Multi-dimensional evaluation reliability aggregation calculation method fusing Kendall harmony coefficient

Abstract

The invention relates to the technical field of evaluation reliability calculation, in particular to a multi-dimensional evaluation reliability aggregation calculation method fusing a Kendall harmony coefficient, and the method comprises the following steps: collecting multi-source evaluation information, arranging the multi-source evaluation information into an evaluation order sequence in order to recognize evaluation behavior characteristics, constructing an independent time window according to the total number of evaluation objects, and calculating the evaluation reliability according to the independent time window; calculating a Kendall harmony coefficient to quantify evaluation consistency, constructing a three-dimensional interaction tensor based on evaluation behavior characteristics and the Kendall coefficient, identifying an initial credibility degree through tensor decomposition, calculating evaluation sorting dispersity, constructing a credibility decline relation, correcting and resetting the initial credibility degree, generating an evaluation credibility index, and performing evaluation on the evaluation credibility index. Establishing a credibility verification closed-loop mechanism according to the index, and generating a credibility evaluation report; according to the invention, a more comprehensive and more accurate multi-dimensional evaluation reliability aggregation calculation method is realized.

Description

Technical Field

[0001] This invention relates to the field of assessment reliability calculation technology, and in particular to a multi-dimensional assessment reliability aggregation calculation method that integrates Kendall's coefficient of harmony. Background Technology

[0002] In current methods for calculating the reliability of assessments, traditional techniques often rely on information from a single source, which often fails to fully reflect the true situation of the assessment object, leading to inaccuracies and insufficient credibility of the assessment results. Especially when faced with multi-dimensional and multi-source assessment information, existing methods lack effective fusion mechanisms and struggle to handle the complexity between information, resulting in one-sidedness and lack of credibility of the assessment results, which affects the scientificity and reliability of decision-making.

[0003] Furthermore, existing audit reliability calculation methods often focus on the impact of time factors when processing behavioral characteristics, failing to effectively establish dynamic audit models. This makes them unable to adapt to rapidly changing environments in practical applications. In particular, during the evaluation of audit targets, there is a lack of sufficient consideration of the dispersion of audit ranking, which increases the bias in the reliability assessment results and ultimately fails to provide users with accurate reference relevance, thus affecting the effectiveness of the audit.

[0004] In practical applications, traditional methods often lack systematicity and comprehensiveness when faced with the aggregation of multi-source audit information, making it impossible to establish an effective credibility verification mechanism. This weakens the application value of audit results, especially in complex decision-making fields where the credibility of information is directly related to the accuracy and effectiveness of decisions. Summary of the Invention

[0005] To address the aforementioned technical problems, this invention proposes a multi-dimensional reliability aggregation calculation method that integrates Kendall's concordance coefficient, thereby resolving at least one of the aforementioned technical issues.

[0006] To achieve the above objectives, a multi-dimensional reliability aggregation calculation method integrating Kendall's concordance coefficient is proposed, including the following steps: Step S1: Collect multi-source assessment information, arrange it in order to form an assessment sequence, and identify the assessment behavior characteristics of the assessment sequence. Step S2: Construct an independent time window based on the total number of evaluation objects from multiple sources of evaluation information, and calculate the Kendall's coefficient of harmony for the evaluation objects within the independent time window; Step S3: Construct a three-dimensional interaction tensor based on the assessment behavior characteristics and Kendall's harmony coefficient, and identify the initial credibility of the assessment information through three-dimensional interaction tensor decomposition; Step S4: Calculate the evaluation ranking dispersion of multi-source evaluation information, and construct a decreasing credibility relationship through the evaluation ranking dispersion; Step S5: Based on the decreasing credibility relationship, perform impact correction on the initial credibility level and reset the credibility index to generate the evaluation credibility index; Step S6: Establish a credibility verification closed-loop mechanism based on the credibility index and generate a credibility evaluation report.

[0007] This invention ensures the comprehensiveness and diversity of data by collecting multi-source evaluation information and identifying evaluation behavior characteristics, thereby enhancing the accuracy and representativeness of the evaluation. The construction of independent time windows enables effective monitoring of the dynamic changes of the evaluation objects. The calculated Kendall's harmony coefficient provides a quantitative basis for subsequent analysis. The construction of a three-dimensional interaction tensor fully reveals the relationships between different evaluation dimensions. Tensor decomposition technology can identify the initial credibility of evaluation information, effectively improving the reliability of evaluation results. In the calculation of ranking dispersion, by analyzing the dispersion of multi-source information, the consistency and stability of evaluation results are revealed. The constructed credibility decreasing relationship provides a basis for the dynamic adjustment of credibility. This provides a scientific basis for the implementation of correction and credibility index reset, optimizing the initial credibility level. The resulting evaluation credibility index provides a more accurate reference for decision-making. The established credibility verification closed-loop mechanism ensures the continuous validity and updating of evaluation results, promotes the effective integration and utilization of multi-source evaluation information, enhances the scientific and systematic nature of the evaluation process, adapts to the multi-dimensional analysis needs in complex environments, provides a more reliable basis for decision-making in various fields, promotes the development of intelligent decision support systems, improves the flexibility and efficiency of information processing, and ultimately achieves a comprehensive improvement and optimization of evaluation credibility, promoting intelligent and refined management of data-driven decision-making. Attached Figure Description

[0008] Figure 1 This is a flowchart illustrating the steps of a multi-dimensional assessment reliability aggregation calculation method that integrates Kendall's concordance coefficient according to the present invention. Figure 2 This is a detailed flowchart illustrating the implementation steps of step S1. Figure 3 This is a flowchart illustrating the detailed implementation steps of step S2. Detailed Implementation

[0009] It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the scope of the invention.

[0010] This application provides a method for calculating the reliability aggregation of multi-dimensional assessments by incorporating Kendall's concordance coefficients. The execution entities of this method include, but are not limited to, mechanical equipment, data processing platforms, cloud server nodes, and network upload devices that can be considered as general computing nodes in this application. The data processing platform includes, but is not limited to, at least one of an audio / image management system, an information management system, and a cloud-based data management system.

[0011] Please see Figures 1 to 3 This invention provides a method for calculating the reliability of multi-dimensional assessments by incorporating Kendall's coefficient of harmony. The method includes the following steps: Step S1: Collect multi-source assessment information, arrange it in order to form an assessment sequence, and identify the assessment behavior characteristics of the assessment sequence. Step S2: Construct an independent time window based on the total number of evaluation objects from multiple sources of evaluation information, and calculate the Kendall's coefficient of harmony for the evaluation objects within the independent time window; Step S3: Construct a three-dimensional interaction tensor based on the assessment behavior characteristics and Kendall's harmony coefficient, and identify the initial credibility of the assessment information through three-dimensional interaction tensor decomposition; Step S4: Calculate the evaluation ranking dispersion of multi-source evaluation information, and construct a decreasing credibility relationship through the evaluation ranking dispersion; Step S5: Based on the decreasing credibility relationship, perform impact correction on the initial credibility level and reset the credibility index to generate the evaluation credibility index; Step S6: Establish a credibility verification closed-loop mechanism based on the credibility index and generate a credibility evaluation report.

