User behavior data mining method and system applied to digital enterprise management

Abstract

The application provides a user behavior data mining method and system applied to digital enterprise management, collects multi-dimensional behavior data of a target user on a business operation interface, performs multi-modal data analysis on the multi-dimensional behavior data, generates a behavior track feature set with time sequence correlation, trains an adaptive time sequence analysis model based on the behavior track feature set, the time sequence analysis model captures long and short term dependence in a user behavior mode through a dynamic window division strategy, and generates a potential loss risk prediction index, constructs an interactive process parameter matrix according to the potential loss risk prediction index, calls the optimized interactive process parameter matrix to drive the reconstruction of the business operation interface, generates an interactive interface adapted to the current user behavior mode, and iteratively updates the time sequence analysis model through an incremental feedback mechanism within a preset verification period. The application can improve the comprehensiveness and accuracy of user behavior mode mining.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of data processing, in particular to a user behavior data mining method and system applied to digital enterprise management. BACKGROUND

[0002] In the field of digital enterprise management, user behavior data analysis as the core means of optimizing business processes and improving user experience, mainly through collecting user operation data to build behavior models to support decision optimization. The existing technology usually builds a static analysis model based on user behavior characteristics (such as click frequency or stay time), predicts user behavior trends through preset rules or simple time sequence algorithms, and uses fixed parameters to drive interface optimization strategies. However, the existing method is difficult to capture the complex behavior patterns such as cross-interface spatial distribution, function module switching logic and business request correlation implied in user operation, resulting in significant deviation in potential loss risk prediction; at the same time, the static model cannot adapt to the dynamic evolution law of user behavior patterns, the interface optimization strategy is out of touch with real-time operation feedback, causing interface reconstruction hysteresis and recommended strategy rigidity, which seriously restricts the response efficiency and user experience improvement of enterprise digital management system. In addition, the isolated data collection, model training and interface optimization links in the traditional technical solution form an open-loop architecture, lacking an incremental learning mechanism based on user real-time behavior feedback, further exacerbating the technical defect of insufficient system adaptive ability. SUMMARY

[0003] The present application provides a user behavior data mining method and system applied to digital enterprise management.

[0004] In a first aspect, the present application embodiment provides a user behavior data mining method applied to digital enterprise management, the method comprising: collecting multi-dimensional behavior data of a target user on a business operation interface; performing multi-modal data analysis on the multi-dimensional behavior data to generate a behavior trajectory feature set with time sequence correlation, wherein different modal data are time-stamped and semantically associated through a cross-modal alignment strategy; training an adaptive time sequence analysis model based on the behavior trajectory feature set, the time sequence analysis model capturing long and short term dependencies in user behavior patterns through a dynamic window division strategy, and generating a potential loss risk prediction index; constructing an interactive process parameter matrix according to the potential loss risk prediction index; calling the optimized interactive process parameter matrix to drive the reconstruction of the business operation interface, generating an interactive interface adapted to the current user behavior pattern, and iteratively updating the time sequence analysis model through an incremental feedback mechanism within a preset verification period.

[0005] In a second aspect, the present application embodiment provides a computer system, comprising: a memory, the memory storing a computer program; a processor for loading the computer program to realize the above-mentioned user behavior data mining method applied to digital enterprise management.

[0006] This invention provides a user behavior data mining method for digital enterprise management. Through multi-dimensional behavioral data and a cross-modal alignment strategy, it achieves deep fusion of spatiotemporal features and semantic associations, breaking through the limitations of traditional data analysis and significantly improving the comprehensiveness and accuracy of user behavior pattern mining. An adaptive temporal analysis model built based on a dynamic window partitioning strategy can intelligently capture the dynamic evolution of long-term and short-term dependencies in user operations, effectively enhancing the reliability of potential churn risk prediction indicators. By constructing an interaction process parameter matrix including interface element weight distribution, functional module priority sequences, and recommendation strategy adjustment coefficients, it achieves deep coupling between user behavior characteristics and business scenarios, driving real-time dynamic optimization of business operation interface reconstruction strategies. Combined with a closed-loop learning link formed by an incremental feedback mechanism, it continuously iterates and updates the temporal analysis model and interface interaction parameters, ensuring synchronous adaptation between user behavior pattern evolution and interface optimization processes. Ultimately, while reducing user churn risk, it comprehensively improves the response efficiency and user experience of the digital management system, providing reliable support for accurate decision-making in complex business scenarios. Attached Figure Description

[0007] Figure 1 This is a flowchart of a user behavior data mining method for digital enterprise management provided by an embodiment of the present invention.

[0008] Figure 2 This is a schematic diagram of the composition of a computer system provided in an embodiment of the present invention. Detailed Implementation

[0009] Please see Figure 1 , Figure 1 A flowchart illustrating a user behavior data mining method for digital enterprise management, provided as an embodiment of the present invention, is shown. This user behavior data mining method for digital enterprise management can be executed by a computer system and includes the following steps:

[0010] Step S100: Collect multi-dimensional behavioral data of the target user on the business operation interface. The multi-dimensional behavioral data includes interface click trajectory, operation duration distribution, functional module switching path and related business data request records.

[0011] The interface click track refers to the track formed by the user's click operation on the business operation interface, which can reflect the user's operation position and sequence on the interface. For example, in a business operation interface of an e-commerce platform, the user clicks the product detail page, the add-to-cart button, the settlement button, etc. in turn, and these click operations constitute the interface click track. The operation duration distribution refers to the time distribution spent by the user in different operation links, which can reflect the user's concentration and input on different operations. For example, the user stays on the product detail page for 5 minutes and stays on the shipping address page for 2 minutes, which is the embodiment of the operation duration distribution. The function module switching path refers to the path of switching between different function modules of the business operation interface, which reflects the user's operation process and business demand. The associated business data request record refers to the record of the user's request for associated business data in the operation process, which contains the user's demand information for business data. For example, in a financial system, the user requests to query the financial report of a certain time period, which is an associated business data request record.

