Methods, devices, and equipment for detecting affective dependence based on dimensionality missing compensation

By adaptively determining the subset of available dimensions and the set of missing dimensions, and using the missing dimension estimation relationship for compensation scoring and error calculation, the problem of unreliable results caused by missing detection dimension data is solved, achieving a balance between usability and reliability in affective dependence risk detection.

CN123115676APending Publication Date: 2026-11-17CHANGSHA HERONG SPACE-TIME TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611337427.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-31
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In the detection of emotional dependence risk, when some detection dimensions are missing data, existing solutions either abandon the detection, resulting in limited data, or directly calculate in the absence of dimensions, leading to unreliable detection results and easy misintervention or insufficient intervention.

Method used

By obtaining the data availability determination results of the detection dimensions, the available dimension subset and the missing dimension set are adaptively determined. The pre-calibrated missing dimension estimation relationship and the measured score estimation are used to compensate for the score and error. The weighted summation is then performed to output the detection result with confidence. When the confidence is lower than the threshold, supplementary data collection is triggered to prevent intervention measures.

Benefits of technology

Even when the detection dimension data is incomplete, it can still output detection results carrying credibility information, taking into account both the usability and credibility of the detection, avoiding false intervention, and ensuring that downstream links can credibly accept the detection results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123115676A_ABST
    Figure CN123115676A_ABST
Patent Text Reader

Abstract

This application relates to a method, apparatus, and device for detecting affective dependence risk based on dimension missing compensation. The method includes: acquiring data availability assessment results for each detection dimension, determining a subset of available dimensions and a set of missing dimensions; when the number of detection dimensions in the subset of available dimensions meets a preset quantity condition, calculating a measured score using a scorer; estimating the compensation score and estimation error of each detection dimension in the set of missing dimensions based on a pre-calibrated missing dimension estimation relationship and the measured score; weighting and summing the measured score and the compensation score to obtain an affective dependence risk value, and determining the confidence level based on weight coverage and estimation error; when the confidence level is lower than a preset confidence threshold, outputting a prompt message and triggering supplementary data collection for the set of missing dimensions, while prohibiting intervention measures. This method can output detection results carrying confidence levels even when detection dimension data is incomplete.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of affective computing technology, and in particular to a method, apparatus and device for detecting affective dependency risk based on dimension missing compensation. Background Technology

[0002] With the development of affective computing and user behavior analysis technologies, detecting users' emotional dependence on services through multi-dimensional behavioral data has become an important foundation for risk warning and compliance intervention. Complete emotional dependence risk detection typically requires data support from multiple detection dimensions, such as interaction intensity, emotional substitution, dependence during vulnerable moments, and withdrawal reactions.

[0003] However, in actual products, the data collection conditions for each detection dimension vary significantly, and data for some detection dimensions is difficult to obtain in most products. For example, the detection of withdrawal symptoms relies on behavioral and emotional sensing data during service unavailability, while the detection of emotional substitution relies on real social control data. When data for a detection dimension is incomplete, existing solutions either abandon the detection, preventing data-limited products from conducting emotional dependence risk detection; or they directly calculate the results when the dimension is missing, resulting in unreliable test results that lack credibility information. Downstream intervention stages cannot grade and accept the test results, easily leading to misinterpretation or insufficient intervention. Summary of the Invention

[0004] Therefore, it is necessary to provide a method, apparatus, and device for detecting affective dependence risk based on dimension missing compensation, which can output detection results with confidence scores even when the detection dimension data is incomplete, in order to address the above-mentioned technical problems.

[0005] An emotion dependence risk detection method based on dimensional missing compensation, the method comprising: Obtain the data availability determination results for each detection dimension in a preset dependency detection dimension set, and determine the available dimension subset and the missing dimension set based on the data availability determination results; wherein, the dependency detection dimension set includes multiple detection dimensions that characterize the user's emotional dependency risk; When the number of detection dimensions in the available dimension subset meets the preset quantity condition, the measured score of each detection dimension is calculated according to the scorer corresponding to each detection dimension in the available dimension subset. Based on the pre-calibrated missing dimension estimation relationship and each of the measured scores, the compensation score and corresponding estimation error of each detection dimension in the missing dimension set are estimated; wherein, the missing dimension estimation relationship represents the mapping relationship between the scores of some detection dimensions and the estimation error of the mapping relationship, which is obtained by offline calibration of a calibration sample set that meets the preset calibration integrity conditions; The measured scores and the compensated scores are weighted and synthesized to obtain the emotional dependence risk value. The confidence level corresponding to the emotional dependence risk value is determined based on the weight coverage of the measured scores in the weighted synthesis and the estimation error of each score. The emotional dependence risk value is gated and output based on the confidence level; wherein, when the confidence level is lower than a preset confidence threshold, a prompt message is output and supplementary data collection for the missing dimension set is triggered, and intervention measures are prohibited based on the emotional dependence risk value.