[0012] This invention, by real-time acquisition of water flow monitoring parameters, accurately understands the water flow status during the operation of the washing and care equipment, avoiding adverse effects on user experience caused by excessive or insufficient water flow. Based on the needs of different washing and care tasks, the system can adjust the water flow output in real time, achieving flexible water flow distribution and ensuring that the user's desired washing and care mode is accurately executed. By constructing a real-time water flow distribution map at the nozzle, the system can intelligently adjust the water flow output at the nozzle according to the actual situation of the user's head, providing a more personalized washing and care service. Multi-frequency wavelet transform decomposition can accurately capture different frequency characteristics of water flow changes, thereby more accurately describing the instantaneous changes and fluctuations in water flow. This helps to effectively identify minute water flow disturbances and avoid the washing and care effect being affected by uneven water flow. Calculating the water flow output pulse width waveform (i.e., pulse width modulation signal) enables precise control of water flow, ensuring that the water flow and water pressure meet preset requirements in each time period, improving comfort and reducing energy waste. By detecting abnormal fluctuations in the time-series water flow pulse width curve, the system can identify irregular fluctuations or instabilities in the water flow in real time and make timely adjustments to avoid discomfort or damage caused by excessively strong or weak water flow. Different areas of a user's head may perceive water pressure differently. By predicting these differences, the system can adjust the water flow intensity and pressure in different areas to ensure optimal balance of comfort across the entire scalp. Combining water pressure perception data from different scalp areas, the device can automatically adjust the water flow output according to each user's needs, providing a customized washing and care experience to meet diverse requirements. Dynamic pulse width duty cycle adjustment effectively controls the intensity and duration of the water flow, preventing uneven washing and care caused by excessively concentrated or dispersed water flow, ensuring balanced care for the scalp and hair. Asynchronous and coordinated adjustment of multiple nozzles ensures precise water flow distribution to different areas while avoiding interference between nozzles. This effectively improves overall washing and care efficiency and enhances user comfort. Through coordinated adjustment of multiple nozzles, the washing and care device can dynamically adjust the output of each nozzle according to the user's scalp condition, improving the accuracy and comfort of the water flow and preventing uneven water flow caused by excessively high or low local water pressure. By simulating and analyzing the water flow pulse width curve, the system can predict potential water flow instability trends and take preventative measures to avoid reduced washing and care effects or equipment damage caused by excessive water flow fluctuations. By predicting water flow instability, the system can optimize the water flow pulse width in real time, ensuring stable water flow, reducing system instability factors, and improving equipment reliability and long-term operational stability. Based on the predicted water flow instability, the system can pre-adjust pulse width parameters to prevent sudden water flow instability and ensure that water flow quality remains at its optimal level throughout the washing and care process. During pulse width adjustment, the system calculates current compensation values ​​in real time based on current monitoring data, ensuring that current fluctuations are effectively controlled while adjusting the water flow, preventing overload or malfunctions caused by current changes.Dynamic current compensation not only helps maintain stable system operation but also optimizes energy efficiency, reduces unnecessary energy consumption, and improves the economic and environmental performance of the equipment. Through intelligent pulse width modulation, the equipment can automatically adjust the pulse width of the water flow output to adapt to different user needs without manual intervention, thus enhancing the intelligence level of the washing and care equipment.

[0013] In the embodiments of the present invention, see Figure 1 This is a flowchart illustrating the steps of a multi-dimensional assessment reliability aggregation calculation method incorporating Kendall's concordance coefficient according to the present invention. In this example, the steps of the multi-dimensional assessment reliability aggregation calculation method incorporating Kendall's concordance coefficient include: Step S1: Collect multi-source assessment information, arrange it in order to form an assessment sequence, and identify the assessment behavior characteristics of the assessment sequence. In this embodiment, the multi-source evaluation information includes scoring records, text feedback, behavior logs, and timestamp data from different evaluation channels. This data can comprehensively reflect the performance of the evaluation object in different dimensions. Evaluation information is collected synchronously from multiple independent evaluation platforms (such as user evaluation systems, expert scoring systems, and automatic quality detection modules). The collected data is encoded in a unified format to form an original evaluation dataset containing the evaluation object identifier, evaluation time, evaluation dimension, and evaluation score. Then, all evaluation information is sorted by time according to the evaluation time field to establish an evaluation order sequence. The sorting rule is set to ascending order to ensure the temporal continuity of the evaluation order. Next, behavioral features were identified in the assessment sequence. Statistical analysis was used to extract the assessment frequency, score fluctuation range, scoring interval, scoring trend slope, and time difference stability for each assessment subject. Assuming that in a sample assessment subject, the continuous assessment time interval is [5min, 7min, 6min, 8min], the average interval is 6.5min, the score fluctuation range is [3.8, 4.5, 4.2, 4.7], and the fluctuation range is 0.9. Based on these features, a clustering algorithm was used to classify the assessment behavior into three categories: "stable," "fluctuating," and "abnormal," and the classification results were recorded.

[0014] Step S2: Construct an independent time window based on the total number of evaluation objects from multiple sources of evaluation information, and calculate the Kendall's coefficient of harmony for the evaluation objects within the independent time window; In this embodiment, Kendall's coefficient of harmony is used to measure the consistency of evaluations of the same assessment object by different evaluators within the same time window. First, the total number of assessment objects in the multi-source assessment information is counted, assuming the number of objects is N=120. The time window width is adaptively divided according to the object size parameter, with the number of windows set to W, and each time window containing a fixed number of assessment samples. For example, if the window width... Then, the ratings of all subjects are tallied every 10 minutes. Then, within each independent time window, the rating matrix R={r_ij} of all subjects is extracted, where r_ij represents the rating of the i-th evaluator for the j-th subject. Next, the Kendall's coefficient of harmony formula is used: Where S represents the ratings and the squared difference, m represents the number of evaluators, and n represents the number of evaluation subjects. The harmony coefficient W for each time window is calculated by examining the R matrix for each time window. Assuming that in window T1, m=5 and n=10, S is calculated to be 145, resulting in W=0.87, indicating high consistency; while in window T2, W=0.52, indicating a difference in evaluations. The W values ​​from all windows are combined into a time series W(t) and plotted graphically to show the trend of consistency over time.

[0015] Step S3: Construct a three-dimensional interaction tensor based on the assessment behavior characteristics and Kendall's harmony coefficient, and identify the initial credibility of the assessment information through three-dimensional interaction tensor decomposition; In this embodiment, the evaluation behavior characteristics and Kendall's harmony coefficients for each time window are input into the three-dimensional interactive tensor construction module. The tensor dimensions are defined as source dimension, time dimension, and object dimension. The source dimension corresponds to three types of evaluation sources: experts, users, and systems. The time dimension corresponds to four time windows, and the object dimension corresponds to 20 evaluation objects. The input data is organized in the form of matrix blocks. Assuming that the constructed core tensor units T(expert,W1,A1)=0.85, T(user,W1,A1)=0.72, and T(system,W1,A1)=0.80, and so on, all units are filled. The system uses the principal component projection method to interactively decompose the three-dimensional tensor, mapping the time dimension components to dynamic consistency factors, the source dimension to behavioral stability factors, and the object dimension to rating deviation factors. After three-dimensional decomposition, the initial credibility dataset is obtained. Assuming that the mean contribution of the expert dimension is 0.83, the user dimension is 0.76, and the system dimension is 0.88, the initial credibility after comprehensive weighting is 0.82.

