A Method and System for Continuous Analysis of Psychological States Based on Multidimensional Behavioral Perception

Abstract

The present application relates to the technical field of behavior feature recognition, in particular to a psychological state continuous analysis method and system based on multi-dimensional behavior perception. The present application converts multi-dimensional behavior data of psychological services into user behavior feature sequences arranged according to continuous sampling time periods, forms behavior deviation values of each behavior feature according to personal behavior baselines, and calculates conditional transfer entropy of different behavior features under different transfer directions and different time lags in a sliding time window, thereby realizing continuous recognition of behavior change sequence and corresponding time interval, and solving the problem that traditional analysis results are difficult to distinguish different psychological behavior change propagation processes. On this basis, by comparing conditional transfer entropy of opposite transfer directions and establishing directed edges according to continuously established transfer directions, short-term common changes do not directly constitute behavior propagation relationships.

Description

Technical Field

[0001] This invention relates to the field of behavioral feature recognition technology, and in particular to a method and system for continuous analysis of psychological states based on multidimensional behavioral perception. Background Technology

[0002] The continuous analysis of psychological states involved in the method of continuous analysis of psychological states refers to the continuous acquisition of information related to an individual's psychological state over time, and the identification, recording and correlation analysis of the state performance at different times, in order to describe the individual's psychological state and its changing process.

[0003] Traditional psychological state analysis typically begins by collecting user psychological assessment results or behavioral statistics from psychological service platforms. It then normalizes, weights, or calculates correlations among the data points, finally determining the user's psychological state based on the overall score or common changes between indicators. While traditional psychological state analysis can reflect a user's overall behavioral level within a specific timeframe, it suffers from several drawbacks. When users continuously undergo psychological assessments, online psychological counseling, emotion recording, and psychological training, different psychological service behaviors often change gradually in a chronological order. Directly merging multiple types of behavioral data into a single score or calculating correlations within the same timeframe compresses the sequence and intervals of behavioral changes into a single analysis result. This makes it difficult to distinguish whether one behavioral change precedes another, and further reveals the propagation of continuous behavioral changes. For example, a change in the browsing category ratio leading to a decrease in consultation interaction time, or a decrease in the training task completion ratio followed by an increase in consultation response intervals, may produce similar overall scores. However, the starting points and directions of the corresponding psychological behavioral changes differ, making it difficult for traditional analysis to differentiate between these processes. Summary of the Invention

[0004] The purpose of this invention is to address the shortcomings of existing technologies by proposing a method and system for continuous analysis of psychological states based on multidimensional behavioral perception.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a continuous analysis method for psychological states based on multidimensional behavioral perception, comprising the following steps:

[0006] Constructing user behavior feature sequences based on multidimensional behavioral data of users' psychological services;

[0007] Calculate the behavioral deviation value corresponding to each type of behavioral feature in the user behavioral feature sequence, and establish an individual baseline deviation sequence;

[0008] The sliding time window is divided based on the individual baseline deviation sequence. Within the sliding time window, the conditional transfer entropy sequence and the transfer lag sequence between different behavioral features are determined. The conditional transfer entropy represents the predictive power of a change in one behavioral feature on the subsequent change in another behavioral feature, and the transfer lag represents the time interval between corresponding changes in two behavioral features.

[0009] Each type of behavior feature in the user behavior feature sequence is taken as a behavior node. The directed edges between behavior nodes are determined based on the conditional transfer entropy sequence and the transfer delay sequence. A directed graph of behavior change is constructed based on the directed edges.

[0010] The user's psychological state can be determined by referring to the directed graph of behavioral changes.

[0011] As a further aspect of the present invention, the multidimensional behavioral data of psychological services specifically refers to the behavioral records collected by users when they log in to the psychological service platform and browse the psychological service function page, based on a continuous sampling period. The continuous sampling period is a series of consecutive time periods divided according to a preset sampling cycle.

[0012] As a further aspect of the present invention, the psychological service function pages include a psychological assessment page, a psychological counseling page, an emotion recording page, and a psychological training page; the behavior records include access frequency, access interval, cumulative counseling interaction time, average counseling response interval, training task completion rate, and browsing category ratio; wherein, the access frequency and access interval are obtained based on the page access records of each psychological service function page, the cumulative counseling interaction time and average counseling response interval are obtained based on the counseling session records and message interaction records of the psychological counseling page, the training task completion rate is obtained based on the training task records of the psychological training page, and the browsing category ratio is obtained based on the page browsing records of each psychological service function page.

[0013] As a further aspect of the present invention, the user behavior feature sequence includes access frequency change features, access interval change features, consultation duration change features, response interval change features, training completion change features, and browsing ratio change features.

[0014] As a further aspect of the present invention, the access frequency variation feature is the difference in the user's access frequency within adjacent sampling periods; the access interval variation feature is the difference in the user's access interval within adjacent sampling periods; the consultation duration variation feature is the difference in the cumulative consultation interaction duration within adjacent sampling periods; the response interval variation feature is the difference in the average consultation response interval within adjacent sampling periods; the training completion variation feature is the difference in the proportion of training tasks completed by the user within adjacent sampling periods; and the browsing ratio variation feature is the value obtained by dividing the sum of the absolute values ​​of the differences in the proportions of each browsing category within adjacent sampling periods by two.

[0015] As a further aspect of the present invention, the step of calculating the behavioral deviation value corresponding to each type of behavioral feature in the user behavioral feature sequence and establishing a personal baseline deviation sequence includes:

[0016] Set a baseline duration, and select from the user behavior feature sequence the access frequency change feature, access interval change feature, consultation duration change feature, response interval change feature, training completion change feature and browsing ratio change feature corresponding to the cumulative duration of the continuous sampling period reaching the baseline duration. The range of continuous sampling periods corresponding to each selected feature is taken as the baseline time period.

[0017] The median and median absolute deviation of each feature within the baseline time period are calculated to form a personal behavior baseline. The difference between each feature and the corresponding median in the personal behavior baseline is calculated for each consecutive sampling period. When the corresponding median absolute deviation is not zero, the difference is divided by the corresponding median absolute deviation to obtain the behavior deviation value of the corresponding feature. When the corresponding median absolute deviation is zero, the difference is used as the behavior deviation value of the corresponding feature. The behavior deviation values ​​corresponding to each feature are arranged in the order of the consecutive sampling periods to obtain the personal baseline deviation sequence.

[0018] As a further aspect of the present invention, the step of dividing a sliding time window based on the individual baseline deviation sequence and determining the conditional transfer entropy sequence and transfer delay sequence between different behavioral characteristics within the sliding time window includes:

[0019] Set the sliding time window, sliding step size, and maximum transmission duration. Generate a time lag set from one preset sampling period to the maximum transmission duration according to integer multiples of the preset sampling period. Based on the user's personal baseline deviation sequence, divide the continuous window according to the sliding time window and sliding step size, and calculate the ratio of each time lag in the time lag set to the preset sampling period to obtain the number of lag sampling intervals.