[0006] In one embodiment, obtaining the data availability determination result for each detection dimension in a preset dependency detection dimension set includes: for each detection dimension in the dependency detection dimension set, detecting whether the corresponding data source exists, whether the data sample size meets the standard, and whether the data authorization is valid; when the data source corresponding to the detection dimension exists, the data sample size meets the standard, and the data authorization is valid, the data for the detection dimension is determined to be available, and the detection dimension is assigned to the available dimension subset; otherwise, the detection dimension is assigned to the missing dimension set; wherein, each detection dimension in the dependency detection dimension set records a data requirement description and a scorer identifier, the data requirement description is used to determine data availability, and the scorer identifier is used to determine the corresponding scorer; triggering supplementary data collection for the missing dimension set includes: generating supplementary collection suggestions based on the data requirement descriptions of each detection dimension in the missing dimension set; wherein, the supplementary collection suggestions include at least one of requesting data authorization and extending the observation period; after performing supplementary data collection according to the supplementary collection suggestions, the data availability determination result for each detection dimension is re-obtained, and the emotional dependency risk value and the confidence level are recalculated based on the re-determined available dimension subset and missing dimension set. The scores for each detection dimension are calculated based on machine-readable behavioral data, such as interaction log event sequences, session duration and frequency distribution, access time distribution, access behavior during service unavailability, and terminal collection status data.

[0007] In one embodiment, the preset quantity condition is that the available dimension subset contains no less than two detection dimensions; when the number of detection dimensions in the available dimension subset does not meet the preset quantity condition, an undetectable state and a list of missing data are output, and the emotional dependence risk value is not generated; the dependence detection dimension set includes at least three detection dimensions from the following: interaction intensity dimension, emotional substitution dimension, vulnerability moment dependence dimension, withdrawal reaction dimension, and time period offset dimension.

[0008] In one embodiment, determining the confidence level corresponding to the emotional dependence risk value based on the weight coverage of the measured scores in the weighted synthesis and each of the estimation errors includes: determining the weight coverage based on the sum of the weights of the measured scores of each detection dimension in the available dimension subset in the weighted synthesis; determining the confidence level based on the weight coverage and the estimation errors of each of the compensation scores; wherein the confidence level is positively correlated with the weight coverage and negatively correlated with each of the estimation errors; gating the output of the emotional dependence risk value based on the confidence level further includes: when the confidence level is not lower than a first preset confidence threshold, outputting the emotional dependence risk value normally; wherein the normally output emotional dependence risk value is used to trigger graded intervention measures; when the confidence level is lower than the first preset confidence threshold but not lower than the preset confidence threshold, marking the emotional dependence risk value as a medium confidence result and reducing the level of the intervention measures triggered based on the emotional dependence risk value.

[0009] In one embodiment, the method further includes: when the measured score of any detection dimension in the available dimension subset hits a preset extreme risk rule, it is not subject to the gating output restriction of the confidence level on the emotional dependence risk value, and an emergency handling process is triggered; the emergency handling process includes at least one of manual review and immediate safety intervention.

[0010] In one embodiment, the detection output of the gated output carries a list of used dimensions, a list of compensated dimensions, and a version identifier of the missing dimension estimation relationship; wherein, the list of used dimensions records the detection dimensions corresponding to the measured scores participating in the weighted comprehensive, and the list of compensated dimensions records the detection dimensions that have been compensated; the method further includes: writing the emotional dependence risk value, the confidence level, the list of used dimensions, the list of compensated dimensions, the version identifier of the missing dimension estimation relationship, and the alarm status into a detection record; the detection record is retained as a compliance event record for review by regulatory authorities with limited scope.

[0011] In one embodiment, the missing dimension estimation relationship is a dimension correlation matrix; each matrix element in the dimension correlation matrix includes the estimation coefficient and estimation error of the score of the first detection dimension on the score of the second detection dimension; wherein, the first detection dimension and the second detection dimension are different detection dimensions in the dependent detection dimension set; the dimension correlation matrix records the calibration sample size and calibration time, and is set in layers according to user groups; the dimension correlation matrix is ​​calibrated offline in the following way: on user groups where data for each detection dimension in the dependent detection dimension set is available, regression calibration is periodically performed to obtain a new version of the dimension correlation matrix; the new version of the dimension correlation matrix is ​​verified offline, and after verification, the dimension correlation matrix used online is replaced with the new version of the dimension correlation matrix in grayscale; wherein, the offline calibration only uses the scoring layer data of each detection dimension and does not use the original interactive content.

[0012] In one embodiment, the preset calibration integrity condition includes: the calibration sample set contains samples in which data for each detection dimension in the dependent detection dimension set is available, or the data coverage of each detection dimension in the calibration sample set is not lower than a preset coverage threshold; the calibration sample set includes at least one of real user samples, synthetic samples generated from real user samples, and partially complete samples that have undergone missing data completion processing.

[0013] An emotion dependence risk detection device based on dimension missing compensation, the device comprising: The availability determination module is used to obtain the data availability determination results for each detection dimension in a preset dependency detection dimension set, and determine the available dimension subset and the missing dimension set based on the data availability determination results; wherein, the dependency detection dimension set includes multiple detection dimensions that characterize the user's emotional dependency risk; The measured scoring module is used to calculate the measured score of each detection dimension according to the scorer corresponding to each detection dimension in the available dimension subset when the number of detection dimensions in the available dimension subset meets the preset number condition. The compensation estimation module is used to estimate the compensation score and corresponding estimation error of each detection dimension in the missing dimension set based on the pre-calibrated missing dimension estimation relationship and each of the measured scores; wherein, the missing dimension estimation relationship represents the mapping relationship between the scores of some detection dimensions and the estimation error of the mapping relationship, which is obtained by offline calibration of a calibration sample set that meets the preset calibration integrity conditions; The comprehensive calculation module is used to perform weighted comprehensive calculation on each of the measured scores and each of the compensated scores to obtain the emotional dependence risk value, and to determine the confidence level corresponding to the emotional dependence risk value based on the weight coverage of the measured scores in the weighted comprehensive calculation and each of the estimation errors. The gating output module is used to gating the output of the emotional dependence risk value based on the confidence level; wherein, when the confidence level is lower than a preset confidence threshold, a prompt message is output and supplementary data collection for the missing dimension set is triggered, and intervention measures are prohibited based on the emotional dependence risk value.