[0016] Step S4: Calculate the evaluation ranking dispersion of multi-source evaluation information, and construct a decreasing credibility relationship through the evaluation ranking dispersion; In this embodiment, the evaluation ranking dispersion is used to measure the degree of difference in the ranking results among different sources. The system standardizes and ranks the ranking results of each source. For example, the expert rating ranking is [2,1,3,5,4], the user rating ranking is [1,3,2,4,5], and the system rating ranking is [2,1,4,3,5]. The average rank deviation value is calculated by accumulating the differences in each ranking. The deviation threshold range is set from 0 to 5. The larger the range, the higher the dispersion. Assuming that the calculated dispersion between experts and users is 2.5, between experts and the system is 1.8, and between users and the system is 3.0, the overall average dispersion value is 2.43. Based on this dispersion, a credibility decreasing relationship curve is constructed. When the dispersion exceeds 2.0, the linear decreasing rate of credibility is set to 0.05 times, and the final decreasing parameter λ=0.12 is obtained, corresponding to a credibility decrease range of [0.82→0.70], forming a credibility decreasing relationship table.

[0017] Step S5: Based on the decreasing credibility relationship, perform impact correction on the initial credibility level and reset the credibility index to generate the evaluation credibility index; In this embodiment, the initial credibility level of each source is corrected according to the decreasing relationship table. The adjustment is made by combining the behavior type and dispersion weight. It is assumed that the initial credibility level of the expert source is 0.83, with a corresponding dispersion weight of 0.9; the initial credibility level of the user source is 0.76, with a corresponding weight of 0.8; and the initial credibility level of the system source is 0.88, with a corresponding weight of 0.95. By superimposing the decreasing parameter λ=0.12 with each weight value, the corrected credibility indices are obtained as follows: 0.81 for experts, 0.72 for users, and 0.85 for systems. The distribution of the corrected indices is recorded to form a credibility matrix, which reflects the distribution of trust contribution of different sources throughout the entire period. A time series trend chart is generated to show the fluctuation range of credibility over time. It is assumed that the overall credibility average value remains between 0.79 and 0.83 during the evaluation period, with a fluctuation range of ±0.02.

[0018] Step S6: Establish a credibility verification closed-loop mechanism based on the credibility index and generate a credibility evaluation report.

[0019] In this embodiment, the generated evaluation credibility index is sampled and verified. Evaluation data from three time windows are randomly selected for review, such as windows W1, W3, and W4. The verification results show that the credibility fluctuation trend is 93% consistent with the historical score changes, meeting the closed-loop verification threshold requirements. Based on this, a real-time feedback module is established. When a credibility decrease of more than 0.05 is detected, a re-evaluation mechanism is automatically triggered to reassess the behavioral data for that time period and update the index results. All processes are recorded in the credibility evaluation log file. Finally, the closed-loop correction results are summarized to generate a credibility evaluation report. The report includes a credibility index distribution table for each time window, a behavioral characteristic statistics table, a dispersion change curve, and a credibility correction details table, ensuring that the evaluation results are statistically stable and traceable.

[0020] In this embodiment, see Figure 2 The diagram below illustrates the detailed implementation steps of step S1. In this embodiment, the specific steps of step S1 are as follows: Collect multi-source evaluation information; Time-series synchronization processing is performed on multi-source evaluation information to obtain time-aligned evaluation information; Arrange the time-aligned assessment information in order to form an assessment sequence; The score difference is calculated for each adjacent sequence in the evaluation sequence, and the score difference is mapped to the local volatility of the sequence score. Extreme ranking frequencies in the evaluation sequence are eliminated by local volatility, and evaluation behavior characteristics in the evaluation sequence are identified.

[0021] In this embodiment, multi-source assessment information is collected, which includes multi-dimensional evaluation data from different channels. These include real-time scoring records from online assessment systems, quantitative results data provided by third-party assessment agencies, and textual scoring results from historical questionnaire databases. To ensure comprehensive sample coverage, the data collection period is set to 30 consecutive days, with daily collection time from 08:00 to 22:00, collected once per hour, and a minimum of 1000 data entries collected per collection. A unified data interface is used to automatically identify and standardize the data format, converting the original assessment data into a structured table format. The fields include assessment object number, assessment timestamp, score value, assessment source identifier, and remarks field, forming a preliminary multi-source assessment dataset.

[0022] The collected multi-source evaluation information is processed for time series synchronization. To address the differences in time distribution of evaluation data from different sources, a synchronization strategy based on a unified reference time axis is adopted. The hourly time is used as the alignment reference, and linear interpolation and time completion algorithms are used to correct the inconsistency of sampling intervals. For example, when a source is missing data between 10:00 and 10:30, the score values ​​of adjacent times are used for interpolation to keep the continuity of the time series within the error range of ±2 minutes. After synchronization processing, time-aligned evaluation information is formed.

[0023] After time synchronization is completed, the time-aligned assessment information is sorted according to the assessment time order to establish an assessment sequence. Each sequence records an assessment event node. The node content includes the object number, timestamp, and score value. The sequence is arranged in ascending order of time as {E1,E2,...,En}. For example, E1 represents the score corresponding to the first assessment time of 10:00, E2 represents the score of 10:05, and so on. The sequence interval is uniformly 5 minutes.

[0024] The difference between the scores of every two adjacent nodes in the evaluation sequence is calculated to obtain the score difference Δscore. This difference is then mapped to the local volatility of the sequence score to reflect the degree of fluctuation of the score in a short period of time. In specific operation, the score difference interval [-5,5] is linearly mapped to the volatility interval [0,1]. For example, when the score difference is 2, the corresponding volatility is 0.4, and when the score difference is -3, the corresponding volatility is 0.6. In this way, a local volatility sequence is constructed, and the volatility state at each time point is recorded to evaluate the stability and trend of the evaluation object in the time series.

[0025] Based on local volatility, extreme ranking frequencies in the evaluation sequence are eliminated. A threshold parameter θ is set, and when the local volatility exceeds 0.9 three times consecutively, it is determined to be an extreme volatility range. A sliding window algorithm is used for smoothing correction to eliminate the influence of abnormal high-frequency changes and retain the data part that reflects the true evaluation dynamics. The smoothed sequence is used to identify evaluation behavior characteristics. By analyzing the volatility distribution characteristics, key feature parameters such as evaluation stability, mean response delay, and preference directionality are extracted. For example, in a certain test, the average volatility was 0.32, the mean response delay was 4.8 minutes, and the preference directionality was positive 0.21, indicating that the overall evaluation behavior was stable and had a slight positive tendency, thus obtaining the final evaluation behavior feature dataset.

[0026] In this embodiment, step S2 specifically includes: Collect the identification of the evaluation object from multi-source evaluation information; count the total number of evaluation objects based on the evaluation object identification; The time window width is matched and calculated based on the total number of evaluation subjects to determine the time span threshold of the independent time window; Multiple independent time windows are constructed for multi-source evaluation information by using the time span threshold of independent time windows; For each independent time window, a pairwise sorting consistency comparison is performed on the evaluation order sequence, and the rank correlation coefficient between the sorted objects is calculated. Kendall's harmonic coefficient transformation is performed on the rank correlation coefficients between the sorted objects to obtain the Kendall's harmonic factor of the evaluated objects within an independent time window. Kendall's harmony factors for all evaluation objects within all independent time windows are arranged in a time series to generate Kendall's harmony coefficients.