[0020] The browsing ratio change feature and consultation duration change feature, consultation duration change feature and response interval change feature, and response interval change feature and training completion change feature in the user behavior feature sequence are respectively formed into behavior feature pairs, and the first and second transmission directions of each behavior feature pair are determined respectively; for each transmission direction, the behavior feature corresponding to the starting point of the transmission direction is taken as the transmission start feature, and the behavior feature corresponding to the ending point of the transmission direction is taken as the transmission arrival feature; within the continuous window, continuous sampling periods are sequentially selected as the benchmark continuous sampling periods, and the behavior deviation values ​​corresponding to the transmission start feature in each benchmark continuous sampling period are extracted from the personal baseline deviation sequence to form the starting feature pre-sequence deviation sequence, and the transmission arrival features are extracted in each benchmark continuous sampling period. The behavioral deviation values ​​corresponding to the quasi-continuous sampling periods form an arrival feature pre-sequence deviation sequence; the continuous sampling periods separated from each benchmark continuous sampling period by the number of lag sampling intervals are determined as lag continuous sampling periods, and the behavioral deviation values ​​corresponding to the arrival features in each lag continuous sampling period are extracted to form an arrival feature subsequent deviation sequence; wherein, the first transmission direction includes a transmission direction starting with the browsing ratio change feature and ending with the consultation duration change feature, a transmission direction starting with the consultation duration change feature and ending with the response interval change feature, and a transmission direction starting with the response interval change feature and ending with the training completion change feature; the second transmission direction is the transmission direction opposite to the first transmission direction for the corresponding behavioral feature pair;

[0021] Based on the preceding deviation sequence of the initial feature, the preceding deviation sequence of the arrival feature, and the subsequent deviation sequence of the arrival feature corresponding to each behavioral feature pair in the first and second transmission directions, the behavioral deviation values ​​corresponding to the access frequency change feature and access interval change feature in the personal baseline deviation sequence during each continuous sampling period are used as conditional information. The conditional transfer entropy algorithm is used to calculate the conditional transfer entropy of each behavioral feature pair under different transmission directions and different time lags. Within the same continuous window, the conditional transfer entropy with the largest value is determined as the maximum conditional transfer entropy of the corresponding behavioral feature pair and the corresponding transmission direction, and the time lag corresponding to the maximum conditional transfer entropy is determined as the transmission lag. According to the order of the continuous window, the maximum conditional transfer entropy and transmission lag of each behavioral feature pair in the first and second transmission directions are arranged to form the conditional transfer entropy sequence and the transmission lag sequence, respectively.

[0022] As a further aspect of the present invention, the directed graph of behavior change includes:

[0023] The user behavior feature sequence includes the access frequency change feature, access interval change feature, consultation duration change feature, response interval change feature, training completion change feature, and browsing ratio change feature, which are respectively used as behavior nodes. A continuous decision window number is set, and the maximum conditional transfer entropy of each behavior feature is compared with the conditional transfer entropy sequence and the transfer delay sequence. When the maximum conditional transfer entropy of a transfer direction is greater than the maximum conditional transfer entropy of the opposite transfer direction in adjacent consecutive windows that reach the continuous decision window number, and the corresponding transfer delay does not exceed the maximum transfer duration, the corresponding transfer direction is determined to meet the comparison condition.

[0024] The behavior node corresponding to the starting point of the propagation direction that satisfies the comparison condition is taken as the starting node of the directed edge, and the behavior node corresponding to the ending point of the propagation direction is taken as the ending node of the directed edge. A directed edge is established from the starting node of the directed edge to the ending node of the directed edge. A directed graph of behavior change is constructed based on the behavior node and the directed edge, and the directed edges that connect the beginning and end of the directed graph of behavior change form the behavior change propagation path.

[0025] As a further aspect of the present invention, the determination of the user's psychological state by referring to the directed graph of behavioral changes includes:

[0026] The behavioral nodes in the directed graph of behavioral changes that have no other directed edges pointing to them and are connected outward by directed edges are identified as the sources of user psychological state changes; the behavioral nodes reached by the propagation path of the behavioral changes corresponding to the latest continuous window are identified as the stages of user psychological state changes; the number of different behavioral nodes traversed by the propagation path of behavioral changes is identified as the scope of influence of user psychological state; the trend of user psychological state changes is determined based on the direction of change of each behavioral deviation value in the personal baseline deviation sequence between adjacent continuous sampling periods; and the user psychological state is represented by combining the sources of user psychological state changes, stages of user psychological state changes, scope of influence of user psychological state changes, trend of user psychological state changes, and propagation path of behavioral changes.

[0027] A continuous analysis system for psychological states based on multidimensional behavioral perception, comprising:

[0028] The feature construction module constructs user behavior feature sequences based on users' multidimensional psychological service behavior data;

[0029] The deviation calculation module calculates the behavioral deviation value corresponding to each type of behavioral feature in the user's behavioral feature sequence and establishes a personal baseline deviation sequence.

[0030] The transfer analysis module divides the sliding time window based on the individual baseline deviation sequence, and determines the conditional transfer entropy sequence and transfer lag sequence between different behavioral characteristics within the sliding time window; whereby, the conditional transfer entropy represents the degree to which a change in one behavioral characteristic predicts the subsequent change in another behavioral characteristic, and the transfer lag represents the time interval between corresponding changes in two behavioral characteristics.

[0031] The graph construction module takes each type of behavior feature in the user behavior feature sequence as a behavior node, determines the directed edges between behavior nodes based on the conditional propagation entropy sequence and the propagation delay sequence, and constructs a directed graph of behavior changes based on the directed edges.

[0032] The state determination module determines the user's psychological state by referring to the directed graph of behavioral changes.