[0012] Step S200: multi-modal data analysis is performed on the multi-dimensional behavior data to generate a behavior track feature set with time sequence correlation, wherein different modal data are time-stamped and semantically associated through a cross-modal alignment strategy.

[0013] Multi-modal data analysis refers to the analysis and processing of different types of multi-dimensional behavior data to extract useful information. In this step, the interface click track, operation duration distribution, function module switching path and associated business data request record are analyzed. The behavior track feature set with time sequence correlation refers to the conversion of multi-dimensional behavior data into a feature set with time sequence and correlation, which can more comprehensively reflect the user's behavior pattern. The cross-modal alignment strategy synchronizes the time stamps of different modal data and semantically aligns them in time and semantics to facilitate better analysis and processing. Time stamp synchronization refers to the unification of time stamps of different modal data, so that they are comparable in time.

[0014] As an embodiment, the above-mentioned multi-dimensional behavior data includes interface click track, operation duration distribution, function module switching path and associated business data request record. Based on this, step S200, multi-modal data analysis is performed on the multi-dimensional behavior data to generate a behavior track feature set with time sequence correlation, which can specifically include the following steps S210-S250:

[0015] Step S210: extract the spatial coordinate sequence in the interface click track, and calculate the interface area attention heat map based on the coordinate distribution density.

[0016] The sequence of spatial coordinates in the interface click track refers to the sequence composed of the position coordinates of the user's click operations on the business operation interface. It can accurately reflect the user's click position on the interface. The interface area attention heat map is a chart used to visualize the user's attention to different areas of the interface, which represents the user's attention to different areas by the depth of color. The deeper the color, the higher the user's attention.

[0017] Extracting the sequence of spatial coordinates can be achieved by processing the interface click track data. Specifically, the position coordinates of each click operation can be extracted from the click track data and arranged in the order of the clicks to obtain the sequence of spatial coordinates. In one example, the business operation interface is divided into several small areas, for example, the interface is divided into a 10x10 grid. Then, the number of clicks in each area is counted, and the more clicks indicate that the coordinate distribution density of that area is greater. Finally, a color value is assigned to each area according to the coordinate distribution density, and the depth of the color value is proportional to the coordinate distribution density, thereby generating the interface area attention heat map.

[0018] Step S220: Analyzing the dwell interval feature in the operation duration distribution, identifying operation clusters composed of continuous operation events exceeding the frequency threshold and abnormal stationary nodes, and high-frequency operation clusters being composed of continuous events with operation frequency exceeding the preset threshold in the sliding window.

[0019] The dwell interval feature in the operation duration distribution refers to the feature presented by the dwell time interval between different operation links. It can reflect the user's operation rhythm and attention concentration. For example, the user stays for a long time between two adjacent operation links, indicating that the user may be thinking or encountering a problem. The operation cluster refers to a set composed of continuous operation events exceeding the frequency threshold, which represents a series of operations concentrated by the user within a period of time. The abnormal stationary node refers to the position where the user stays for an abnormally long time during the operation process, which may indicate that the user encounters difficulties or the operation process has problems. The high-frequency operation cluster refers to a cluster composed of continuous operation events with operation frequency exceeding the preset threshold in the sliding window.

[0020] Analyzing the dwell interval feature can be achieved by analyzing the operation duration distribution data. For example, the dwell time interval between adjacent operation links is calculated, and statistical analysis is performed on these intervals, such as calculating the mean, standard deviation, etc., to extract the dwell interval feature.

[0021] Step S230: Constructing the transition probability matrix of the function module switching path, calculating the state transition weights between the function modules, and generating the path prediction sequence based on the hidden Markov model.

[0022] The transition probability matrix of the function module switching path is a matrix representing the probability of a user switching from one function module to another function module. The rows of the matrix represent the starting function module, the columns represent the target function module, and the elements in the matrix represent the transition probability from the starting function module to the target function module. The state transition weight refers to the importance of the transition from one function module to another function module during the function module switching process. It can be calculated according to factors such as transition probability and operation frequency. For example, the number of times a user switches between different function modules is counted to obtain a switching frequency matrix. Then, each row element of the switching frequency matrix is divided by the sum of the elements in that row to obtain the transition probability matrix. The state transition weight can be calculated by weighting factors such as transition probability and operation frequency. For example, the state transition weight can be obtained by multiplying the transition probability by the operation frequency.

[0023] Step S240: Map the associated business data request record into a request type code vector, and construct a request load evaluation index through the request response time and the result data volume.

[0024] Mapping the associated business data request record into a request type code vector means representing different types of associated business data request records in the form of a vector. Request response time refers to the time taken from the user issuing a data request to the system returning a response result. Result data volume refers to the data size of the response result returned by the system. The request load evaluation index is an index for evaluating the system's processing request load, which can reflect the system's pressure and performance when processing requests.

[0025] Step S250: Perform multi-modal feature fusion on the attention heat map, dwell interval feature, transition probability matrix, and request load evaluation index, and generate a time series embedding vector containing global context information as the behavior trajectory feature set using a time axis alignment strategy.

[0026] Multi-modal feature fusion refers to the fusion of different types of features to obtain more comprehensive and valuable feature representations. In this step, the attention heat map, dwell interval feature, transition probability matrix, and request load evaluation index are fused. The time axis alignment strategy is a method for processing multi-modal data, which aligns different modal data on the time axis to make them consistent in time. The time series embedding vector is a vector that embeds time series data, which can convert information with time order into a low-dimensional vector space. Global context information refers to the overall information of multi-dimensional behavior data, which can more comprehensively reflect the user's behavior patterns.