[0014] A computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the above-described method for detecting affective dependence risk based on dimensionality missing compensation.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for detecting emotion dependency risk based on dimensionality missing compensation.

[0016] A computer program product includes a computer program that, when executed by a processor, implements the steps of the above-described method for detecting affective dependence risk based on dimensionality missing compensation.

[0017] The aforementioned method, apparatus, and equipment for detecting affective dependence based on missing dimension compensation adaptively determine the available dimension subset and the missing dimension set by acquiring the data availability assessment results for each detection dimension, enabling the detection to continue even when some dimension data is missing. When the available dimension subset meets a preset quantity condition, the measured score is calculated based on the scorer corresponding to each detection dimension. Then, based on a pre-calibrated missing dimension estimation relationship, the compensated score and estimation error for the missing dimensions are estimated from the measured score. This mapping relationship, where the scores of some detection dimensions are used to estimate the scores of other detection dimensions, compensates for the information gaps in the missing dimensions, avoiding direct abandonment of detection or failure to detect data when dimensions are missing. In the absence of data, the system directly calculates the risk value; then, it weights and combines the measured score and the compensated score to obtain the emotional dependence risk value, and determines the confidence level based on the weight coverage and estimation error, so that the detection result carries credibility information; finally, it gates the output of the emotional dependence risk value based on the confidence level. When the confidence level is lower than the preset confidence threshold, it only outputs a prompt message and triggers the collection of supplementary data for the missing dimension set, and prohibits the triggering of intervention measures. Downstream intervention links can classify the acceptance of detection results according to the confidence level, and improve the confidence level of subsequent detections by collecting supplementary data, so as to balance the usability and credibility of emotional dependence risk detection even when the detection dimension data is incomplete. Attached Figure Description

[0018] Figure 1This is an application scenario diagram of an emotion dependency risk detection method based on dimension missing compensation in one embodiment; Figure 2 This is a flowchart illustrating an emotion dependency risk detection method based on dimension missing compensation in one embodiment. Figure 3 This is a structural block diagram of an emotion dependence risk detection device based on dimension missing compensation in one embodiment; Figure 4 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0020] The emotion dependence risk detection method based on dimensional missing compensation provided in this application can be applied to, for example... Figure 1 In the application environment shown, terminal 102 communicates with server 104 via a network. A user-facing application runs on terminal 102, generating user behavior data during operation. Server 104 collects this behavior data and performs emotional dependency risk detection based on it. Terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. Server 104 can be a standalone server or a server cluster consisting of multiple servers.

[0021] In one embodiment, such as Figure 2 As shown, an emotion dependence risk detection method based on dimension missing compensation is presented, and this method is applied to... Figure 1 Taking the server in the example, the following steps are included: Step 202: Obtain the data availability determination results for each detection dimension in the preset dependency detection dimension set, and determine the available dimension subset and the missing dimension set based on the data availability determination results.

[0022] The dependency detection dimension set includes multiple detection dimensions characterizing the user's emotional dependency risk, such as interaction intensity dimension, emotional substitution dimension, vulnerability moment dependency dimension, withdrawal reaction dimension, time period shift dimension, and consumption slope dimension. It can include at least three of the following dimensions: interaction intensity dimension, emotional substitution dimension, vulnerability moment dependency dimension, withdrawal reaction dimension, and time period shift dimension. Each detection dimension records a data requirement description and a scorer identifier. The data requirement description specifies the data source, sample size, and authorization requirements upon which the scoring of that detection dimension depends. The scorer identifier identifies the corresponding scorer for that detection dimension. Data availability determination is performed for each detection dimension individually, based on the existence of the corresponding data source, whether the data sample size meets the requirements, and whether the data authorization is valid. The available dimension subset is the set of detection dimensions for which data is available, and the missing dimension subset is the set of detection dimensions for which data is not available. This step, by determining availability before defining the detection dimension range, allows the detection process to adapt to the data conditions of different products, rather than assuming all dimensions are available as a prerequisite for detection.

[0023] Step 204: When the number of detection dimensions in the available dimension subset meets the preset quantity condition, calculate the measured score of each detection dimension according to the scorer corresponding to each detection dimension in the available dimension subset.

[0024] The preset quantity condition is the minimum number of dimensions required for detection to be operational; for example, the number of detectable dimensions in the available dimension subset must be no less than 2. When the number of detectable dimensions in the available dimension subset does not meet the preset quantity condition, the sentiment dependency risk value is not output, but an undetectable state and a list of missing data are output to avoid outputting unreliable results when information is severely insufficient. The scorer is a scoring calculation unit corresponding to the detection dimension. It is located according to the scorer identifier and is used to convert the behavioral data corresponding to the detection dimension into a measured score.