[0033] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0034] This invention transforms multidimensional behavioral data from psychological services into a sequence of user behavioral features arranged according to continuous sampling periods. Based on individual behavioral baselines, it generates behavioral deviation values ​​for each feature and calculates the conditional propagation entropy of different behavioral features under different transmission directions and time lags within a sliding time window. This enables continuous identification of the sequence and corresponding time intervals of behavioral changes, solving the problem that traditional analysis results struggle to distinguish the propagation processes of different psychological and behavioral changes. Furthermore, by comparing the conditional propagation entropy in opposite transmission directions and establishing directed edges based on consistently valid transmission directions, short-term common changes do not directly constitute behavioral propagation relationships. By connecting these directed edges into a directed graph of behavioral changes, the source, current stage, scope of influence, and propagation path of user psychological state changes can be further determined. This transforms the single-score psychological state results into a process analysis result that describes the starting point, propagation direction, propagation time, and current location of psychological and behavioral changes. Attached Figure Description

[0035] Figure 1 This is the overall flowchart of the continuous analysis method for psychological states of the present invention;

[0036] Figure 2 A flowchart for establishing an individual baseline deviation sequence for this invention;

[0037] Figure 3 This is a flowchart for determining the conditional transfer entropy sequence and the transfer delay sequence in this invention;

[0038] Figure 4 A flowchart for constructing a directed graph of behavioral changes for this invention;

[0039] Figure 5 This is a flowchart for determining the user's psychological state in this invention. Detailed Implementation

[0040] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0041] Please see Figure 1-5 This invention provides a technical solution, a method for continuous analysis of psychological states based on multidimensional behavioral perception, comprising the following steps:

[0042] S1: Construct a user behavior feature sequence based on multidimensional behavioral data of users' psychological services;

[0043] Users log in to the psychological service platform via mobile phone, tablet, or computer during continuous psychological assessments, online psychological counseling, emotion recording, and psychological training. Before data collection begins, a preset sampling period is entered and converted into a fixed time unit. Starting from the selected start time, a time boundary is set after each preset sampling period, forming a continuous sampling period between two adjacent time boundaries. The end time of the previous continuous sampling period coincides with the start time of the next continuous sampling period. When a user opens the psychological assessment page, psychological counseling page, emotion recording page, or psychological training page, the user identifier, psychological service function page category, page opening time, and page closing time are recorded. When a user conducts a consultation on the psychological counseling page, the consultation session identifier, consultation session start time, consultation session end time, user message sending time, and consultation response time are recorded. When a user accepts a training task on the psychological training page, the training task identifier and training task completion status are recorded. Based on the occurrence time of the behavior record, behavior records located between adjacent time boundaries in the same group are grouped into the same continuous sampling period. Page access records represent the access history of users entering the psychological service function pages. Access frequency = number of page access records for psychological assessment pages + number of page access records for psychological counseling pages + number of page access records for emotion recording pages + number of page access records for psychological training pages. Page access records within the same continuous sampling period are arranged from earliest to latest page opening time. The interval between adjacent page accesses = page opening time of the next page access record minus page opening time of the previous page access record. The total interval between accesses is the sum of all adjacent page access intervals ÷ number of adjacent page access intervals. Consultation session records represent the time record of a psychological counseling session from start to finish. Consultation session start and end times are corresponding to the same consultation session identifier. The duration of a single consultation interaction = consultation session end time - consultation session start time. When a consultation session spans adjacent continuous sampling periods, the actual duration of the consultation session within each continuous sampling period is obtained separately. The cumulative consultation interaction duration = sum of all consultation interaction durations within the same continuous sampling period. The message interaction log represents the time correspondence between user messages and consultation responses. Based on the same consultation session identifier and message occurrence order, each user message is matched with the first consultation response that follows it. The single consultation response interval = consultation response time - user message sending time. The average consultation response interval = the sum of all single consultation response intervals ÷ the number of user messages corresponding to the completed responses. The training task log represents the records of users receiving and completing training tasks on the psychological training page. The training task completion rate = the number of training tasks corresponding to the completion status ÷ the number of training tasks within the same continuous sampling period.Page browsing history represents the category records formed by users browsing different psychological service function pages. The total number of page browsing history = the number of page browsing history for psychological assessment pages + the number of page browsing history for psychological counseling pages + the number of page browsing history for emotion recording pages + the number of page browsing history for psychological training pages. The proportion of a single browsing category = the number of page browsing history for a single psychological service function page ÷ the total number of page browsing history. Finally, multidimensional behavioral data of psychological services corresponding to each continuous sampling period are formed.

[0044] The multidimensional behavioral data of psychological services are arranged from earliest to latest according to the start time of the continuous sampling period. The values ​​of similar behaviors in two adjacent continuous sampling periods are matched one-to-one according to their time position. Starting from the second continuous sampling period, the behavioral changes in the current continuous sampling period relative to the previous continuous sampling period are obtained. Access frequency change feature = Access frequency of the current continuous sampling period - Access frequency of the previous continuous sampling period. Access frequency change feature represents the change in the number of visits to the psychological service function page within adjacent continuous sampling periods. Access interval change feature = Access interval of the current continuous sampling period - Access interval of the previous continuous sampling period. Access interval change feature represents the change in the time interval between page visits within adjacent continuous sampling periods. Consultation duration change feature = Cumulative consultation interaction duration of the current continuous sampling period - Cumulative consultation interaction duration of the previous continuous sampling period. Consultation duration change feature represents the change in the total duration of psychological consultation interaction within adjacent continuous sampling periods. Response interval change feature = Average consultation response interval of the current continuous sampling period - Average consultation response interval of the previous continuous sampling period. Response interval change feature represents the change in the time interval between consultation responses within adjacent continuous sampling periods. The training completion change characteristic = the proportion of training tasks completed in the current continuous sampling period - the proportion of training tasks completed in the previous continuous sampling period. This characteristic represents the change in the proportion of training tasks completed within adjacent continuous sampling periods. It corresponds to the browsing category proportions of the psychological assessment page, psychological counseling page, emotion recording page, and psychological training page within adjacent continuous sampling periods. The difference in the proportion of a single browsing category = the proportion of a single browsing category in the current continuous sampling period - the proportion of a single browsing category in the previous continuous sampling period. The browsing proportion change characteristic = (absolute value of the difference in the browsing category proportions of the psychological assessment page + absolute value of the difference in the browsing category proportions of the psychological counseling page + absolute value of the difference in the browsing category proportions of the emotion recording page + absolute value of the difference in the browsing category proportions of the psychological training page) ÷ 2. The browsing proportion change characteristic uses the absolute values ​​of the differences in the proportions of each browsing category for accumulation, without distinguishing between increases or decreases in the proportion of a single browsing category, only retaining the values ​​showing changes in the browsing composition among the four types of psychological service function pages. The characteristics of changes in access frequency, access interval, consultation duration, response interval, training completion, and browsing ratio corresponding to the same continuous sampling period are placed at the same time position and arranged in the order of the continuous sampling periods to form a user behavior feature sequence.

