance score based on the first answer information, determine the corresponding first psychological warning level based on the first importance score, and then determine the corresponding second psychological expectation level based on the second importance score. The student's psychological state is determined by a dual-path parallel intelligent psychological decision-making mechanism through the first psychological warning level and the second psychological warning level.

Images