[0025] Step 206: Based on the pre-calibrated missing dimension estimation relationship and each measured score, estimate the compensation score and corresponding estimation error of each detection dimension in the missing dimension set.

[0026] The missing dimension estimation relationship represents the mapping relationship between the scores of some detection dimensions and the scores of the remaining detection dimensions, along with the estimation error of the mapping relationship. This is obtained through offline calibration using a calibration sample set that meets preset calibration integrity conditions. The compensation score is an estimate of the scores of the detection dimensions in the missing dimension set, calculated from the measured scores of the available dimension subset and the missing dimension estimation relationship, while simultaneously recording the estimation error. By using the estimation relationship between dimensions to compensate for missing dimensions, the information gaps in the missing dimensions are filled, allowing detection to continue even with missing dimensions. The uncertainty of the compensation is explicitly retained in the form of estimation error, providing a basis for subsequent confidence calculations. The missing dimension estimation relationship is not limited to a specific estimation method; for example, it can be a dimension correlation matrix, a Bayesian estimation model, a tree model, or a sub-model trained by combining missing dimensions.

[0027] In one embodiment, the compensation score for each detection dimension in the missing dimension set is calculated as follows:

[0028] in, S represents the compensation score for the j-th detection dimension in the set of missing dimensions; S represents the subset of available dimensions. This represents the measured score of the i-th detection dimension in the available dimension subset; To estimate the coefficients, we characterize the strength of the mapping from the scores of the available dimensions to the scores of the missing dimension. The combination coefficients are normalized according to the weights of the available dimensions, and the sum of the combination coefficients of all available dimensions is 1.

[0029] The composite estimation error for each missing dimension is calculated as follows:

[0030] in, Indicates the composite estimation error. This represents the estimation error when estimating the j-th missing dimension from the i-th available dimension.

[0031] The total error of the missing dimension set is calculated as follows:

[0032] in, E This represents the total error of the set of missing dimensions. M For the set of missing dimensions, Let be the weight of the j-th missing dimension in the weighted synthesis.

[0033] Weight coverage is calculated as follows:

[0034] Wherein, CV stands for weight coverage, which is the weight coverage of the measured score in the weighted summation. The weight of the i-th detection dimension in the weighted synthesis from the available dimension subset; D For the set of dependent detection dimensions; This represents the weight of the k-th detection dimension in the weighted synthesis within the dependency detection dimension set.

[0035] The confidence level is calculated as follows:

[0036] in, This represents the confidence level corresponding to the risk value of emotional dependence. This is the preset upper limit for total error normalization; The normalized total error is capped at 1; the calculation results are truncated in the interval from 0 to 1.

[0037] Boundary cases are handled as follows: when the missing dimension set is empty, the total error is 0, the weight coverage is 1, and the confidence level is 1; when the number of detectable dimensions in the available dimension subset is less than two, the undetectable state and the list of missing data are output, and no sentiment dependency risk value is generated; when the missing dimension estimation relation is missing an item for a certain missing dimension, the missing dimension is not compensated, is removed from the weighted aggregation and the weights are renormalized, and the weight coverage decreases accordingly.

[0038] In one embodiment, numerical calculations are performed using five detection dimensions D1 to D5 as an example. The weights of each detection dimension in the weighted summation are as follows: The three-dimensional data of a user, D1, D3, and D5, can be obtained. Data for dimensions D2 and D4 is missing. In the missing dimension estimation relation, the estimated coefficient of D2 is estimated from D1. The estimation error is The estimated coefficients of D4 are estimated from D5 as follows: The estimation error is Both missing dimensions have only one usable dimension for estimation, and the combination coefficient is... The compensation scores are respectively and The risk value of emotional dependence is... Weight coverage is The total error is The upper limit of the total error normalization is set to 0.12, and the confidence level is [value missing]. If the value falls within the middle confidence interval, the emotional dependence risk value is marked as a middle confidence result and downgraded to trigger intervention measures.

[0039] Step 208: Weight the measured scores and the compensated scores to obtain the emotional dependence risk value, and determine the confidence level corresponding to the emotional dependence risk value based on the weight coverage of the measured scores in the weighted synthesis and the estimation errors.

[0040] The affective dependence risk score is a weighted composite result of the scores from various detection dimensions, with the actual scores and compensation scores participating in the composite according to their respective weights. Weight coverage is the sum of the weights of the actual scores in the weighted composite, reflecting the proportion of actual information in the detection result. Confidence level is determined based on weight coverage and the estimation errors; higher weight coverage and smaller estimation errors result in higher confidence. The detection result carries a confidence level output, enabling downstream processes to understand the reliability of the detection result.

[0041] Step 210: Gated output of emotional dependence risk value based on confidence level.

[0042] Gated output controls the output method and subsequent use of affective dependence risk values ​​based on confidence levels. When the confidence level is lower than a preset threshold, a prompt message is output and supplementary data collection for the missing dimension set is triggered, prohibiting intervention measures based on affective dependence risk values ​​and avoiding low-reliability detection results directly driving intervention. It should be noted that the affective dependence risk values ​​output in this application are used for interaction risk management and technical control in anthropomorphic interactive services and do not constitute disease diagnosis results or medical diagnosis conclusions.