[0045] S2: Calculate the behavioral deviation value corresponding to each type of behavioral feature in the user behavioral feature sequence, and establish an individual baseline deviation sequence, including:

[0046] After the user behavior feature sequence is formed, the baseline duration is recorded and converted into a time unit with the same preset sampling period. The baseline duration represents the continuous time length used to describe the initial change level of a user's behavior. Starting from the earliest continuous sampling period in the user behavior feature sequence, the following features are selected sequentially according to their order: access frequency change feature, access interval change feature, consultation duration change feature, response interval change feature, training completion change feature, and browsing ratio change feature. The duration of the selected continuous sampling periods is accumulated simultaneously. The accumulated duration = number of selected continuous sampling periods × preset sampling period. If the accumulated duration is less than the baseline duration, the next continuous sampling period is selected. When the accumulated duration first reaches the baseline duration, selection stops, and the time from the start time of the first selected continuous sampling period to the end time of the last selected continuous sampling period is taken as the baseline time period. The values ​​of the six behavioral features within the baseline time period are arranged in ascending order. When the number of behavioral feature values ​​is odd, the median = (number of behavioral feature values ​​+ 1) ÷ 2 after sorting; when the number of behavioral feature values ​​is even, the median = (number of behavioral feature values ​​÷ 2 after sorting + (number of behavioral feature values ​​÷ 2 + 1) after sorting) ÷ 2. The median represents the level of behavioral change of the same behavioral feature in the middle of the sorted range within the baseline time period. For each behavioral feature value within the baseline time period, the absolute deviation relative to the corresponding median is calculated: Absolute deviation = |behavioral feature value - corresponding median|. All absolute deviations for the same behavioral feature are then arranged in ascending order. When the number of absolute deviations is odd, the median absolute deviation = (number of absolute deviations + 1) ÷ 2 after sorting; when the number of absolute deviations is even, the median absolute deviation = (number of absolute deviations ÷ 2 after sorting + (number of absolute deviations ÷ 2 + 1) after sorting) ÷ 2. Median absolute deviation represents the deviation of the same behavioral characteristic from the corresponding median during the baseline time period, ultimately forming an individual behavioral baseline composed of the median and median absolute deviation corresponding to the six behavioral characteristics.

[0047] After the personal behavior baseline is established, the characteristics of visit frequency change, visit interval change, consultation duration change, response interval change, training completion change, and browsing ratio change for each consecutive sampling period are read in chronological order. The corresponding median and median absolute deviation from the personal behavior baseline are then read according to the behavioral feature name. The behavioral feature difference is calculated as: Behavioral feature value for the current consecutive sampling period minus the median of the corresponding behavioral feature in the personal behavior baseline. The behavioral feature difference indicates the direction and magnitude of the deviation of the behavioral change in the current consecutive sampling period from the median level of the personal behavior baseline. When the median absolute deviation of the corresponding behavioral feature in the personal behavior baseline is not 0, the behavioral deviation value is calculated as: Behavioral feature difference ÷ Median absolute deviation of the corresponding behavioral feature in the personal behavior baseline. When the median absolute deviation of the corresponding behavioral feature in the personal behavior baseline is 0, the behavioral deviation value is calculated as: Behavioral feature difference. A behavioral deviation value greater than 0 indicates that the current behavioral feature value is higher than the corresponding median in the personal behavior baseline; a behavioral deviation value less than 0 indicates that the current behavioral feature value is lower than the corresponding median in the personal behavior baseline; and a behavioral deviation value equal to 0 indicates that the current behavioral feature value is consistent with the corresponding median in the personal behavior baseline. The behavioral deviation values ​​corresponding to the six behavioral characteristics within the same continuous sampling period are placed at the same time position and then arranged in chronological order along the continuous sampling period to form an individual baseline deviation sequence.

[0048] S3: Divide the sliding time window based on the individual baseline deviation sequence, and determine the conditional transfer entropy sequence and transfer delay sequence between different behavioral characteristics within the sliding time window, including:

[0049] After the individual baseline deviation sequence is formed, the sliding time window, sliding step size, and maximum transmission duration are recorded, and these are converted to the same time unit. The sliding time window represents the continuous time range for each joint reading of the individual baseline deviation sequence; the sliding step size represents the shift time between the starting positions of adjacent continuous windows; and the maximum transmission duration represents the longest allowed time interval between a change in one behavioral characteristic and a subsequent change in another. The earliest continuous sampling period in the individual baseline deviation sequence is used as the starting position of the first continuous window. Continuous sampling periods covered by the sliding time window are selected from the starting position of the first continuous window. After the first continuous window is selected, the starting position of the continuous window is shifted backward by the number of continuous sampling periods corresponding to the sliding step size. Then, continuous sampling periods covered by the sliding time window are selected again from the shifted starting position, and subsequent continuous windows are formed in the same manner. A preset sampling period is used as the minimum time lag in the time lag set. The preset sampling period is increased by one each time. The increase continues as long as the current time lag is less than the maximum transmission duration, and stops when the current time lag reaches the maximum transmission duration. The number of lag sampling intervals = time lag ÷ preset sampling period. The number of lag sampling intervals represents the number of consecutive sampling periods spanned by a time lag. Continuous windows are arranged from earliest to latest according to their start time, and the time lag sets are arranged from smallest to largest according to their time lag, ultimately forming the continuous windows and the number of lag sampling intervals corresponding to each time lag.

[0050] Within each continuous window, the browsing ratio change feature and the consultation duration change feature are considered as one behavioral feature pair, the consultation duration change feature and the response interval change feature are considered as another behavioral feature pair, and the response interval change feature and the training completion change feature are considered as yet another behavioral feature pair. A behavioral feature pair represents two behavioral features whose directional change relationship needs to be examined within the same continuous window. The browsing ratio change feature pointing towards the consultation duration change feature, the consultation duration change feature pointing towards the response interval change feature, and the response interval change feature pointing towards the training completion change feature are respectively considered as the first transmission direction of the corresponding behavioral feature pair, and the transmission direction opposite to the first transmission direction is considered as the second transmission direction. For each transmission direction, the behavioral feature corresponding to the starting point of the transmission direction is considered as the transmission start feature, and the behavioral feature corresponding to the ending point of the transmission direction is considered as the transmission arrival feature. For a continuous window and a number of lag sampling intervals, baseline continuous sampling periods are sequentially set starting from the earliest continuous sampling period within the continuous window. The baseline continuous sampling period represents the time position at which the current behavioral deviation value of the transmission start feature and the current behavioral deviation value of the transmission arrival feature are read. The arrangement position of the lag continuous sampling period = the arrangement position of the baseline continuous sampling period + the number of lag sampling intervals. The lag continuous sampling period represents the time position at which the subsequent behavioral deviation value of the transmitted arrival feature is read. When the lag continuous sampling period is still within the current continuous window, the corresponding baseline continuous sampling period is retained; when the lag continuous sampling period exceeds the current continuous window, the addition of the baseline continuous sampling period is stopped. The behavioral deviation values ​​corresponding to the transmission start feature in each baseline continuous sampling period are read from the individual baseline deviation sequence and arranged in the order of the baseline continuous sampling periods; the behavioral deviation values ​​corresponding to the transmission arrival feature in the same baseline continuous sampling period are read and kept at the same time position as the transmission start feature; then the behavioral deviation values ​​of the transmission arrival feature in the corresponding lag continuous sampling period are read and arranged in the order of the corresponding baseline continuous sampling periods, finally forming the preceding deviation sequence of the start feature, the preceding deviation sequence of the arrival feature, and the subsequent deviation sequence of the arrival feature for each behavioral feature pair under the first transmission direction, the second transmission direction, and each time lag.