[0027] The multi-modal features can be fused in various ways, such as feature splicing, weighted summation, etc. In this step, the attention heat map, dwell interval feature, transition probability matrix and request load evaluation index can be spliced to obtain a fused feature vector. The process of generating a time series embedding vector using a time axis alignment strategy is as follows: first, arrange the fused feature vector in chronological order to form a time series feature sequence. Then, use embedding algorithms (such as Word2Vec, GloVe, etc.) to convert the time series feature sequence into a time series embedding vector. This time series embedding vector contains global context information and can more comprehensively reflect the user's behavior patterns, so it can be used as a behavior trajectory feature set. When performing multi-modal feature fusion, different weights can be assigned to each feature according to its importance. For features of different dimensions, the assignment of weights can balance their influence in the fusion. For example, the weight of each feature is determined by domain knowledge, experimental verification or machine learning algorithms (such as feature selection algorithms). Multiply each feature by its corresponding weight and then fuse them, so that features of different dimensions can reasonably participate in the fusion process. Alternatively, use min-max normalization to scale the value range of each feature to the [0, 1] interval, and then fuse them. In other aspects of the present embodiment, if similar technical content of different dimensional feature fusion is involved, it should be understood that those skilled in the art can normalize the dimensions based on their own common technical knowledge for further feature operations.

[0028] Step S300: training an adaptive time series analysis model based on the behavior trajectory feature set, the time series analysis model capturing long and short term dependencies in user behavior patterns through a dynamic window division strategy and generating a potential churn risk prediction indicator.

[0029] The adaptive time series analysis model is a time series analysis model that can automatically adjust its parameters and structure based on input data. It can better adapt to different user behavior patterns and data characteristics. The dynamic window division strategy is a method for processing time series data, which can dynamically adjust the size and position of the window according to the characteristics of the data and the analysis requirements, to capture long and short term dependencies in user behavior patterns. Long and short term dependencies refer to the long and short term relationships between data at different time points in time series data. The potential churn risk prediction indicator is an indicator for predicting whether a user has a churn risk, which can help enterprises take measures in advance to prevent user churn.

[0030] As an implementation, step S300, training an adaptive time series analysis model based on the behavior trajectory feature set, can specifically include steps S310-S350:

[0031] Step S310: input the time sequence embedding vector into the bidirectional long short-term memory network, capture the forward and reverse time sequence dependency relationship through the gating mechanism, and generate an initial hidden state sequence.

[0032] Specifically, the time sequence embedding vector is input into the BiLSTM network in chronological order. At each time step, the gating mechanism calculates the states of the input gate, the forget gate, and the output gate according to the current input and the hidden state of the last time step, thereby controlling the inflow, retention, and output of information. Through the gating mechanism, the BiLSTM network can effectively capture the forward and reverse time sequence dependency relationship. Finally, the outputs of the forward LSTM layer and the reverse LSTM layer are spliced to obtain the initial hidden state sequence.

[0033] Step S320: connect a multi-head self-attention mechanism layer at the back end of the bidirectional long short-term memory network, calculate the correlation weights between different time step features, and perform attention weighted aggregation on the initial hidden state sequence to obtain an aggregated feature sequence.

[0034] The correlation weight refers to the importance of the correlation between different time step features in the multi-head self-attention mechanism. The process of connecting a multi-head self-attention mechanism layer at the back end of the bidirectional long short-term memory network is as follows: first, input the initial hidden state sequence into the multi-head self-attention mechanism layer. In each head, the correlation weights between different time step features in the input sequence are calculated. Specifically, the correlation weights are obtained by calculating the similarity between the query vector, the key vector, and the value vector. Then, the input sequence is weighted and summed according to the correlation weights to obtain the output of each head. Finally, the outputs of multiple heads are spliced to obtain the aggregated feature sequence.

[0035] Step S330: use a dynamic convolution kernel group to perform multi-scale feature extraction on the aggregated feature sequence to obtain multi-scale features, and the dynamic convolution kernel group adaptively adjusts the convolution kernel size and dilation coefficient according to the current input feature dimension.

[0036] The dynamic convolution kernel group is a group of convolution kernels that can adaptively adjust the size and dilation coefficient of the convolution kernel according to the dimension of the current input feature. Multi-scale feature extraction refers to extracting feature information of different scales from the input feature sequence to more comprehensively describe the features of the data.

[0037] As an implementation, step S330, using a dynamic convolution kernel group to perform multi-scale feature extraction on the aggregated feature sequence to obtain multi-scale features, can specifically include the following steps S331-S335:

[0038] Step S331: calculate a convolution kernel size adjustment factor according to the time step and channel dimension of the input feature sequence, and the adjustment factor is negatively correlated with the local variance of the feature sequence.

[0039] The time step of the input feature sequence refers to the length of the feature sequence in the time dimension, and the channel dimension refers to the number of feature sequences in the feature dimension. The convolution kernel size adjustment factor is a factor for adjusting the size of the convolution kernel, which is negatively correlated with the local variance of the feature sequence. The local variance refers to the variance of the feature sequence in the local region, which can reflect the fluctuation degree of the feature sequence.

[0040] The process of calculating the convolution kernel size adjustment factor according to the time step and channel dimension of the input feature sequence is as follows: first, calculate the variance of the input feature sequence in different local regions. The feature sequence can be divided into several local regions, each region containing several time steps and channels. Then, statistical analysis is performed on the variance of each local region to obtain the average value of the local variance. Finally, the convolution kernel size adjustment factor is calculated according to the average value of the local variance, and the adjustment factor is negatively correlated with the local variance. For example, an inverse proportional function can be used to calculate the adjustment factor, so that the larger the local variance, the smaller the adjustment factor, thereby reducing the size of the convolution kernel.