[0027] In this embodiment, the evaluation object identifiers are collected from multi-source evaluation information. The evaluation object identifier is used to uniquely correspond to the individual or entity participating in the evaluation. It includes the user number in the online evaluation system, the evaluation sample number of the third-party organization, and the respondent code in the questionnaire survey. To ensure the uniqueness and traceability of the identifier, all identifiers are uniformly formatted as a 16-character combination. The first four characters are the source code, the middle eight characters are the timestamp number, and the last four characters are a random check code. For example, the source code 0001 represents the internal evaluation system of the enterprise, and the source code 0003 represents the external evaluation platform. The collected results are uniformly written into the standardized field object_id field set to form the evaluation object identifier dataset.

[0028] The total number of objects is counted based on the evaluation object identifiers. All identifiers are grouped by source category and duplicates are removed. A deduplication threshold rule is used, and when the same identifier appears in different sources, the earliest time is taken as the main one. After the statistics, the total number of evaluation objects N_total is obtained. For example, if there are 3200 identifiers in the internal evaluation records of an enterprise and 2800 identifiers in the external evaluation platform records, of which 400 are duplicate identifiers, then the total number of objects is calculated to be 5600. This result is used as the basis parameter for subsequent time window width matching calculation.

[0029] The time window width is calculated based on the total number of evaluation subjects, and an initial time window span baseline value is set. Calculate the matching ratio within a 5-minute time window. When the total number of objects exceeds 5000, the time window width is automatically expanded. The calculated time window width is as follows: For example, when N_total is 5600, the calculated time window span threshold is 28 minutes, meaning that each independent time window covers a 28-minute interval of evaluation data.

[0030] After obtaining the time span threshold, multiple independent time windows are constructed for multi-source evaluation information based on this threshold. The evaluation data for the entire period is segmented in chronological order, with each segment being 28 minutes long. Window overlap is allowed by 5 minutes to eliminate boundary effects. For example, evaluation data for two consecutive hours can be divided into 9 independent time windows. Each window contains an average of about 600 records. Each record has a unified timestamp and score value field, forming an independent time window dataset.

[0031] For each independent time window, a pairwise sorting consistency comparison is performed on the evaluation sequence. Evaluation objects within the same time period within the window are arranged in descending order of their scores, forming two sorted lists, List_A and List_B, from different sources. The rank difference statistic is calculated by comparing the pairwise order consistency between the two lists. If two objects are in the same order in both lists, it is counted as a match pair (1); otherwise, it is counted as a match pair (-1). For example, if there are 100 evaluation objects in window T1, and 82 pairs are in the same order, the rank correlation ratio is 0.82. The rank correlation coefficient within the window is then obtained accordingly. .

[0032] The rank correlation coefficients among the sorted objects are transformed using Kendall's coordination coefficient, and a linear normalization method based on the rank difference distribution is employed. Mapped to the interval [0,1], where 0 represents complete inconsistency and 1 represents complete consistency, the transformed result is defined as the Kendall harmony factor. Each independent time window generates a Value, for example when When =0.82, we get =0.87.

[0033] Arrange the Kendall harmony factors of all evaluation objects within the independent time window according to the time series to generate a time series trajectory sequence of Kendall harmony coefficients, with each time point corresponding to one. This value allows us to observe the trend of consistency in the evaluation process over time, for example, within a two-hour evaluation cycle. The mean is 0.76, and the peak occurs in the sixth window. =0.91, the lowest value appears in the ninth window. =0.68, reflecting that the evaluation behavior tends to stabilize in the middle and fluctuates in the end, thus obtaining the final Kendall harmony coefficient sequence dataset.

[0034] In this embodiment, the specific steps of step S3 are as follows: Extract behavioral patterns and behavioral stability parameters from the behavioral characteristics assessed. Calculate the mean and variance of the ranking position of each reviewed object based on the review ranking sequence; A three-dimensional interaction tensor is generated by constructing tensors using behavioral stability parameters, Kendall's harmony coefficient, and ordination variance as three independent dimensions. The coupling degree of behavior-harmony-ordering is calculated based on the three-dimensional interaction tensor, and the coupling degree is mapped to the interaction influence coefficient. The credibility of multi-source evaluation information is mapped based on the interaction influence coefficient to obtain the initial credibility level.

[0035] In this embodiment, behavioral patterns and behavioral stability parameters are extracted from the evaluation behavioral features. The behavioral patterns are used to characterize the repetitive behavioral trends and response patterns of the evaluation object in multiple evaluation processes. For example, in three periodic evaluations, the increase, decrease, or stable change of the same object's score are all summarized as pattern features. The behavioral stability parameters are used to quantify the fluctuation range and continuity of the pattern. By calculating the score difference of the continuous evaluation data sequence, the score change range is set to [-5, 5] points. When the absolute value of the score difference of three consecutive evaluations does not exceed 1 point, it is defined as high-stability behavior. When the difference exceeds 3 points, it is defined as low-stability behavior. The pattern encoding and stability parameters are stored together in the behavior_feature field to form a structured behavioral feature dataset.

[0036] The mean and variance of the ranking of each reviewed object are calculated based on the review ranking sequence. All evaluation objects are ranked and statistically analyzed according to the review rounds divided by time windows. The ranking of each round is in the range of [1, N], where N is the total number of objects. For example, when N=200, if object A's ranking in the five rounds of evaluation is 15, 19, 22, 17, and 18, then the mean ranking is 18.2 and the variance ranking is 6.3. The mean can reflect its central trend in the overall ranking, and the variance can reflect the ranking volatility. The mean and variance results of all objects are summarized to generate the ranking_stat dataset.

[0037] A three-dimensional interaction tensor is generated by constructing a tensor with behavioral stability parameter, Kendall's harmony coefficient, and ranking variance as three independent dimensions. The three axes of the three-dimensional interaction tensor correspond to behavioral stability s, harmony τ, and volatility σ, respectively. A three-dimensional matrix structure is formed by mapping the coordinates of the three. The matrix unit value corresponds to the feature point value of a specific evaluation object in the three-dimensional space. For example, when an object has a stability parameter of 0.83, a Kendall harmony coefficient of 0.79, and a ranking variance of 5.1, its coordinate point in the three-dimensional tensor is (0.83, 0.79, 5.1). This structure is used to express the interactive distribution relationship between behavioral features.

[0038] Based on the coupling degree of the three-dimensional interaction tensor computation behavior – harmony – sorting, the degree of interaction and aggregation of different object features is reflected by analyzing the Euclidean distance distribution and density change rate between adjacent feature points in the three-dimensional tensor. When the feature points of multiple objects are concentrated and the average distance is less than 1.5, it is defined as a high coupling region, and when the average distance exceeds 3, it is defined as a low coupling region. In the calculation process, the average distance of the window is used as the coupling degree index. For example, if there are 150 object samples in the window and the average distance is 1.2, it is judged as a high coupling level. The coupling result is mapped to the interaction influence coefficient. The interaction influence coefficient takes values ​​in the range of [0,1]. The larger the value, the stronger the consistency and stability among the object groups.