[0051] In MATLAB, the preceding deviation sequences of the initial feature, the preceding deviation sequence of the arrival feature, and the subsequent deviation sequences of the arrival feature, all corresponding to the same continuous window, the same behavioral feature pair, the same transmission direction, and the same time lag, are imported into the same workspace. Simultaneously, the behavioral deviation values ​​of the access frequency variation feature and the access interval variation feature, corresponding to the same baseline continuous sampling period, are imported from the individual baseline deviation sequence. The access frequency variation feature and the access interval variation feature are used as conditional information, representing the page access activity changes synchronously given when examining the predictive power of a change in one behavioral feature on the subsequent change in another. Following the arrangement of the baseline continuous sampling periods, the preceding behavioral deviation values ​​of the initial feature, the preceding behavioral deviation values ​​of the arrival feature, the subsequent behavioral deviation values ​​of the arrival feature, the behavioral deviation values ​​of the access frequency variation feature, and the behavioral deviation values ​​of the access interval variation feature are written into the same row, so that each row corresponds to one baseline continuous sampling period. The behavioral deviation values ​​in each row are statistically analyzed by combining the number of times each of the five behavioral deviation values ​​occurs, the number of times the preceding behavioral deviation value of the transmission start feature and the preceding behavioral deviation value of the transmission arrival feature, along with the two conditional information values, the number of times the subsequent behavioral deviation value of the transmission arrival feature, the preceding behavioral deviation value of the transmission arrival feature, and the two conditional information values, and the number of times the preceding behavioral deviation value of the transmission arrival feature and the two conditional information values ​​occur. The probability of the five values ​​occurring together = the number of times each of the five behavioral deviation values ​​occurs ÷ the number of consecutive sampling periods used in the current calculation. The conditional probability after adding the transmission start feature = the number of times each of the five behavioral deviation values ​​occurs ÷ the number of times each of the preceding behavioral deviation values ​​of the transmission start feature, the preceding behavioral deviation value of the transmission arrival feature, the behavioral deviation value of the access frequency change feature, and the behavioral deviation value of the access interval change feature occurs. The conditional probability without the initiating feature is calculated as follows: The number of times the following four values ​​occur together: the subsequent behavioral deviation of the arriving feature, the preceding behavioral deviation of the arriving feature, the behavioral deviation of the access frequency change feature, and the behavioral deviation of the access interval change feature. This number is then divided by the total number of times these three values ​​occur together. The amount of information transmitted by a single joint value is calculated as: the combined probability of the five values ​​× log2 (conditional probability with the initiating feature ÷ conditional probability without the initiating feature). The conditional transmission entropy is the sum of the amounts of information transmitted by all single joint values. Conditional transmission entropy represents the additional predictive information about subsequent changes in the arriving feature after adding a preceding change to the initiating feature, given that the preceding changes, access frequency changes, and access interval changes of the arriving feature are already provided.The conditional transfer entropy of each behavioral feature pair within the same continuous window is calculated sequentially under the first transfer direction, the second transfer direction, and each time lag. The conditional transfer entropy corresponding to different time lags is compared one by one. When the conditional transfer entropy read later is greater than the currently retained conditional transfer entropy, the later read conditional transfer entropy is taken as the current maximum conditional transfer entropy, and the corresponding time lag is recorded synchronously. When the conditional transfer entropy read later is less than the current maximum conditional transfer entropy, the current maximum conditional transfer entropy and its corresponding time lag are retained. After all time lag comparisons are completed, the time lag corresponding to the maximum conditional transfer entropy is read as the transfer lag. The maximum conditional transfer entropy and transfer lag of each behavioral feature pair are arranged according to the order of the continuous window in the first and second transfer directions, ultimately forming a conditional transfer entropy sequence and a transfer lag sequence.

[0052] S4: Each type of behavior feature in the user behavior feature sequence is taken as a behavior node. Directed edges between behavior nodes are determined based on the conditional propagation entropy sequence and the propagation delay sequence. A directed graph of behavior changes is constructed based on these edges, including:

[0053] After the conditional transfer entropy sequence and transfer delay sequence are formed, the number of continuous decision windows is recorded. The number of continuous decision windows is an integer greater than 0, representing the number of adjacent consecutive windows that need to continuously satisfy the comparison condition in the same transfer direction. Following the order of the consecutive windows, starting from the earliest consecutive window position that can cover the number of continuous decision windows, the same number of adjacent consecutive windows as the number of continuous decision windows are selected consecutively. The maximum conditional transfer entropy and corresponding transfer delay for the same behavioral feature pair in the first and second transfer directions are read respectively. The maximum conditional transfer entropy of the two transfer directions within each adjacent consecutive window is compared one by one, and the transfer delay corresponding to the transfer direction with the larger maximum conditional transfer entropy value is checked. The first transmission direction satisfies the comparison condition if its maximum conditional transmission entropy is greater than that of the second transmission direction in all adjacent consecutive windows reaching the number of sustained decision windows, and if all transmission delays corresponding to the first transmission direction satisfy the condition that transmission delay ≤ maximum transmission duration. Similarly, the second transmission direction satisfies the comparison condition if its maximum conditional transmission entropy is greater than that of the first transmission direction in all adjacent consecutive windows reaching the number of sustained decision windows, and if all transmission delays corresponding to the second transmission direction satisfy the condition that transmission delay ≤ maximum transmission duration. If the maximum conditional transmission entropy of a transmission direction in any adjacent consecutive window is not greater than that of the opposite transmission direction, then a transmission direction does not satisfy the comparison condition. If the maximum conditional transmission entropy of a transmission direction is greater than that of the opposite transmission direction in all adjacent consecutive windows, but any corresponding transmission delay satisfies the condition that transmission delay > maximum transmission duration, then a transmission direction does not satisfy the comparison condition. After one comparison, the starting position of the selected adjacent consecutive windows is moved forward by one consecutive window, and the same processing is performed on the next group of adjacent consecutive windows, ultimately forming the transmission directions that satisfy the comparison conditions for each behavioral feature pair.