[0041] Step S332: generating a set of basic convolution kernel sizes based on the adjustment factor, and assigning a learnable dilation coefficient matrix in each convolution kernel, which dynamically adjusts the receptive field range according to the feature difference between adjacent time steps.

[0042] The set of basic convolution kernel sizes refers to a group of convolution kernel sizes generated according to the convolution kernel size adjustment factor. The learnable dilation coefficient matrix is a matrix whose elements can be learned and adjusted during training. The dilation coefficient matrix dynamically adjusts the receptive field range according to the feature difference between adjacent time steps, so that it can better capture feature information of different scales.

[0043] The process of generating a set of basic convolution kernel sizes based on the adjustment factor is as follows: first, determine a basic convolution kernel size range according to the convolution kernel size adjustment factor. Then, generate a group of convolution kernels with different sizes in this range to form a set of basic convolution kernel sizes. The process of assigning a learnable dilation coefficient matrix in each convolution kernel is as follows: assign a learnable dilation coefficient matrix to each convolution kernel, and initialize the elements of the matrix to a suitable value. During training, the elements of the dilation coefficient matrix are dynamically adjusted according to the difference between adjacent time steps. Specifically, a neural network can be used to learn the dilation coefficient matrix, with the feature difference between adjacent time steps as input and the adjusted dilation coefficient matrix as output. By adjusting the dilation coefficient matrix, the receptive field range of the convolution kernel can be changed, thereby extracting feature information of different scales.

[0044] Step S333: applying multiple dynamic convolution kernels to the same input feature sequence in parallel to generate feature maps with different scale characteristics.

[0045] A dynamic convolution kernel refers to a convolution kernel that adjusts its size and dilation factor adaptively according to the input feature dimension. A feature map refers to an output feature map obtained after a convolution operation, which reflects the feature information of the input feature sequence at different scales.

[0046] The process of applying multiple dynamic convolution kernels to the same input feature sequence is as follows: the input feature sequence is simultaneously input into multiple dynamic convolution kernels, each with different sizes and dilation factors. Each convolution kernel will perform a convolution operation on the input feature sequence to generate a feature map. Due to the different sizes and dilation factors of different convolution kernels, the generated feature maps have different scale characteristics. By applying multiple dynamic convolution kernels in parallel, different scale feature information can be extracted simultaneously, and feature maps with different scale characteristics can be obtained.

[0047] Step S334: Channel attention weighting is performed on the feature map, and the importance scores of each channel feature are calculated and the channel weights are redistributed.

[0048] The importance score of each channel feature refers to the score obtained by evaluating the importance of each channel feature in the feature map. Redistributing the channel weight refers to adjusting the weight of the channel according to the importance score of each channel feature, so that important channels have greater weights. First, a global average pooling operation is performed on each channel of the feature map, compressing the feature map of each channel into a scalar value. Then, these scalar values are input into a fully connected neural network, which will output the importance score of each channel. According to the importance score, the weight of the channel is redistributed. For example, a Sigmoid function can be used to map the importance score to the [0, 1] interval, and then the value is used as the weight of the channel. Finally, each channel of the feature map is multiplied by the corresponding weight to obtain the channel attention weighted feature map.

[0049] Step S335: The weighted feature map is concatenated along the channel dimension, and dimension reduction processing is performed through a separable convolution layer to generate multi-scale features.

[0050] Specifically, the multiple channel attention weighted feature maps are concatenated in the channel dimension, i.e., their channels are arranged in sequence to form a new feature map. A depth separable convolution is used to perform convolution operation on the concatenated feature map, and the depth separable convolution will perform convolution on each channel separately, and then concatenate the results. Then, a pointwise convolution is used to perform convolution operation on the output of the depth separable convolution, and the pointwise convolution will perform linear combination on each channel to achieve dimension reduction. Finally, the dimension-reduced feature map, i.e., the multi-scale feature, is obtained.

[0051] Step S340: Residual connection is performed between the multi-scale features and the original time-series embedding vectors, and a fully connected layer is used to map them into a latent vector space containing abstract representations of user behavior patterns.

[0052] Specifically, the multi-scale features and the original time-series embedding vectors are element-wise added to obtain a residual connected feature vector. The residual connected feature vector is input into a fully connected layer, which performs linear transformation and non-linear activation on the input vector to map it into a low-dimensional vector space. This low-dimensional vector space is the latent vector space, which contains abstract representations of user behavior patterns and can be used for subsequent analysis and prediction.

[0053] Step S350: A joint training framework of contrastive learning task and prediction task is constructed in the latent vector space. The contrastive learning task enhances the model's ability to recognize behavior pattern differences through positive and negative sample pairs, and the prediction task optimizes the accuracy of the latent attrition risk prediction indicator through the cross-entropy loss function.

[0054] For example, a set of feature vectors of user behavior patterns is first sampled from the latent vector space, and positive sample pairs are generated based on the time-series continuity of the feature vectors. Negative sample pairs are generated by randomly perturbing or pairing non-adjacent time period feature vectors to form a training data set for the contrastive learning task.

[0055] The latent vector space is a low-dimensional vector space that contains abstract representations of user behavior patterns. The set of feature vectors refers to a set of feature vectors sampled from the latent vector space, each representing a user's behavior pattern. The positive sample pair refers to a pair of feature vectors with similar behavior patterns, generated based on the time-series continuity of the feature vectors. The negative sample pair refers to a pair of feature vectors with different behavior patterns, generated by randomly perturbing or pairing non-adjacent time period feature vectors. The training data set for the contrastive learning task refers to a data set composed of positive and negative sample pairs, used to train the contrastive learning model.