[0039] The credibility of multi-source evaluation information is mapped based on the interaction influence coefficient, and a linear correspondence table between the interaction influence coefficient and the credibility is established. When the interaction influence coefficient is greater than 0.85, a high credibility value (between 0.9 and 1) is assigned. When the interaction influence coefficient is between 0.6 and 0.85, it is defined as a medium credibility level. When it is less than 0.6, it is defined as a low credibility level. For example, in a comprehensive evaluation, the interaction influence coefficient is 0.78, which corresponds to an initial credibility level of 0.8. The credibility levels of all objects are summarized to form a credibility mapping table.

[0040] In this embodiment, the specific steps of step S4 are as follows: Extract the ranking results of each evaluation source from the multi-source evaluation information; The ranking difference is calculated based on the evaluation ranking results to obtain the ranking deviation. Variance is calculated based on the ranking deviation, and the dispersion of the ranking of multi-source evaluation information is determined by the variance. By evaluating the dispersion of the ranking and the initial confidence level, a correlation analysis was conducted to obtain dispersion-confidence correlation data; Calculate the gradient rate of change for dispersion-confidence correlation data and extract the decreasing gradient coefficient of confidence as dispersion changes. A confidence-decreasing relationship is constructed based on the decreasing gradient coefficients.

[0041] In this embodiment, the evaluation ranking results of each evaluation source in the multi-source evaluation information are extracted. The evaluation sources include the enterprise's internal evaluation system, third-party professional evaluation agencies, and external cooperation platforms. Each evaluation source independently generates a corresponding evaluation result list. Each list contains the evaluation object identifier, evaluation score, and ranking position fields. To ensure the alignment of the ranking results, the evaluation time range of different evaluation sources is first unified to ensure that the same object is comparable within the same period. For example, if the sampling period of the enterprise's internal system is 10 minutes and that of the third-party platform is 15 minutes, then the time step is unified to 10 minutes by linear interpolation. The synchronized evaluation results are written into the ranking_result field to form a multi-source evaluation ranking dataset.

[0042] The ranking difference is calculated based on the evaluation ranking results to obtain the ranking deviation. The ranking values ​​of the same object in different evaluation sources are compared in pairs to calculate their ranking difference. For example, if object A ranks 12 in the internal evaluation and 20 in the third-party platform, the ranking deviation of object A is 8. When object B ranks 5, 6 and 4 in the three evaluation sources respectively, the average deviation is 1.3. The overall ranking deviation dataset ranking_bias is obtained by calculating the ranking deviation of all objects. The larger the deviation value, the higher the degree of inconsistency of the ranking results between different evaluation sources.

[0043] Variance is calculated based on the ranking deviation, and the ranking dispersion of multi-source assessment information is determined by the variance. The ranking deviation values ​​of all objects are aggregated according to the assessment source category, and the variance of each assessment source is calculated to characterize the dispersion of the results from each source. For example, when the variances of the internal and external platforms are 3.8 and 5.1 respectively, it indicates that the data from the external platform is more dispersed. The overall dispersion index D_disperse is obtained by taking the weighted average of the variances of all assessment sources. For example, when D_disperse takes the value in the range of [0,10], the larger the value, the more significant the ranking difference, thus forming the assessment ranking dispersion dataset.

[0044] By performing correlation analysis on the dispersion and initial credibility of the evaluation ranking, dispersion-credibility correlation data is obtained. The dispersion value of each evaluation object is mapped and matched with the initial credibility generated in step S3 to form binary data pairs. Each sample pair contains two parameters: dispersion D_disperse and credibility C_trust. The linear or non-linear correspondence between the two is identified by the association clustering algorithm. For example, in the sample set, when D_disperse is less than 2, C_trust is higher than 0.85, while when D_disperse exceeds 6, C_trust generally drops to below 0.6, indicating that there is a significant negative correlation between the two, thus generating the dispersion-credibility correlation dataset.

[0045] Gradient rate of change is calculated for dispersion-confidence correlation data. The decreasing gradient coefficient of confidence as dispersion changes is extracted. The correlation data is sorted in ascending order of dispersion value, and the rate of change of confidence for adjacent sample pairs is calculated. For example, when dispersion increases from 3.0 to 4.5, confidence decreases from 0.82 to 0.74, and the gradient rate of change is -0.08. The decreasing trend is calculated for all data using a sliding window statistical method, and the average gradient rate of change is extracted as the decreasing gradient coefficient. Used to characterize the rate of decline in credibility, for example A value of -0.06 indicates that for every unit increase in dispersion, the confidence level decreases by approximately 6%.

[0046] Construct a confidence-decreasing relationship based on the decreasing gradient coefficients, and Corresponding to the dispersion threshold range, a segmented decreasing relationship table is established. When the dispersion D_disperse is in [0,2], the confidence remains unchanged from the initial value. When D_disperse is in [2,5], the confidence is adjusted according to... The confidence level decreases linearly. When D_disperse exceeds 5, a saturated decreasing model is used to maintain the lower limit of confidence at 0.5. For example, when the dispersion of an object is 4.2, the confidence level is calculated to be 0.77, while when the dispersion is 6.8, the confidence level drops to 0.58, thus forming a complete confidence decreasing relationship model.

[0047] In this embodiment, the specific steps for constructing a decreasing confidence relationship based on the decreasing gradient coefficients are as follows: Identify the gradient change magnitude in the decreasing gradient coefficients, and divide the decreasing gradient coefficients into segments based on the gradient change magnitude to generate a set of distinguishable gradient intervals. By comparing the rising and falling directions of adjacent intervals in the gradient interval set, the overall gradient decreasing direction and local reversal points are extracted to identify the gradient decreasing trend. Based on the gradient decreasing trend, a smoothing correction mechanism is introduced at the local reversal point to obtain a stable decreasing trend line. By accumulating gradient changes in layers, the decay rate and inflection points of the stable decreasing trend line are extracted to generate a trajectory with decreasing credibility. The confidence decreases by dividing the decreasing boundary and steady-state interval based on the confidence decrease trajectory, and then generating the confidence decrease relationship.

[0048] In this embodiment, the gradient change magnitude in the decreasing gradient coefficients is identified, and the decreasing gradient coefficients are segmented based on the gradient change magnitude to generate a set of distinguishable gradient intervals. The decreasing gradient coefficient sequence is then... Arrange the gradients in chronological order over time and calculate the absolute difference between adjacent gradient coefficients. and with The range of change is used as the basis for segment division. A value less than 0.03 is defined as a stable region. The range between 0.03 and 0.07 is defined as the transition zone. A value greater than 0.07 is defined as a fluctuation range, for example, when continuously observed... When the values ​​are -0.05, -0.06, -0.08, -0.15, and -0.09, a stable segment A is formed between -0.05 and -0.08, a fluctuating segment B is formed between -0.08 and -0.15, and a transition segment C is formed between -0.15 and -0.09. After segmentation, a gradient interval set containing multiple segment identifiers and corresponding coefficient ranges is generated.