[0054] After obtaining the transmission directions that satisfy the comparison conditions, the characteristics of visit frequency change, visit interval change, consultation duration change, response interval change, training completion change, and browsing ratio change are registered as behavior nodes. Each behavior node represents the position in the directed graph of behavior changes corresponding to a behavior feature. Each transmission direction that satisfies the comparison conditions is read one by one. The behavior node corresponding to the transmission start feature is placed at the starting position of the directed edge, and the behavior node corresponding to the transmission arrival feature is placed at the ending position of the directed edge. The two behavior nodes are then connected according to the transmission direction. A directed edge represents the behavior change transmission relationship from one behavior node to another. When the first transmission direction satisfies the comparison conditions, a directed edge is established according to the first transmission direction; when the second transmission direction satisfies the comparison conditions, a directed edge is established according to the second transmission direction; when neither the first nor the second transmission direction satisfies the comparison conditions, no directed edge is established between the corresponding behavior nodes. After all directed edges are established, the behavior nodes are arranged according to the pointing relationships of the directed edges. Read each directed edge sequentially, mapping the terminating node of one directed edge to the starting node of another. If two positions correspond to the same behavior node, append the later directed edge to the earlier one, and then use the terminating node of the later directed edge to continue searching for a continuation edge. If two positions correspond to different behavior nodes, do not continuation. If the terminating node of the current directed edge is connected to only one subsequent directed edge, continue arranging along the corresponding directed edge. If the terminating node of the current directed edge is connected to two or more subsequent directed edges, form the corresponding connection order along each subsequent directed edge. If the terminating node of the current directed edge has no subsequent directed edges, stop extending the current connection order. Treat the overall connection relationship between behavior nodes and directed edges as a directed graph of behavior changes. Arrange the head-to-tail directed edges according to their pointing order to form behavior change propagation paths, ultimately forming the directed graph of behavior changes and the behavior change propagation path corresponding to the latest continuous window.

[0055] S5: Determining a User's Mental State Using a Directed Graph of Behavioral Changes. Determining a user's mental state using a directed graph of behavioral changes includes:

[0056] For the directed graph of behavior changes corresponding to the latest continuous window, the number of times each behavior node acts as the termination node of a directed edge and the number of times it acts as the starting node of a directed edge are counted. When a behavior node acts as the termination node of a directed edge 0 times and the number of times it acts as the starting node of a directed edge is greater than 0, the corresponding behavior node is regarded as the source of the user's psychological state change. The source of the user's psychological state change represents the behavior feature that is first transmitted to subsequent behavior nodes in the behavior change propagation path. Following the direction of the directed edges of the behavior change propagation path, subsequent behavior nodes are read sequentially starting from the source of the user's psychological state change. The behavior node that the behavior change propagation path finally reaches is regarded as the stage of the user's psychological state change. The stage of the user's psychological state change represents the position of the behavior feature that the behavior change propagation path has reached in the latest continuous window. Duplicate behavior nodes are removed according to the behavior feature name in the behavior change propagation path. The same behavior feature name is counted only once. The influence range of the user's psychological state equals the number of different behavior nodes passed through by the behavior change propagation path. Next, the behavioral deviation values ​​corresponding to each behavioral node traversed by the behavioral change propagation path are read from the individual baseline deviation sequence within adjacent consecutive sampling periods. The change in behavioral deviation value = behavioral deviation value of the subsequent consecutive sampling period minus behavioral deviation value of the previous consecutive sampling period. When the change in behavioral deviation value is greater than 0, the direction of change of the corresponding behavioral feature is recorded as increasing; when the change in behavioral deviation value is less than 0, the direction of change of the corresponding behavioral feature is recorded as decreasing; when the change in behavioral deviation value is equal to 0, the direction of change of the corresponding behavioral feature is recorded as unchanged. The direction of change corresponding to each behavioral node is arranged according to the order of the behavioral nodes in the behavioral change propagation path to form the trend of user psychological state change. During a user's continuous psychological assessment, counseling, emotion recording, and training, if the behavioral change propagation path sequentially follows this pattern: changes in browsing ratio lead to changes in counseling duration, then to changes in response interval, and finally to changes in training completion, and further, the cumulative counseling interaction time decreases relative to the individual's behavioral baseline, the average counseling response interval increases relative to the individual's behavioral baseline, and the training task completion rate decreases relative to the individual's behavioral baseline, then it can be described as a continuous psychological behavioral change where the user first changed their attention allocation across different psychological service function pages, subsequently reduced psychological counseling interaction, gradually slowed counseling responses, and further exhibited a decline in the completion rate of psychological training tasks. If the behavioral change propagation path only involves changes in browsing ratio leading to changes in counseling duration, then it can be described as a change in the user's attention allocation across different psychological service function pages, which has begun to affect psychological counseling interaction, but the pace of psychological counseling responses and the completion status of psychological training tasks have not yet entered the corresponding propagation path. If the propagation path of behavioral change is that the change in response interval points to the change in training completion, and after the average consultation response interval increases relative to the individual's behavioral baseline, the proportion of training task completion decreases relative to the individual's behavioral baseline, then it can be described as the user first experiencing a slowdown in psychological counseling response, followed by a decrease in the degree of psychological training execution.If the propagation path of behavioral change is from training completion changes to response interval changes, and the proportion of training task completion decreases relative to the individual's behavioral baseline first, followed by a lengthening of the average consultation response interval, then it can be described as the user first experiencing a decrease in persistence in psychological training, followed by a slowdown in psychological counseling responses. Combining the source of the user's psychological state change, the stage of the user's psychological state change, the scope of the user's psychological state influence, the trend of the user's psychological state change, and the propagation path of behavioral change ultimately forms the user's psychological state.

[0057] Traditional methods using correlation coefficients to measure the degree of co-change between two behavioral characteristics within the same time frame can determine whether two behavioral characteristics change in the same or opposite directions, but they cannot determine which of the browsing ratio change and consultation duration change occurred first, nor can they determine how long it takes for one behavioral change to correspond to the subsequent change of the other. Traditional weighted scoring combines multiple behavioral records into a single score according to preset weights, and also eliminates the original temporal order of the behavioral characteristics during the weighting process. This invention first converts different behavioral changes into behavioral deviation values ​​relative to an individual's behavioral baseline, and then calculates the conditional transmission entropy of the same behavioral characteristic pair under two opposite transmission directions and different time lags within a sliding time window. It uses access frequency change characteristics and access interval change characteristics as conditional information to distinguish the historical information of the transmission initiation characteristic from the overall page access activity changes, thereby obtaining the transmission direction and transmission lag of the behavioral change. Compared to applying transfer entropy directly within a single continuous window, this invention also compares the maximum conditional transfer entropy of two opposite transfer directions. It requires that the same transfer direction be consistently greater than the opposite transfer direction within adjacent continuous windows that reach the required number of continuous decision windows, and that the corresponding transfer delays do not exceed the maximum transfer duration. This prevents short-term directional relationships formed by a single psychological assessment, temporary events, or accidental failure to complete a training task from being directly converted into directed edges. The directed edges retained after continuity judgment can be connected to form a behavioral change propagation path, enabling the psychological state analysis results to indicate from which behavior the psychological behavioral change started, to which behavior it is currently propagating, how many behaviors it has affected, and the direction of change of each behavioral deviation value within adjacent continuous sampling periods.