[0056] Random sampling methods can be used to select several feature vectors from the latent vector space to form a set of feature vectors. For each feature vector in the set of feature vectors, select a feature vector adjacent in time as the positive sample to form a positive sample pair. Random perturbation can be performed on the feature vector to generate a new feature vector, which is then paired with the original feature vector to form a negative sample pair; or non-adjacent time period feature vectors are paired to form a negative sample pair. Finally, the positive and negative sample pairs are combined to form a training data set for the contrastive learning task.

[0057] After that, the similarity scores of the feature vectors in the positive sample pair are calculated and compared with the similarity scores of the negative sample pair to generate a contrast loss value reflecting the consistency degree of the behavior patterns. In this step, the similarity scores of the feature vectors in the positive sample pair and the similarity scores of the feature vectors in the negative sample pair are calculated. The contrast loss value is a loss value for measuring the difference in similarity between the positive sample pair and the negative sample pair, which can reflect the consistency degree of the behavior patterns.

[0058] The similarity score between two feature vectors can be calculated using methods such as cosine similarity, Euclidean distance, etc. Then, the similarity scores of the positive sample pair are compared with the similarity scores of the negative sample pair. The contrast loss value can be calculated using a loss function, for example, using a triplet loss function, taking the difference between the similarity scores of the positive sample pair and the negative sample pair as the loss value. Through the contrast loss value, the model can learn the difference between the positive sample pair and the negative sample pair, thereby enhancing the model's ability to recognize the difference in behavior patterns.

[0059] Next, the feature vectors in the latent vector space are input into the prediction task classifier to obtain the user churn risk prediction result, and the deviation degree of the prediction result and the true label is calculated through the cross-entropy loss function to generate a prediction loss value. The prediction task classifier is a classifier for predicting user churn risk, which classifies the input feature vectors into high churn risk or low churn risk. The user churn risk prediction result refers to the prediction result of the prediction task classifier on the user churn risk. The true label refers to the actual churn risk label of the user, which can be obtained through historical data or actual observation.

[0060] Then, the contrast loss value and the prediction loss value are fused to generate a joint training loss function. The joint training loss function considers the training targets of both the contrast learning task and the prediction task. Through the joint training loss function, the model can learn the difference in behavior patterns while improving the accuracy of predicting user churn risk. For example, the contrast loss value and the prediction loss value are fused using the weighted sum method.

[0061] After that, the parameters of the time series analysis model are iteratively updated based on the joint training loss function, while optimizing the clustering tightness of the behavior pattern features in the latent vector space and the clarity of the prediction task classification boundary. According to the gradient of the joint training loss function, the parameters of the model are constantly adjusted so that the joint training loss function gradually decreases. The clustering tightness refers to the tightness of the feature vectors with similar behavior patterns in the latent vector space. The clarity of the prediction task classification boundary refers to the clarity of the high churn risk and low churn risk regions divided by the prediction task classifier in the latent vector space.

[0062] The parameters of the model are updated according to the gradient of the joint training loss function using an optimization algorithm such as stochastic gradient descent, Adam, etc. In each iteration, the gradient of the joint training loss function with respect to the model parameters is calculated, and then the model parameters are adjusted according to the direction and size of the gradient. At the same time, the clustering tightness of the behavior pattern features in the latent vector space and the clarity of the classification boundary of the prediction task are optimized. By continuously adjusting the parameters of the model, the feature vectors with similar behavior patterns are more closely clustered together in the latent vector space, and the high churn risk and low churn risk regions divided by the prediction task classifier in the latent vector space are more clear. This can improve the model's ability to recognize behavior patterns and the accuracy of predicting user churn risk.

[0063] Finally, after each parameter update, the generation strategy of positive and negative sample pairs is re-evaluated, and when it is detected that the contribution of the contrast learning task to the accuracy of the prediction task is less than the preset threshold, the temporal span of the positive sample pairs and the semantic difference of the negative sample pairs are dynamically enhanced. When the model converges, the joint training framework makes the feature vectors of high churn risk users far away from the low risk user cluster in the latent vector space, while preserving the similarity of behavior patterns of users with the same risk level. After each parameter update, the contribution of the contrast learning task to the accuracy of the prediction task is calculated, i.e. the accuracy of the prediction task classifier is compared in the presence and absence of the contrast learning task. When it is detected that the contribution of the contrast learning task to the accuracy of the prediction task is less than the preset threshold, it means that the current positive and negative sample pair generation strategy may not be effective enough and needs to be adjusted.

[0064] Step S400: Construct an interactive process parameter matrix according to the latent churn risk prediction indicator, which includes interface element weight distribution, function module priority sequence, and recommendation strategy adjustment coefficient.

[0065] The latent churn risk prediction indicator is used to predict whether a user has a churn risk and can be calculated based on the user's behavior patterns and historical data. The interactive process parameter matrix is a matrix that includes interface element weight distribution, function module priority sequence, and recommendation strategy adjustment coefficient, which can be used to optimize the interactive process of the business operation interface and improve user experience. Interface element weight distribution refers to the weight distribution obtained by evaluating the importance of each element in the business operation interface, which can be used to drive interface layout rendering. Function module priority sequence refers to the sequence obtained by sorting the priority of each function module in the business operation interface, which can be used for navigation path optimization. The recommendation strategy adjustment coefficient is a coefficient used to adjust the recommended content, which can be used for dynamic content push strategy generation.

[0066] Specifically, step S400 can specifically include the following steps S410-S450:

[0067] Step S410: Extracting user behavior deviation degree features from potential churn risk prediction indicators, the behavior deviation degree features including matching difference results of interface operation frequency distribution and historical behavior trajectory, and abnormal fluctuation coefficients of function module stay duration.