[0049] By comparing the rising and falling directions of adjacent intervals in the gradient interval set, the overall gradient decreasing direction and local reversal points are extracted to identify the gradient decreasing trend. The interval set is compared in ascending and descending order according to the segment number. When the central gradient value of the later segment is less than that of the previous segment, it is determined to be a decreasing direction. When the central gradient value of the adjacent segment increases in the opposite direction, it is marked as a local reversal point. For example, if the central gradient of segment A is -0.06, segment B is -0.12, and segment C is -0.10, then the overall trend is decreasing, but there is a reversal point between B and C. By statistically analyzing the positions of all reversal points, the overall direction of gradient change can be identified, thereby determining the initial shape of the decreasing trend curve and obtaining the gradient decreasing trend dataset.

[0050] Based on the gradient decreasing trend, a smoothing correction mechanism is introduced at local reversal points to obtain a stable decreasing trend line. Five gradient samples are extracted from the segments before and after each reversal point and smoothed by moving average to compress excessively drastic fluctuations. For example, at the reversal point of segment B, the original gradient change amplitude is 0.09, and after smoothing, the change amplitude is reduced to 0.04, making the trend line smoother and avoiding misjudgment of credibility due to single abnormal fluctuations. After processing all local reversal points through this smoothing correction mechanism, a continuous, monotonous, and abrupt stable decreasing trend line is formed.

[0051] By accumulating gradient changes in layers, the decay rate and inflection points of a stable decreasing trend line are extracted to generate a confidence-decreasing trajectory. The entire trend line is divided into several stages, each containing 10 to 20 sample points. The average gradient change within each stage is accumulated to reflect the decay intensity. For example, if the average gradient is -0.04 in the early stage, -0.07 in the middle stage, and -0.02 in the later stage, it indicates that the decay rate first increases and then decreases, suggesting that the confidence decrease is most significant in the middle stage. By analyzing the decay rate curve, inflection points can be identified. For example, a gradient convergence point appears when the cumulative change reaches 70%, which is marked as the main decay inflection point, thus generating a continuous confidence-decreasing trajectory dataset.

[0052] Based on the decreasing confidence trajectory, decreasing boundaries and steady-state intervals are defined to generate a decreasing confidence relationship. The segment of the decreasing trajectory with a decay rate greater than 0.06 is defined as the rapid decreasing region, the segment with a rate between 0.03 and 0.06 is defined as the slow decreasing region, and the segment with a rate less than 0.03 is defined as the steady-state interval. For example, in the sample data, the first segment of the decreasing trajectory lasts for 15 minutes, corresponding to a rate of 0.05, which belongs to the slow decreasing region; the middle segment lasts for 10 minutes, corresponding to a rate of 0.08, which belongs to the rapid decreasing region; and the last segment lasts for 20 minutes, corresponding to a rate of 0.02, which belongs to the steady-state interval. Through these boundary divisions, a complete decreasing confidence relationship model is formed.

[0053] In this embodiment, the specific steps of step S5 are as follows: A credibility comparison structure is established by pairing credibility levels with initial credibility data based on the decreasing credibility relationship; Based on the credibility comparison structure, the initial credibility data is segmented and sorted by time and group dimensions to obtain the credibility change path sequence. Abrupt change points are identified in the confidence change path sequence, and stabilization weights and balance weights are applied to the sudden drop segment and the sudden rise segment, respectively, to generate a balance adjustment sequence; The credibility index is generated by recalculating the credibility based on the balanced adjustment sequence.

[0054] In this embodiment, a credibility correspondence pairing is performed based on the credibility decrease relationship and the initial credibility data to establish a credibility comparison structure. The credibility decrease relationship model is mapped one-to-one with the initial credibility data. Each evaluation object corresponds to a decrease relationship curve and an initial credibility value. During the pairing process, the object identifier ID is used as the index key, and joint matching is performed through timestamp and data source type to ensure the temporal consistency and source uniqueness of the data. For example, when the initial credibility of object A is 0.82 and the corresponding decrease interval is the slow decrease zone, the pairing relationship is {ID:A, C_init:0.82, Decline_zone:slow}. When the initial credibility of object B is 0.71 and is in the fast decrease zone, the corresponding relationship is {ID:B, C_init:0.71, Decline_zone:fast}. The credibility comparison structure table is formed through the pairing operation. This structure table contains object identifier, initial credibility, decrease interval type, and timestamp fields.

[0055] Based on the credibility comparison structure, the initial credibility data is segmented and sorted by time and group dimensions to obtain a credibility change path sequence. All evaluation objects are divided according to the evaluation time order and their respective groups (such as expert group, external institution group, system evaluation group). The data within each group is arranged in ascending order of time. The credibility difference ΔC between adjacent time periods represents the direction of change. When ΔC is positive, it indicates that the credibility is increasing, and when ΔC is negative, it indicates that the credibility is decreasing. For example, in the expert group, the credibility of object A decreases from 0.83 to 0.78 from T1 to T2, so ΔC is -0.05. The credibility of object B increases from 0.75 to 0.80 from T1 to T2, so ΔC is +0.05. The credibility change path sequence is generated by sorting the difference and direction. The path sequence records the dynamic trajectory of the credibility of each object in the time dimension, forming a structured data table.

[0056] Abrupt change points are identified in the confidence change path sequence. Stabilization weights and balance weights are applied to the sudden drop and rise segments, respectively, to generate a balance adjustment sequence. Abrupt change points in the path sequence where the confidence change rate exceeds the threshold of 0.07 are identified and marked as abnormal change segments. A negative rate is defined as a sudden drop segment, and a positive rate is defined as a sudden rise segment. A stabilization weight coefficient W_suppress is applied to the sudden drop segment, with a value range of [0.3, 0.6], to suppress the effect of excessive decline. For example, if object A suddenly drops by 0.12 at time T3, the weighted adjustment is 0.12 × 0.5 = 0.06. A balance weight coefficient W_balance is applied to the sudden rise segment, with a value range of [0.4, 0.7], to weaken the fluctuations caused by excessively rapid rise. For example, if object B suddenly rises by 0.10 at time T3, the weighted adjustment is 0.10 × 0.6 = 0.06. The sequence generated after weighted correction is called the balance adjustment sequence.

[0057] The credibility reset calculation is performed based on the balancing adjustment sequence to generate the evaluation credibility index. The last period of the credibility time series of each object in the balancing adjustment sequence is taken as the stable credibility value. At the same time, the credibility is normalized by combining the decreasing interval and the weighted correction parameter to keep the final credibility distribution range between [0,1]. For example, after multiple rounds of adjustment, the final credibility of object A is 0.79, object B is 0.74, and object C is 0.88. The final results of all objects are weighted and averaged to obtain the overall evaluation credibility index C_index. In a sample of 50 evaluation objects, the C_index is calculated to be 0.81. This index is used to reflect the credibility level of the overall evaluation results of the system.

[0058] In this embodiment, the specific steps for identifying abrupt change points in the confidence change path sequence and applying stabilization weights and balancing weights to the sudden drop and rise segments respectively are as follows: The difference between adjacent nodes in the confidence change path sequence is calculated, the magnitude and direction information of the confidence change are extracted, and the path fluctuation difference is generated. Based on the path fluctuation difference, abrupt change points are identified in the path sequence of credibility changes to obtain the location indexes of sudden drop points and sudden rise points. Based on the location index of the sudden drop point, the sudden drop point interval of the confidence change path sequence is located, and the sudden drop point interval is stabilized and balanced to generate a confidence recovery sequence. Based on the location index of the rise point, the rise point interval of the credibility change path sequence is located, and the rise point interval is converged and adjusted to generate a credibility balance sequence. The confidence stabilization sequence and the confidence equilibrium sequence are merged and progressively corrected to generate the equilibrium adjustment sequence.