[0043] In the aforementioned method for detecting affective dependence risk based on missing dimension compensation, the available dimension subset and the missing dimension set are adaptively determined through data availability judgment, enabling the detection to continue even when some dimension data is missing. The missing dimension estimation relationship is used to estimate the scoring and estimation errors from the measured scores, thus filling the information gaps of the missing dimensions. The affective dependence risk value is obtained through weighted synthesis and the confidence level is determined, so that the detection results carry credibility information. The gated output is used to match the intervention steps with the credibility of the detection results, thereby taking into account both the usability and credibility of the detection even when the detection dimension data is incomplete.

[0044] In one embodiment, the server checks the existence of the corresponding data source, the data sample size, and the validity of the data authorization for each detection dimension in the dependent detection dimension set. When the data source corresponding to the detection dimension exists, the data sample size meets the requirements, and the data authorization is valid, the data for that detection dimension is determined to be available, and the detection dimension is included in the available dimension subset; otherwise, the detection dimension is included in the missing dimension set. In this embodiment, all three determination conditions must be met simultaneously to determine that the data is available; if any condition is not met, the dimension is included in the missing dimension set, avoiding misjudgments caused by a single condition. For example, if the data source for a certain detection dimension has been connected but the user authorization has expired, the data for that detection dimension is still determined to be unavailable, ensuring the compliance of data use from the detection process. The data requirement description provides the data standards for each detection dimension for availability determination, and the scorer identifier is used to locate the scorer corresponding to each detection dimension in step 204.

[0045] In one embodiment, when supplementary data collection for a set of missing dimensions is triggered, the server generates supplementary collection suggestions based on the data requirement descriptions of each detection dimension in the set of missing dimensions. These suggestions include at least one of requesting data authorization and extending the observation period. For example, for detection dimensions included in the set of missing dimensions due to invalid data authorization, a suggestion to request data authorization is generated to guide the user to complete the authorization; for detection dimensions included in the set of missing dimensions due to insufficient data sample size, a suggestion to extend the observation period is generated to accumulate samples through extended observation. After supplementary data collection is performed according to the supplementary collection suggestions, the data availability determination results for each detection dimension are re-acquired, and the emotional dependence risk value and confidence level are recalculated based on the newly determined subset of available dimensions and the set of missing dimensions. Supplementary data collection expands the subset of available dimensions, shrinks the set of missing dimensions, thereby increasing weight coverage, reducing compensation scores, and correspondingly increasing confidence, forming a closed loop where low-confidence detection guides data supplementation, and data supplementation improves detection confidence.

[0046] In one embodiment, the gating output includes three stages of processing. When the confidence level is not lower than a first preset confidence threshold, the emotional dependence risk value is output normally, and this normally output emotional dependence risk value is used to trigger tiered intervention measures. When the confidence level is lower than the first preset confidence threshold but not lower than a preset confidence threshold, the emotional dependence risk value is marked as a medium-confidence result, and the level of the intervention measures triggered based on the emotional dependence risk value is reduced. When the confidence level is lower than the preset confidence threshold, only a prompt message is output, and supplementary data collection is triggered. In this embodiment, the intervention intensity is matched step-by-step with the credibility of the detection result, and the intervention measures corresponding to the medium-confidence result are executed at a lower level, which not only preserves the reference value of the detection result but also reduces the risk of erroneous intervention. Tiered intervention measures are technical control actions executed by the system, such as reducing the frequency of interaction responses, limiting the session duration, pausing specific interaction functions, switching interaction modes, or requesting manual review.

[0047] In one embodiment, when the measured score of any detection dimension in the available dimension subset matches a preset extreme risk rule, the gating output of the confidence level on the emotional dependence risk value is not restricted, triggering an emergency handling process. This emergency handling process includes at least one of manual review and immediate safety intervention. It should be noted that the gating constraint is the acceptance of the weighted summation of the emotional dependence risk value, while the extreme risk rule applies to extreme signals of a single measured dimension, based on measured data rather than compensation estimation results. The bypass gating of extreme signals in a single measured dimension and the gating constraint do not conflict. Therefore, even users with extremely limited data can receive timely handling of high-risk situations.

[0048] In one embodiment, the server determines the weight coverage based on the sum of the weights of the measured scores of each detection dimension in the available dimensional subset in the weighted summation. Then, it determines the confidence level based on the weight coverage and the estimation errors of each compensation score. The confidence level is positively correlated with the weight coverage and negatively correlated with each estimation error. For example, the weight coverage can be used as the basis for the confidence level, and each estimation error can be used as a penalty. The higher the weight of the measured score and the lower the uncertainty of the compensation estimate, the higher the confidence level. This embodiment explicitly links the confidence level with the proportion of measured information and the compensation uncertainty, ensuring that the confidence level accurately reflects the composition of the measured and compensation components in the detection results.

[0049] In one embodiment, the missing dimension estimation relationship is represented by a dimension correlation matrix. Each element of the matrix includes the estimated coefficient and estimation error of the score of the first detection dimension on the score of the second detection dimension. The first and second detection dimensions are different detection dimensions from the dependent detection dimension set. The estimation coefficient is used to estimate the compensated score of the second detection dimension from the measured score of the first detection dimension, and the estimation error characterizes the uncertainty of the estimate. The dimension correlation matrix records the calibration sample size and calibration time, facilitating traceability of the matrix's calibration basis. It is also stratified by user group, with different user groups using corresponding stratified dimension correlation matrices to improve the fit between the compensated estimate and the user group.