[0068] The user behavior deviation degree features refer to features for measuring the difference between the user's current behavior and historical behavior. The matching difference results of interface operation frequency distribution and historical behavior trajectory refer to the matching degree between the user's current interface operation frequency distribution and historical behavior trajectory, which can reflect whether the user's operation habits have changed. The abnormal fluctuation coefficients of function module stay duration refer to the abnormal fluctuation degree of the user's stay duration on different function modules, which can reflect whether the user's interest in different function modules has changed.

[0069] For example, the matching difference results of interface operation frequency distribution and historical behavior trajectory are calculated. The current interface operation frequency distribution can be compared with the operation frequency distribution of historical behavior trajectory to calculate the similarity score between them, and the lower the similarity score, the greater the matching difference. Then, the abnormal fluctuation coefficients of function module stay duration are determined, for example, the stay duration of the user on different function modules is statistically analyzed to calculate statistical quantities such as standard deviation or coefficient of variation, which can reflect the abnormal fluctuation degree of the stay duration. By extracting the user behavior deviation degree features, the user's behavior change can be better understood, providing a basis for subsequent interactive process parameter matrix construction.

[0070] Step S420: Based on the matching difference results of interface operation frequency distribution, the dynamic weight distribution values of each interface element are calculated, and the interface element weight distribution is generated, wherein the weight values of the areas with matching difference results lower than the preset difference threshold value are positively correlated with the increase.

[0071] The dynamic weight distribution values of interface elements refer to the weight values assigned to each interface element according to the interface operation frequency distribution and matching difference results, which can reflect the importance of interface elements. The interface element weight distribution can be used to drive the interface layout rendering. The preset difference threshold value is a pre-set threshold value for determining whether the matching difference between the interface operation frequency distribution and the historical behavior trajectory is within the normal range.

[0072] For example, the number of element clicks and the stay duration sequence in the interface operation frequency distribution are analyzed, the spatial coordinate set of the operation area exceeding the frequency threshold and the corresponding time distribution characteristics are extracted, and the initial weight mapping table is generated. The number of element clicks in the interface operation frequency distribution refers to the number of times the user clicks on each element on the interface, and the stay duration sequence refers to the time sequence of the user staying on each element. The frequency threshold is a pre-set threshold for determining whether the number of clicks on the element exceeds the normal range. The spatial coordinate set of the operation area refers to the set of position coordinates of the operation area exceeding the frequency threshold. The time distribution characteristics refer to the time distribution characteristics of the operation area exceeding the frequency threshold, such as the time interval of the operation, the peak period of the operation, etc. The initial weight mapping table is a table for recording the initial weight of each interface element, which can provide a basis for subsequent weight allocation.

[0073] Next, according to the initial weight mapping table and the region division rule of the interface layout, the density distribution gradient of each interface sub-region is calculated, and the region density distribution based on the spatial aggregation degree is generated. The initial weight mapping table is a table that records the initial weight of each interface element, and the region division rule of the interface layout refers to the rule of dividing the interface into different sub-regions. The density distribution gradient of the interface sub-region refers to the change gradient of the element weight in different sub-regions of the interface, which can reflect the aggregation degree of the element in space. The region density distribution based on the spatial aggregation degree refers to the region density distribution generated according to the aggregation degree of the element in space, which can be used for subsequent weight allocation and interface layout optimization.

[0074] Then, the matching difference result is decomposed into the difference quantization value of each interface region, the abnormal operation region exceeding the normal fluctuation range is identified through the pre-set difference threshold, and the abnormal region spatial identifier is generated. The matching difference result refers to the matching difference result of the interface operation frequency distribution and the historical behavior trajectory, which can reflect whether the user's operation habit has changed. The difference quantization value of each interface region refers to the quantization value obtained by decomposing the matching difference result to each interface region, which can more accurately reflect the change of the operation habit of each interface region. The abnormal operation region refers to the interface region whose operation habit changes beyond the normal range. The abnormal region spatial identifier is an identifier for identifying the position of the abnormal operation region, which can provide a basis for subsequent weight adjustment and interface optimization.

[0075] After that, dynamic weight distribution is performed based on the region density distribution and the abnormal region spatial identifier, wherein the initial weight value of the region exceeding the preset density threshold is adjusted by a reverse compensation mechanism of the difference quantization value to generate a dynamic weight distribution value of each interface element. The region density distribution refers to the region density distribution generated according to the aggregation degree of the element in space, which can reflect the distribution of the element in different interface regions. Dynamic weight distribution refers to real-time adjustment of the weight value of each interface element according to the region density distribution and the abnormal region spatial identifier. The preset density threshold is a pre-set threshold for judging whether the element aggregation degree of the interface region is too high. The reverse compensation mechanism of the difference quantization value refers to adjusting the initial weight value of the region when the operation habit of the interface region abnormally changes, so as to change in the opposite direction to balance the operation habit of the user.

[0076] Specifically, for each interface element, the region density distribution and the abnormal region spatial identifier of the interface region where the element is located are obtained. Then, it is judged whether the region density of the region exceeds the preset density threshold. If the preset density threshold is exceeded, it means that the element aggregation degree of the region is too high, and the initial weight value of the region needs to be adjusted. Specifically, according to the difference quantization value of the region, a reverse compensation mechanism is used for adjustment.

[0077] Finally, the dynamic weight distribution value is subjected to cross-region smoothing processing to eliminate weight mutation noise between adjacent interface elements, and an interface element weight distribution conforming to the visual continuity constraint is generated; wherein the maximum weight value of the interface element weight distribution is determined by the dynamic modification result of the region density distribution and the abnormal region spatial identifier, and the weight distribution process is real-time associated with the difference evolution state of the user operation frequency and the historical trajectory.