[0058] A continuous analysis system for psychological states based on multidimensional behavioral perception, comprising:

[0059] The feature construction module constructs user behavior feature sequences based on users' multidimensional psychological service behavior data;

[0060] The deviation calculation module calculates the behavioral deviation value corresponding to each type of behavioral feature in the user's behavioral feature sequence and establishes a personal baseline deviation sequence.

[0061] The transfer analysis module divides the sliding time window based on the individual baseline deviation sequence, and determines the conditional transfer entropy sequence and transfer lag sequence between different behavioral characteristics within the sliding time window; whereby, the conditional transfer entropy represents the degree to which a change in one behavioral characteristic predicts the subsequent change in another behavioral characteristic, and the transfer lag represents the time interval between corresponding changes in two behavioral characteristics.

[0062] The graph construction module takes each type of behavior feature in the user behavior feature sequence as a behavior node, determines the directed edges between behavior nodes based on the conditional propagation entropy sequence and the propagation delay sequence, and constructs a directed graph of behavior changes based on the directed edges.

[0063] The state determination module determines the user's psychological state by referring to the directed graph of behavioral changes.

[0064] The psychological state continuous analysis system provided by this invention is used to execute the above-mentioned psychological state continuous analysis method, and has the same or corresponding technical features and effects. Specifically, the feature construction module executes step S1 to construct a user behavior feature sequence based on the user's multi-dimensional psychological service behavior data; the deviation calculation module executes step S2 to calculate the behavior deviation value corresponding to each type of behavior feature in the user behavior feature sequence and establish a personal baseline deviation sequence; the transfer analysis module executes step S3 to divide a sliding time window according to the personal baseline deviation sequence and determine the conditional transfer entropy sequence and transfer delay sequence between different behavior features within the sliding time window; the graph construction module executes step S4 to take each type of behavior feature in the user behavior feature sequence as a behavior node, determine the directed edges between behavior nodes according to the conditional transfer entropy sequence and transfer delay sequence, and construct a directed graph of behavior changes based on the directed edges; and the state determination module executes step S5 to determine the user's psychological state by referring to the directed graph of behavior changes.

[0065] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention in any other way. Any person skilled in the art may make changes or modifications to the above-disclosed technical content to create equivalent embodiments that can be applied to other fields. However, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the protection scope of the present invention.

Claims

1. A method for continuous analysis of psychological states based on multidimensional behavioral perception, characterized in that, Includes the following steps: Constructing user behavior feature sequences based on multidimensional behavioral data of users' psychological services; Calculate the behavioral deviation value corresponding to each type of behavioral feature in the user behavioral feature sequence, and establish an individual baseline deviation sequence; The sliding time window is divided based on the individual baseline deviation sequence. Within the sliding time window, the conditional transfer entropy sequence and the transfer lag sequence between different behavioral features are determined. The conditional transfer entropy represents the predictive power of a change in one behavioral feature on the subsequent change in another behavioral feature, and the transfer lag represents the time interval between corresponding changes in two behavioral features. Each type of behavior feature in the user behavior feature sequence is taken as a behavior node. The directed edges between behavior nodes are determined based on the conditional transfer entropy sequence and the transfer delay sequence. A directed graph of behavior change is constructed based on the directed edges. The user's psychological state can be determined by referring to the directed graph of behavioral changes.

2. The method for continuous analysis of psychological states based on multidimensional behavioral perception according to claim 1, characterized in that, The multidimensional behavioral data of psychological services specifically refers to the behavioral records collected by users when they log in to the psychological service platform and browse the psychological service function page, based on continuous sampling periods. The continuous sampling period is a series of consecutive time periods divided according to a preset sampling period.

3. The method for continuous analysis of psychological states based on multidimensional behavioral perception according to claim 2, characterized in that, The psychological service function pages include a psychological assessment page, a psychological counseling page, an emotion recording page, and a psychological training page; the behavior records include access frequency, access interval, cumulative counseling interaction time, average counseling response interval, training task completion rate, and browsing category ratio; wherein, access frequency and access interval are obtained based on the page access records of each psychological service function page, cumulative counseling interaction time and average counseling response interval are obtained based on the counseling session records and message interaction records of the psychological counseling page, training task completion rate is obtained based on the training task records of the psychological training page, and browsing category ratio is obtained based on the page browsing records of each psychological service function page.

4. The method for continuous analysis of psychological states based on multidimensional behavioral perception according to claim 1, characterized in that, The user behavior feature sequence includes access frequency change features, access interval change features, consultation duration change features, response interval change features, training completion change features, and browsing ratio change features.

5. The method for continuous analysis of psychological states based on multidimensional behavioral perception according to claim 4, characterized in that, The access frequency variation feature is the difference in access frequency between adjacent sampling periods; the access interval variation feature is the difference in access interval between adjacent sampling periods; the consultation duration variation feature is the difference in cumulative consultation interaction duration between adjacent sampling periods; the response interval variation feature is the difference in average consultation response interval between adjacent sampling periods; the training completion variation feature is the difference in the proportion of training tasks completed by users between adjacent sampling periods; and the browsing ratio variation feature is the sum of the absolute values ​​of the differences in the proportions of each browsing category between users between adjacent sampling periods, divided by two.

6. The method for continuous analysis of psychological states based on multidimensional behavioral perception according to claim 5, characterized in that, The step of calculating the behavioral deviation value corresponding to each type of behavioral feature in the user behavioral feature sequence and establishing a personal baseline deviation sequence includes: A baseline duration is set. Following the sequential order of continuous sampling periods, the user behavior feature sequence selects the following features: cumulative duration of access frequency changes, access interval changes, consultation duration changes, response interval changes, training completion changes, and browsing ratio changes, corresponding to the baseline duration. The range of continuous sampling periods corresponding to each selected feature is used as the baseline time period. The median and median absolute deviation of each feature within the baseline time period are calculated to form a personal behavior baseline. Calculate the difference between each feature corresponding to each consecutive sampling period and the corresponding median in the personal behavior baseline. When the corresponding median absolute deviation is not zero, divide the difference by the corresponding median absolute deviation to obtain the behavior deviation value of the corresponding feature. When the corresponding median absolute deviation is zero, use the difference as the behavior deviation value of the corresponding feature. Arrange the behavior deviation values ​​corresponding to each feature in the order of the consecutive sampling periods to obtain the personal baseline deviation sequence.