[0050] In another embodiment, the missing dimension estimation relation is a Bayesian estimation model. Based on the joint prior distribution of scores for each detection dimension, the posterior distribution of the missing dimension scores is calculated according to the measured scores of the available dimension subset. The posterior mean is used as the compensation score, and the posterior variance or posterior standard deviation is used as the estimation error. In yet another embodiment, the missing dimension estimation relation is a tree model. A regression tree or regression tree ensemble is trained using the scores of each detection dimension of a user group with complete dimension data. The compensation score for the missing dimension is predicted based on the measured scores of the available dimension subset, and the variance of the sample scores within the leaf nodes or the cross-validation error is used as the estimation error. In yet another embodiment, a sub-model is trained for each combination of available dimensions. During detection, the corresponding sub-model is selected based on the current available dimension subset, and the selected sub-model outputs the compensation score and estimation error. All the above implementations output compensation scores and corresponding estimation errors, and all belong to the implementation forms of the missing dimension estimation relation.

[0051] In one embodiment, the dimensional relevance matrix is ​​calibrated offline as follows: Regression calibration is periodically performed on a user group where data for each detection dimension in the dependent detection dimension set is available, resulting in a new version of the dimensional relevance matrix; the new version of the dimensional relevance matrix is ​​then validated offline, and upon successful validation, it is gradually replaced with the online dimensional relevance matrix. In this embodiment, calibration is performed offline on a user group with complete dimensional data, without affecting online detection; the gradual replacement with a new version of the matrix reduces the risk of fluctuations introduced by matrix updates; the calibration sample size and calibration time are recorded with the matrix version, supporting version traceability. Furthermore, the offline calibration uses only the scoring layer data for each detection dimension, without using the original interaction content. The user's original interaction content is not involved in the calibration or transmitted back, and the calibration process meets the data minimization requirement.

[0052] The calibration sample set must meet preset calibration integrity conditions. These preset conditions may include, for example, samples where data for all detection dimensions is available, or data coverage for each detection dimension in the calibration sample set is not lower than a preset coverage threshold. The calibration sample set can consist of real user samples with complete data for each detection dimension, or it can include synthetic samples generated from real user samples, partially complete samples after missing data completion, or samples corresponding to estimated relationships that have passed verification in this business domain after calibration using complete data from adjacent business domains through transfer learning. As long as the preset calibration integrity conditions are met, the sample set can participate in offline calibration.

[0053] In one embodiment, the gated output of the detection results carries a list of dimensions used, a list of compensated dimensions, and a version identifier for the estimation relationship of missing dimensions. The list of dimensions used records the detection dimensions corresponding to the measured scores participating in the weighted generalization, and the list of compensated dimensions records the detection dimensions that were compensated. Downstream intervention processes determine the dimensional composition of the detection results based on the list of dimensions used and the list of compensated dimensions, and locate and calculate the version of the estimation relationship used based on the version identifier, facilitating result verification and version traceability.

[0054] In one embodiment, after the gating output, the server writes the sentiment dependency risk value, confidence level, list of dimensions used, list of compensated dimensions, version identifier of the missing dimension estimation relationship, and alarm status into the detection record for use by downstream intervention and auditing processes. By retaining the dimensional composition and confidence level of the detection results, downstream processes can understand which dimensions the detection results are based on, and the auditing process can review the data usage during the detection process. The detection record can serve as a compliance event record, written into the forward hash chain of the trusted interaction record retention system, and available for limited regulatory review, making the quality of fulfillment of detection obligations verifiable during regulatory review.

[0055] In one embodiment, an evaluation method for assessing the ability to detect affective dependence risk is also provided: a synthetic behavioral sequence carrying known affective dependence risk labels is constructed; the synthetic behavioral sequence is injected into the detection system according to the availability combinations of each detection dimension; the detection rate and false alarm rate under each availability combination are statistically analyzed; and the impact of dimension missing compensation on detection quality is evaluated. The evaluation results are used to determine the values ​​of preset quantitative conditions, preset confidence thresholds, and the upper limit of total error normalization. This evaluation method is independent of the online detection implementation and can also be separated as an independent application topic.

[0056] It should be understood that, although Figure 2 The steps in the flowchart are shown sequentially as indicated by the arrows, but these steps are not necessarily executed in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order in which these steps are executed, and they can be performed in other orders. Figure 2 At least some of the steps in the process may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed in turn or alternately with other steps or at least some of the sub-steps or stages of other steps.

[0057] In one embodiment, such as Figure 3As shown, an emotion dependence risk detection device based on dimensional missing compensation is provided, including: an availability determination module 302, a measured scoring module 304, a compensation estimation module 306, a comprehensive calculation module 308, and a gated output module 310, wherein: The availability determination module 302 is used to obtain the data availability determination results for each detection dimension in the preset dependency detection dimension set, and determine the available dimension subset and the missing dimension set based on the data availability determination results; wherein, the dependency detection dimension set includes multiple detection dimensions that characterize the user's emotional dependency risk; The measured scoring module 304 is used to calculate the measured score of each detection dimension according to the scorer corresponding to each detection dimension in the available dimension subset when the number of detection dimensions in the available dimension subset meets the preset number condition. The compensation estimation module 306 is used to estimate the compensation score and corresponding estimation error of each detection dimension in the missing dimension set based on the pre-calibrated missing dimension estimation relationship and each measured score; wherein, the missing dimension estimation relationship represents the mapping relationship between the scores of some detection dimensions and the estimation error of the mapping relationship, which is obtained by offline calibration of a calibration sample set that meets the preset calibration integrity conditions. The comprehensive calculation module 308 is used to perform weighted comprehensive calculation on each measured score and each compensated score to obtain the emotional dependence risk value, and to determine the confidence level corresponding to the emotional dependence risk value based on the weight coverage of the measured scores in the weighted comprehensive calculation and each estimation error. The gating output module 310 is used to gating the output of the emotional dependence risk value based on the confidence level; when the confidence level is lower than the preset confidence threshold, a prompt message is output and supplementary data collection for the missing dimension set is triggered, and intervention measures are prohibited based on the emotional dependence risk value.