[0078] Cross-region smoothing processing refers to smoothing processing of the dynamic weight distribution value of adjacent interface elements to eliminate weight mutation noise. The weight mutation noise refers to the sudden change between the weight values of adjacent interface elements, which will affect the visual continuity and user experience of the interface. The visual continuity constraint refers to the principle that the weight distribution of the interface element should conform to the visual continuity, that is, the weight values of adjacent interface elements should gradually change, rather than suddenly change. The maximum weight value of the interface element weight distribution refers to the interface element weight value obtained after cross-region smoothing processing, which is determined by the dynamic modification result of the region density distribution and the abnormal region spatial identifier. The weight distribution process is real-time associated with the difference evolution state of the user operation frequency and the historical trajectory, which means that in the weight distribution process, the difference change of the user operation frequency and the historical trajectory needs to be considered in real time, and the weight value is dynamically adjusted according to the evolution state of the difference.

[0079] Step S430: According to the function module stay duration abnormal fluctuation coefficient, combined with the module switching path frequency in the user historical operation trajectory, the module transfer probability graph is constructed, and the function module priority sequence is generated by the in-degree weight ordering of the nodes in the graph.

[0080] The function module stay duration abnormal fluctuation coefficient refers to the abnormal fluctuation degree of the user's stay duration on different function modules, which can reflect whether the user's interest in different function modules has changed. The module switching path frequency in the user historical operation trajectory refers to the frequency of the user switching between different function modules, which can reflect the user's operation habit and business demand. The module transfer probability graph is a directed weighted graph, which represents the probability of the user switching from one function module to another function module. The in-degree weight of the node in the graph is the sum of the weights of the edges pointing to the node, which can reflect the importance and popularity of the function module.

[0081] Specifically, the time sequence fluctuation characteristics in the function module stay duration abnormal fluctuation coefficient can be analyzed first, the abnormal module set exceeding the preset fluctuation threshold is identified, and the abnormal module identifier and fluctuation intensity association table is generated. The time sequence fluctuation characteristics in the function module stay duration abnormal fluctuation coefficient refers to the fluctuation characteristics of the coefficient in time, which can reflect the change of the user's interest in the function module over time. The preset fluctuation threshold is a pre-set threshold for judging whether the fluctuation of the function module stay duration exceeds the normal range. The abnormal module set refers to the set of function modules whose function module stay duration fluctuation exceeds the preset fluctuation threshold. The abnormal module identifier is an identifier for identifying the abnormal module, and the fluctuation intensity refers to the fluctuation degree of the function module stay duration of the abnormal module. The abnormal module identifier and fluctuation intensity association table is a table for recording the abnormal module identifier and the corresponding fluctuation intensity, which can provide a basis for subsequent module transfer path weight matrix correction.

[0082] Then, the function module switching path sequence of continuous access in the user historical operation trajectory is extracted, the jump frequency and jump direction between modules are counted, and the module transfer path weight matrix is generated. The function module switching path sequence of continuous access in the user historical operation trajectory refers to the sequence composed of the function modules accessed in sequence by the user in the historical operation process. The jump frequency between modules refers to the number of times the user switches from one function module to another function module, and the jump direction refers to the direction of the user switching function modules. The module transfer path weight matrix is a matrix that records the weight of the user switching from one function module to another function module, and the weight can be calculated according to the jump frequency and jump direction and other factors.

[0083] For example, the historical operation trajectory data of the user is analyzed and processed, and the information of the function modules continuously accessed is extracted to form a function module switching path sequence. The function module switching path sequence is traversed, and the number of jumps and the jump direction between adjacent function modules are counted. According to the jump frequency and the jump direction, a weight is assigned to each module transfer path to form a module transfer path weight matrix. For example, the jump frequency can be used as the weight, and the higher the jump frequency, the greater the weight.

[0084] Next, the abnormal module identifier is associated with the module transfer path weight matrix, the weight decay factor of the abnormal module in the transfer path is calculated, and the module transfer path weight matrix is dynamically corrected. The association of the abnormal module identifier with the module transfer path weight matrix means that the abnormal module identifier is associated with the corresponding elements in the module transfer path weight matrix to adjust the transfer path weight of the abnormal module. The weight decay factor of the abnormal module in the transfer path refers to a factor for adjusting the transfer path weight of the abnormal module, which can be calculated according to factors such as the fluctuation intensity of the abnormal module. The dynamic correction of the module transfer path weight matrix means that according to the weight decay factor, the transfer path weight of the abnormal module in the module transfer path weight matrix is adjusted to reflect the change in the user's interest in the abnormal module. For example, the abnormal module identifier and the fluctuation intensity association table are traversed to find the elements corresponding to the abnormal module in the module transfer path weight matrix, and they are associated. The weight decay factor is calculated according to the fluctuation intensity of the abnormal module, and the greater the fluctuation intensity, the smaller the weight decay factor.

[0085] After that, a directed weighted graph structure is constructed based on the corrected module transfer path weight matrix, with function modules as nodes and corrected transfer path weights as edge weights, to generate a module transfer probability graph. For example, each function module is abstracted as a node in the graph. Then, for each non-zero element in the corrected module transfer path weight matrix, a directed edge from the starting function module node to the target function module node is added in the graph, and the value of the element is taken as the weight of the edge.