7. The method for continuous analysis of psychological states based on multidimensional behavioral perception according to claim 6, characterized in that, The step of dividing a sliding time window based on an individual baseline deviation sequence and determining the conditional transfer entropy sequence and transfer delay sequence between different behavioral characteristics within the sliding time window includes: Set the sliding time window, sliding step size, and maximum transmission duration. Generate a time lag set from one preset sampling period to the maximum transmission duration according to integer multiples of the preset sampling period. Based on the user's personal baseline deviation sequence, divide the continuous window according to the sliding time window and sliding step size, and calculate the ratio of each time lag in the time lag set to the preset sampling period to obtain the number of lag sampling intervals. The browsing ratio change feature and consultation duration change feature, consultation duration change feature and response interval change feature, and response interval change feature and training completion change feature in the user behavior feature sequence are respectively formed into behavior feature pairs, and the first and second transmission directions of each behavior feature pair are determined respectively; for each transmission direction, the behavior feature corresponding to the starting point of the transmission direction is taken as the transmission start feature, and the behavior feature corresponding to the ending point of the transmission direction is taken as the transmission arrival feature; within the continuous window, continuous sampling periods are sequentially selected as the benchmark continuous sampling periods, and the behavior deviation values ​​corresponding to the transmission start feature in each benchmark continuous sampling period are extracted from the personal baseline deviation sequence to form the starting feature pre-sequence deviation sequence, and the transmission arrival features are extracted in each benchmark continuous sampling period. The behavioral deviation values ​​corresponding to the quasi-continuous sampling periods form an arrival feature pre-sequence deviation sequence; the continuous sampling periods separated from each benchmark continuous sampling period by the number of lag sampling intervals are determined as lag continuous sampling periods, and the behavioral deviation values ​​corresponding to the arrival features in each lag continuous sampling period are extracted to form an arrival feature subsequent deviation sequence; wherein, the first transmission direction includes a transmission direction starting with the browsing ratio change feature and ending with the consultation duration change feature, a transmission direction starting with the consultation duration change feature and ending with the response interval change feature, and a transmission direction starting with the response interval change feature and ending with the training completion change feature; the second transmission direction is the transmission direction opposite to the first transmission direction for the corresponding behavioral feature pair; Based on the preceding deviation sequence of the initial feature, the preceding deviation sequence of the arrival feature, and the subsequent deviation sequence of the arrival feature corresponding to each behavioral feature pair in the first and second transmission directions, the behavioral deviation values ​​corresponding to the access frequency change feature and access interval change feature in the personal baseline deviation sequence during each continuous sampling period are used as conditional information. The conditional transfer entropy algorithm is used to calculate the conditional transfer entropy of each behavioral feature pair under different transmission directions and different time lags. Within the same continuous window, the conditional transfer entropy with the largest value is determined as the maximum conditional transfer entropy of the corresponding behavioral feature pair and the corresponding transmission direction, and the time lag corresponding to the maximum conditional transfer entropy is determined as the transmission lag. According to the order of the continuous window, the maximum conditional transfer entropy and transmission lag of each behavioral feature pair in the first and second transmission directions are arranged to form the conditional transfer entropy sequence and the transmission lag sequence, respectively.

8. The method for continuous analysis of psychological states based on multidimensional behavioral perception according to claim 7, characterized in that, The constructed behavior change directed graph includes: The user behavior feature sequence includes the access frequency change feature, access interval change feature, consultation duration change feature, response interval change feature, training completion change feature, and browsing ratio change feature, which are respectively used as behavior nodes. A continuous decision window number is set, and the maximum conditional transfer entropy of each behavior feature is compared with the conditional transfer entropy sequence and the transfer delay sequence. When the maximum conditional transfer entropy of a transfer direction is greater than the maximum conditional transfer entropy of the opposite transfer direction in adjacent consecutive windows that reach the continuous decision window number, and the corresponding transfer delay does not exceed the maximum transfer duration, the corresponding transfer direction is determined to meet the comparison condition. The behavior node corresponding to the starting point of the propagation direction that satisfies the comparison condition is taken as the starting node of the directed edge, and the behavior node corresponding to the ending point of the propagation direction is taken as the ending node of the directed edge. A directed edge is established from the starting node of the directed edge to the ending node of the directed edge. A directed graph of behavior change is constructed based on the behavior node and the directed edge, and the directed edges that connect the beginning and end of the directed graph of behavior change form the behavior change propagation path.

9. The method for continuous analysis of psychological states based on multidimensional behavioral perception according to claim 8, characterized in that, The directed graph of reference behavior changes determines the user's psychological state, including: The behavioral nodes in the directed graph of behavioral changes that have no other directed edges pointing to them and are connected outward by directed edges are identified as the sources of user psychological state changes; the behavioral nodes reached by the propagation path of the behavioral changes corresponding to the latest continuous window are identified as the stages of user psychological state changes; the number of different behavioral nodes traversed by the propagation path of behavioral changes is identified as the scope of influence of user psychological state; the trend of user psychological state changes is determined based on the direction of change of each behavioral deviation value in the personal baseline deviation sequence between adjacent continuous sampling periods; and the user psychological state is represented by combining the sources of user psychological state changes, stages of user psychological state changes, scope of influence of user psychological state changes, trend of user psychological state changes, and propagation path of behavioral changes.

10. A continuous analysis system for psychological states based on multidimensional behavioral perception, characterized in that, The system includes: The feature construction module constructs user behavior feature sequences based on users' multidimensional psychological service behavior data; The deviation calculation module calculates the behavioral deviation value corresponding to each type of behavioral feature in the user's behavioral feature sequence and establishes a personal baseline deviation sequence. The transfer analysis module divides the sliding time window based on the individual baseline deviation sequence, and determines the conditional transfer entropy sequence and transfer lag sequence between different behavioral characteristics within the sliding time window; whereby, the conditional transfer entropy represents the degree to which a change in one behavioral characteristic predicts the subsequent change in another behavioral characteristic, and the transfer lag represents the time interval between corresponding changes in two behavioral characteristics. The graph construction module takes each type of behavior feature in the user behavior feature sequence as a behavior node, determines the directed edges between behavior nodes based on the conditional propagation entropy sequence and the propagation delay sequence, and constructs a directed graph of behavior changes based on the directed edges. The state determination module determines the user's psychological state by referring to the directed graph of behavioral changes.

Images