[0058] Specific limitations regarding the dimensionality missing compensation-based affective dependency risk detection device can be found in the limitations of the dimensionality missing compensation-based affective dependency risk detection method described above, and will not be repeated here. Each module in the aforementioned dimensionality missing compensation-based affective dependency risk detection device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0059] In one embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 4As shown, the computer device includes a processor, memory, and a network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with external terminals via a network connection. When the computer program is executed by the processor, it implements an emotion dependency risk detection method based on dimensionality missing compensation.

[0060] Those skilled in the art will understand that Figure 4 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0061] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, the processor executing the computer program to perform the following steps: Obtain the data availability assessment results for each detection dimension in the preset dependency detection dimension set, and determine the available dimension subset and the missing dimension set based on the data availability assessment results; wherein, the dependency detection dimension set includes multiple detection dimensions that characterize the user's emotional dependency risk; When the number of detection dimensions in the available dimension subset meets the preset quantity condition, the measured score of each detection dimension is calculated according to the scorer corresponding to each detection dimension in the available dimension subset. Based on the pre-calibrated missing dimension estimation relationship and each measured score, the compensation score and corresponding estimation error of each detection dimension in the missing dimension set are estimated; wherein, the missing dimension estimation relationship represents the mapping relationship between the scores of some detection dimensions and the estimation error of the mapping relationship, which is obtained by offline calibration of a calibration sample set that meets the pre-set calibration integrity conditions; The measured scores and the compensated scores are weighted and synthesized to obtain the emotional dependence risk value. The confidence level corresponding to the emotional dependence risk value is determined based on the weight coverage of the measured scores in the weighted synthesis and the estimation errors. The emotional dependence risk value is gated and output based on confidence level; when the confidence level is lower than the preset confidence threshold, a prompt message is output and supplementary data collection for the missing dimension set is triggered, and intervention measures are prohibited based on the emotional dependence risk value.

[0062] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method for detecting emotion dependency risk based on dimensionality missing compensation.

[0063] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above method embodiments.

[0064] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory, programmable read-only memory, electrically programmable read-only memory, electrically erasable programmable read-only memory, or flash memory. Volatile memory can include random access memory or external cache memory.

[0065] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0066] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A method for detecting the risk of emotional dependence based on dimensional missing compensation, characterized in that, The method includes: Obtain the data availability determination results for each detection dimension in a preset dependency detection dimension set, and determine the available dimension subset and the missing dimension set based on the data availability determination results; wherein, the dependency detection dimension set includes multiple detection dimensions that characterize the user's emotional dependency risk; When the number of detection dimensions in the available dimension subset meets the preset quantity condition, the measured score of each detection dimension is calculated according to the scorer corresponding to each detection dimension in the available dimension subset. Based on the pre-calibrated missing dimension estimation relationship and the measured scores, the compensation score and corresponding estimation error of each detection dimension in the missing dimension set are estimated; wherein, the missing dimension estimation relationship represents the mapping relationship between the scores of some detection dimensions and the estimation error of the mapping relationship, which is obtained by offline calibration of a calibration sample set that meets the preset calibration integrity conditions; The measured scores and the compensated scores are weighted and synthesized to obtain the emotional dependence risk value. The confidence level corresponding to the emotional dependence risk value is determined based on the weight coverage of the measured scores in the weighted synthesis and the estimation error of each score. The emotional dependence risk value is gated and output based on the confidence level; wherein, when the confidence level is lower than a preset confidence threshold, a prompt message is output and supplementary data collection for the missing dimension set is triggered, and intervention measures are prohibited based on the emotional dependence risk value.

2. The method according to claim 1, characterized in that, Obtain the data availability determination results for each detection dimension in the preset dependency detection dimension set, including: For each detection dimension in the set of dependent detection dimensions, the existence of the corresponding data source, whether the data sample size meets the standard, and whether the data authorization is valid are detected respectively. When the data source corresponding to the detection dimension exists, the data sample size meets the standard, and the data authorization is valid, it is determined that the data for the detection dimension is available, and the detection dimension is assigned to the available dimension subset; otherwise, the detection dimension is assigned to the missing dimension set. Each detection dimension in the dependency detection dimension set records a data requirement description and a scorer identifier. The data requirement description is used to determine data availability, and the scorer identifier is used to determine the corresponding scorer. Triggering supplementary data collection for the missing dimension set includes: generating supplementary collection suggestions based on the data requirement descriptions of each detection dimension in the missing dimension set; performing supplementary data collection according to the supplementary collection suggestions; re-acquiring the data availability determination results of each detection dimension; and recalculating the emotional dependence risk value and the confidence level based on the newly determined available dimension subset and missing dimension set. The supplementary data collection recommendations include at least one of requesting data authorization and extending the observation period.