[0059] In this embodiment, the difference between adjacent nodes in the credibility change path sequence is calculated, the magnitude and direction information of credibility change are extracted, and the path fluctuation difference is generated. The credibility change path sequence C_path generated in step S5 is arranged in chronological order, and each node represents the credibility value of a certain evaluation object in a specific time period. The difference ΔC between adjacent nodes is calculated one by one to reflect the magnitude and direction of credibility change. For example, if the credibility of object A decreases from 0.82 to 0.76 from T1 to T2, then ΔC is -0.06, indicating a downward trend. If the credibility of object B increases from 0.75 to 0.81 from T1 to T2, then ΔC is +0.06, indicating an upward trend.

[0060] Based on the path fluctuation difference, abrupt change points are identified in the path sequence of credibility change. The position indices of abrupt decrease points and abrupt increase points are obtained. A change threshold is set. For example, when the absolute value of ΔC is greater than 0.07, it is marked as a change point. Each ΔC value in the path sequence is compared. If ΔC ≤ -0.07, it is marked as a decrease point. If ΔC ≥ 0.07, it is marked as an increase point. For example, if object A has ΔC = -0.12 at time T3, it is identified as a decrease point with a position index of 3. If object B has ΔC = 0.10 at time T4, it is identified as an increase point with a position index of 4. Through this operation, the index set of all decrease points and increase points is obtained.

[0061] Based on the location index of the drop point, the drop point interval of the confidence change path sequence is located, and the drop point interval is stabilized and balanced to generate a confidence recovery sequence. The interval formed by the drop point and its adjacent nodes before and after it is taken as the drop interval. A recovery weight W_suppress is applied to the confidence value in the interval, for example, with a value of 0.4 to 0.6, to reduce the drop amplitude proportionally and make the downward trend more gradual. For example, if object A drops by 0.12 in the interval from T3 to T4, after adjusting the weight by 0.5, the actual drop amplitude becomes 0.06. The adjusted confidence value is recorded to generate the recovery sequence.

[0062] Based on the location index of the rise point, the rise point interval of the credibility change path sequence is located, and the rise point interval is converged and adjusted to generate a credibility balance sequence. The interval formed by the rise point and its adjacent nodes before and after it is taken as the rise interval. A balance weight W_balance is applied to the credibility value in the interval, for example, with a value of 0.5 to 0.7, to slow down the trend of rising too fast. For example, if object B rises by 0.10 in the interval from T4 to T5, after the weight adjustment of 0.6, the actual rise is 0.06, thus generating a balance sequence.

[0063] The confidence stabilization sequence and the confidence equilibrium sequence are merged and progressively corrected to generate a balance adjustment sequence C_adjust. The stabilization sequence and the equilibrium sequence are merged in chronological order. If necessary, adjacent nodes are linearly smoothed, with the adjustment magnitude not exceeding 0.02 to ensure that the sequence is smooth and continuous. For example, the adjusted confidence sequences of objects A and B are [0.82,0.79,0.73,0.71] and [0.75,0.78,0.84,0.85], respectively. After smoothing and merging, the final balance adjustment sequence [0.82,0.79,0.73,0.78,0.84,0.85] is obtained. This sequence retains the original trend of change and eliminates abrupt changes.

[0064] In this embodiment, the specific steps of step S6 are as follows: The credibility index data of the assessment is divided into stages to obtain a staged credibility sequence. Deviation comparison is performed on the phased credibility sequence, and stable and abnormal regions are identified by detecting differences between the preceding and following phases. A recursive comparison relationship is established between the stable region and the abnormal region, and a closed-loop mechanism for credibility verification is constructed through the recursive comparison relationship. A closed-loop credibility verification mechanism is used to identify records that deviate from credibility. Extract the stability, fluctuation range and recovery trend of each stage in the credibility deviation record, and aggregate them into a credibility assessment report.

[0065] In this embodiment, the evaluation credibility index data is divided into stages to obtain a staged credibility sequence. The balance adjustment sequence generated in step S5 is divided into several stages according to the time or group dimension. Each stage contains 3 to 5 consecutive nodes. Assuming the credibility sequence is [0.82,0.79,0.73,0.78,0.84,0.85,0.80,0.77], it can be divided into stage 1: [0.82,0.79,0.73], stage 2: [0.78,0.84,0.85], and stage 3: [0.80,0.77].

[0066] The confidence sequence is compared for deviations. Stable and abnormal regions are identified by detecting differences between the preceding and following stages. The average value and the maximum and minimum deviations of the confidence values ​​within each stage are calculated. For example, the average value of stage 1 is 0.78, the maximum value is 0.82, the minimum value is 0.73, and the deviation is 0.09. The average value of stage 2 is 0.823 and the deviation is 0.07. The average value of stage 3 is 0.785 and the deviation is 0.03. Stages with smaller deviations are marked as stable regions, such as stage 2 and stage 3, while stages with larger deviations are marked as abnormal regions, such as stage 1, thus forming distinguishing labels between stable and abnormal regions.

[0067] Based on the establishment of a recursive comparison relationship between the stable region and the abnormal region, a closed-loop mechanism for credibility verification is constructed through the recursive comparison relationship. The credibility value of each stage is compared with the average value of the previous stage to identify the deviation magnitude and record the abnormal change trend. For example, the average value of stage 2 (0.823) increases by 0.043 compared with the average value of stage 1 (0.78). The deviation is within the acceptable range, so the recursive verification is normal. If the deviation exceeds 0.05, the closed-loop correction is triggered. Through this mechanism, the stability and abnormal trends of each stage are mapped to each other to form a closed-loop verification chain.

[0068] The credibility deviation record is determined by using a credibility verification closed-loop mechanism. For each stage, the deviation magnitude, direction and impact range are recorded. For example, in stage 1, the deviation magnitude is 0.09 and the direction is decreasing; in stage 2, the deviation magnitude is 0.043 and the direction is increasing; and in stage 3, the deviation magnitude is 0.038 and the direction is decreasing. This information is used to generate a deviation record table.

[0069] The stability, fluctuation range, and recovery trend of each stage in the credibility deviation record are collected and aggregated into a credibility assessment report. The average value, maximum and minimum deviation, deviation range, and trend information of each stage are integrated. For example, stage 1 has moderate stability, fluctuation range of 0.09, and upward recovery trend; stage 2 has high stability, fluctuation range of 0.07, and stable recovery trend; stage 3 has high stability, fluctuation range of 0.03, and slightly downward recovery trend. All stages are summarized to form a complete credibility assessment report, generating a detailed report that includes stage division, stability analysis, anomaly area marking, and recovery trend.

[0070] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0071] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement it. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for calculating the reliability of multi-dimensional assessments by integrating Kendall's coefficient of harmony, characterized in that, Includes the following steps: Step S1: Collect multi-source assessment information, arrange it in order to form an assessment sequence, and identify the assessment behavior characteristics of the assessment sequence. Step S2: Construct an independent time window based on the total number of evaluation objects from multiple sources of evaluation information, and calculate the Kendall's coefficient of harmony for the evaluation objects within the independent time window; Step S3: Construct a three-dimensional interaction tensor based on the assessment behavior characteristics and Kendall's harmony coefficient, and identify the initial credibility of the assessment information through three-dimensional interaction tensor decomposition; Step S4: Calculate the evaluation ranking dispersion of multi-source evaluation information, and construct a decreasing credibility relationship through the evaluation ranking dispersion; Step S5: Based on the decreasing credibility relationship, perform impact correction on the initial credibility level and reset the credibility index to generate the evaluation credibility index; Step S6: Establish a credibility verification closed-loop mechanism based on the credibility index and generate a credibility evaluation report.

2. The multi-dimensional assessment reliability aggregation calculation method integrating Kendall's concordance coefficient as described in claim 1, characterized in that, The specific steps of step S1 are as follows: Collect multi-source evaluation information; Time-series synchronization processing is performed on multi-source evaluation information to obtain time-aligned evaluation information; The time-aligned assessment information is arranged in order to form an assessment sequence. The score difference is calculated for each adjacent sequence in the evaluation sequence, and the score difference is mapped to the local volatility of the sequence score. Extreme ranking frequencies in the evaluation sequence are eliminated by local volatility, and evaluation behavior characteristics in the evaluation sequence are identified.

3. The multi-dimensional assessment reliability aggregation calculation method integrating Kendall's concordance coefficient as described in claim 1, characterized in that, The specific steps of step S2 are as follows: Collect the identification of the evaluation object from multi-source evaluation information; count the total number of evaluation objects based on the evaluation object identification; The time window width is matched and calculated based on the total number of evaluation subjects to determine the time span threshold of the independent time window; Multiple independent time windows are constructed for multi-source evaluation information by using the time span threshold of independent time windows; For each independent time window, a pairwise sorting consistency comparison is performed on the evaluation order sequence, and the rank correlation coefficient between the sorted objects is calculated. Kendall's harmonic coefficient transformation is performed on the rank correlation coefficients between the sorted objects to obtain the Kendall's harmonic factor of the evaluated objects within an independent time window. Kendall's harmony factors for all evaluation objects within all independent time windows are arranged in a time series to generate Kendall's harmony coefficients.

4. The multi-dimensional assessment reliability aggregation calculation method integrating Kendall's concordance coefficient as described in claim 1, characterized in that, The specific steps of step S3 are as follows: Extract behavioral patterns and behavioral stability parameters from the behavioral characteristics assessed. Calculate the mean and variance of the ranking position of each reviewed object based on the review ranking sequence; A three-dimensional interaction tensor is generated by constructing tensors using behavioral stability parameters, Kendall's harmony coefficient, and ordination variance as three independent dimensions. The coupling degree of behavior-harmony-ordering is calculated based on the three-dimensional interaction tensor, and the coupling degree is mapped to the interaction influence coefficient. The credibility of multi-source evaluation information is mapped based on the interaction influence coefficient to obtain the initial credibility level.

5. The multi-dimensional assessment reliability aggregation calculation method integrating Kendall's concordance coefficient as described in claim 1, characterized in that, The specific steps of step S4 are as follows: Extract the ranking results of each evaluation source from the multi-source evaluation information; The ranking difference is calculated based on the evaluation ranking results to obtain the ranking deviation. Variance is calculated based on the ranking deviation, and the dispersion of the ranking of multi-source evaluation information is determined by the variance. By evaluating the dispersion of the ranking and the initial confidence level, a correlation analysis was conducted to obtain dispersion-confidence correlation data; Calculate the gradient rate of change for dispersion-confidence correlation data and extract the decreasing gradient coefficient of confidence as dispersion changes. A confidence-decreasing relationship is constructed based on the decreasing gradient coefficients.

6. The multi-dimensional assessment reliability aggregation calculation method integrating Kendall's concordance coefficient as described in claim 5, characterized in that, The specific steps for constructing a decreasing confidence relationship based on the decreasing gradient coefficients are as follows: Identify the gradient change magnitude in the decreasing gradient coefficients, and divide the decreasing gradient coefficients into segments based on the gradient change magnitude to generate a set of distinguishable gradient intervals. By comparing the rising and falling directions of adjacent intervals in the gradient interval set, the overall gradient decreasing direction and local reversal points are extracted to identify the gradient decreasing trend. Based on the gradient decreasing trend, a smoothing correction mechanism is introduced at the local reversal point to obtain a stable decreasing trend line. By accumulating gradient changes in layers, the decay rate and inflection points of the stable decreasing trend line are extracted to generate a trajectory with decreasing credibility. The confidence decreases by dividing the decreasing boundary and steady-state interval based on the confidence decrease trajectory, and then generating a confidence decrease relationship.

7. The multi-dimensional assessment reliability aggregation calculation method integrating Kendall's concordance coefficient as described in claim 1, characterized in that, The specific steps of step S5 are as follows: A credibility comparison structure is established by pairing credibility levels with initial credibility data based on the decreasing credibility relationship; Based on the credibility comparison structure, the initial credibility data is segmented and sorted by time and group dimensions to obtain the credibility change path sequence. Abrupt change points are identified in the confidence change path sequence, and stabilization weights and balance weights are applied to the sudden drop segment and the sudden rise segment, respectively, to generate a balance adjustment sequence; The credibility index is generated by recalculating the credibility based on the balanced adjustment sequence.

8. The multi-dimensional assessment reliability aggregation calculation method integrating Kendall's concordance coefficient as described in claim 1, characterized in that, The specific steps for identifying abrupt change points in the confidence change path sequence and applying stabilization weights and balancing weights to the sudden drop and rise segments, respectively, are as follows: The difference between adjacent nodes in the confidence change path sequence is calculated, the magnitude and direction information of the confidence change are extracted, and the path fluctuation difference is generated. Based on the path fluctuation difference, abrupt change points are identified in the path sequence of credibility changes to obtain the location indexes of sudden drop points and sudden rise points. Based on the location index of the sudden drop point, the sudden drop point interval of the confidence change path sequence is located, and the sudden drop point interval is stabilized and balanced to generate a confidence recovery sequence. Based on the location index of the rise point, the rise point interval of the credibility change path sequence is located, and the rise point interval is converged and adjusted to generate a credibility balance sequence. The confidence stabilization sequence and the confidence equilibrium sequence are merged and progressively corrected to generate the equilibrium adjustment sequence.

9. The multi-dimensional assessment reliability aggregation calculation method integrating Kendall's concordance coefficient as described in claim 1, characterized in that, The specific steps of step S6 are as follows: The credibility index data of the assessment is divided into stages to obtain a staged credibility sequence. Deviation comparison is performed on the phased credibility sequence, and stable and abnormal regions are identified by detecting differences between the preceding and following phases. A recursive comparison relationship is established between the stable region and the abnormal region, and a closed-loop mechanism for credibility verification is constructed through the recursive comparison relationship. A closed-loop credibility verification mechanism is used to identify records that deviate from credibility. Extract the stability, fluctuation range and recovery trend of each stage in the credibility deviation record, and aggregate them into a credibility assessment report.

Images