3. The method according to claim 1, characterized in that, The preset quantity condition is that the available dimension subset contains no fewer than two detection dimensions; When the number of detectable dimensions in the available dimension subset does not meet the preset quantity condition, an undetectable state and a list of missing data are output, and the emotional dependence risk value is not generated. The dependency detection dimension set includes at least three of the following: interaction intensity dimension, emotional substitution dimension, vulnerability moment dependency dimension, withdrawal reaction dimension, and time period offset dimension.

4. The method according to claim 1, characterized in that, Based on the weight coverage of the measured scores in the weighted aggregation and the estimation errors, the confidence level corresponding to the sentiment dependence risk value is determined, including: The weight coverage is determined by summing the weights of the measured scores of each detection dimension in the available dimension subset in the weighted summation. The confidence level is determined based on the weight coverage and the estimation error of each of the compensation scores; wherein the confidence level is positively correlated with the weight coverage and negatively correlated with each of the estimation errors; Gated output of the affective dependence risk value based on the confidence level also includes: When the confidence level is not lower than the first preset confidence threshold, the emotional dependence risk value is output normally; wherein, the normally output emotional dependence risk value is used to trigger graded intervention measures; When the confidence level is lower than the first preset confidence threshold but not lower than the preset confidence threshold, the emotional dependence risk value is marked as a medium confidence result, and the level of the intervention measures triggered based on the emotional dependence risk value is reduced.

5. The method according to claim 1, characterized in that, The method further includes: when the measured score of any detection dimension in the available dimension subset hits the preset extreme risk rule, it is not subject to the gating output restriction of the confidence level on the emotional dependence risk value, and an emergency handling process is triggered. The emergency response process includes at least one of manual review and immediate security intervention.

6. The method according to claim 1, characterized in that, The detection output of the gated output carries a list of dimensions used, a list of dimensions to be compensated, and a version identifier of the missing dimension estimation relationship; Wherein, the list of dimensions used is a list of detection dimensions corresponding to the measured scores participating in the weighted generalization, and the list of compensated dimensions is a list of detection dimensions that are compensated. The method further includes: writing the emotional dependence risk value, the confidence level, the list of dimensions used, the list of compensated dimensions, the version identifier of the missing dimension estimation relationship, and the alarm status into the detection record; The aforementioned testing records are retained as compliance event records and are available for review by regulators with limited access.

7. The method according to claim 1, characterized in that, The missing dimension estimation relationship is a dimension correlation matrix; Wherein, each matrix element in the dimensional correlation matrix includes the estimated coefficient and estimation error of the score of the first detection dimension on the score of the second detection dimension; the first detection dimension and the second detection dimension are different detection dimensions in the dependent detection dimension set; The dimensional correlation matrix records the calibration sample size and calibration time, and is stratified according to user groups.

8. The method according to claim 7, characterized in that, The preset calibration integrity conditions include: the calibration sample set contains samples in which data for each detection dimension in the dependent detection dimension set is available, or the data coverage of each detection dimension in the calibration sample set is not lower than a preset coverage threshold. The calibration sample set includes at least one of real user samples, synthetic samples generated from real user samples, and partially complete samples that have undergone missing data completion. The dimension correlation matrix is ​​calibrated offline in the following way: regression calibration is performed periodically on user groups where data for each detection dimension in the dependent detection dimension set is available, to obtain a new version of the dimension correlation matrix; The new version of the dimensional correlation matrix is ​​verified offline, and after the verification is passed, the online dimensional correlation matrix is ​​replaced with the new version of the dimensional correlation matrix in grayscale. The offline calibration uses only the scoring layer data of each detection dimension and does not use the original interactive content.

9. A device for detecting the risk of emotional dependence based on dimensional missing compensation, characterized in that, The device includes: The availability determination module is used to obtain the data availability determination results for each detection dimension in a preset dependency detection dimension set, and determine the available dimension subset and the missing dimension set based on the data availability determination results; wherein, the dependency detection dimension set includes multiple detection dimensions that characterize the user's emotional dependency risk; The measured scoring module is used to calculate the measured score of each detection dimension according to the scorer corresponding to each detection dimension in the available dimension subset when the number of detection dimensions in the available dimension subset meets the preset number condition. The compensation estimation module is used to estimate the compensation score and corresponding estimation error of each detection dimension in the missing dimension set based on the pre-calibrated missing dimension estimation relationship and each of the measured scores; wherein, the missing dimension estimation relationship represents the mapping relationship between the scores of some detection dimensions and the estimation error of the mapping relationship, which is obtained by offline calibration of a calibration sample set that meets the preset calibration integrity conditions; The comprehensive calculation module is used to perform weighted comprehensive calculation on each of the measured scores and each of the compensated scores to obtain the emotional dependence risk value, and to determine the confidence level corresponding to the emotional dependence risk value based on the weight coverage of the measured scores in the weighted comprehensive calculation and each of the estimation errors. The gating output module is used to gating the output of the emotional dependence risk value based on the confidence level; wherein, when the confidence level is lower than a preset confidence threshold, a prompt message is output and supplementary data collection for the missing dimension set is triggered, and intervention measures are prohibited based on the emotional dependence risk value.

10. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.