[0086] Then, the computing module calculates the in-degree comprehensive weight of each node in the module transition probability graph. The in-degree comprehensive weight is generated by weighting and fusing the edge weight pointing to the node and the volatility intensity of the corresponding source node. The in-degree comprehensive weight is a weight value obtained by comprehensively considering the weight of the edge pointing to the node. The edge weight pointing to the node reflects the possibility of the transition of other functional modules to the node, and the volatility intensity of the corresponding source node reflects the stability and user attention of the source node. By weighting and fusing these two factors, the importance of the node can be more accurately evaluated. For example, for each node in the graph, all edges pointing to the node are traversed. For each edge, its edge weight and the volatility intensity of the corresponding source node are obtained. The edge weight and the volatility intensity are respectively assigned a weight coefficient, then the edge weight is multiplied by the edge weight coefficient, the volatility intensity is multiplied by the volatility intensity coefficient, and the two are added to obtain the contribution value of the edge to the in-degree comprehensive weight of the target node. The contribution values of all edges pointing to the node are added to obtain the in-degree comprehensive weight of the node.

[0087] Finally, the nodes are arranged in descending order according to the in-degree comprehensive weight, a functional module priority sequence is generated, and a redundant elimination process is performed on the loop path nodes in the sequence; wherein the ranking result of the functional module at the preset priority in the functional module priority sequence simultaneously satisfies the core hub characteristics of high in-degree weight, low volatility abnormality and high frequency access path.

[0088] Loop path nodes refer to nodes that form a loop in the module transition probability graph. The preset priority is a preset priority range. The functional modules within the range should simultaneously have the core hub characteristics of high in-degree weight, low volatility abnormality and high frequency access path. For example, the in-degree comprehensive weights of all nodes in the module transition probability graph are extracted, and a sorting algorithm (such as quicksort, mergesort, etc.) is used to arrange these weights in descending order. According to the sorting result, a functional module priority sequence is obtained.

[0089] Step S440: Extract the service preference tag set in the user portrait, perform semantic matching between the service preference tag and the functional module priority sequence, calculate the association strength between the recommended content and the target module, and generate a recommended strategy adjustment coefficient.

[0090] The business preference label set in the user portrait refers to a set of labels reflecting the user's business preferences extracted by analyzing and mining the user's historical operation data, behavior patterns, etc. These labels can include information such as the user's interest in business areas, function modules, product types, etc. Semantic matching refers to matching the business preference labels with the modules in the function module priority sequence in terms of semantics to determine their relevance. The association strength of the recommended content and the target module refers to the degree of relevance between the recommended content and the function module, which can be calculated through the results of semantic matching. The recommendation strategy adjustment coefficient is a coefficient used to adjust the recommendation strategy, which can be generated according to the association strength of the recommended content and the target module to optimize the push effect of the recommended content.

[0091] For example, first, extract the function module identifier set in the historical operation record that exceeds the preset access frequency threshold from the user portrait, and generate the business preference label set in combination with the user's actively labeled preference data. The preset access frequency threshold is a pre-set threshold used to filter out the function modules frequently accessed by the user. The function module identifier set refers to a set of identifiers of function modules that exceed the preset access frequency threshold. The business preference label set is a label set composed of the function module identifier set and the user's actively labeled preference data, which can more comprehensively reflect the user's business preferences.

[0092] Then, align the module identifiers in the function module priority sequence with the business preference label set in terms of semantic features, extract the context-related word vectors of the module description text and the preference label, and generate a semantic association degree matrix. The module identifier in the function module priority sequence refers to the identifier of the function module sorted by priority, which is used to identify each function module. Semantic feature alignment refers to aligning the module identifier with the labels in the business preference label set in terms of semantics to determine their relevance. The module description text refers to the text describing the function, purpose, etc. of the function module, and the preference label is the label in the business preference label set. The context-related word vector refers to the vector representation of the related words extracted by analyzing the context of the module description text and the preference label. The semantic association degree matrix is a matrix that records the semantic association degree between the function module and the business preference label.

[0093] For example, for each module identifier in the function module priority sequence, find the corresponding module description text. Process the module description text and the labels in the business preference label set using natural language processing techniques to extract their semantic features. Align the semantic features of the module description text with the semantic features of the business preference labels and calculate their similarity. Analyze the module description text and the preference labels using context analysis algorithms (such as dependency syntax analysis, co-occurrence analysis, etc.) to extract their contextual association words. Convert these association words into vector representations to obtain the contextual association word vectors. The semantic association degree between each function module and each business preference label is used as the elements of the matrix to form a semantic association degree matrix. The rows of the matrix represent the function modules, the columns represent the business preference labels, and the values of the elements represent the semantic association degree between the function modules and the business preference labels.

[0094] After that, the matching weight values of each function module and business preference label are calculated based on the semantic association degree matrix, wherein the matching weight of the function module at the preset priority is adjusted by the product of the label coverage and the semantic similarity. The matching weight value refers to the weight value used to measure the matching degree between the function module and the business preference label, which can be calculated according to the semantic association degree matrix. The label coverage refers to the proportion of the number of labels related to a certain function module in the business preference label set to the total number of labels, which reflects the coverage of the business preference label to the function module. The semantic similarity refers to the semantic similarity between the function module and the business preference label, which can be represented by the element value in the semantic association degree matrix. The function module at the preset priority refers to the function module within the preset priority range in the function module priority sequence.

[0095] Then, the matching weight values are normalized to generate the association strength distribution of the recommended content and the target module in combination with the sorting results of the function module priority sequence, and the association strength distribution reflects the recommendation adaptability of different priority modules in the user preference dimension. Normalization (such as linear normalization) can convert the matching weight values to a preset range (such as [0, 1]) for comparison and analysis. The association strength distribution refers to the distribution of the association strength between the recommended content and the target module on different priority modules, which can reflect the recommendation adaptability of different priority modules in the user preference dimension.

[0096] Then, the recommendation trigger condition is dynamically set according to the correlation strength distribution, which can specifically include the following steps: a recommendation content display frequency threshold, an inter-module recommendation interval period, and a content presentation form priority, and a recommendation strategy adjustment coefficient containing multi-dimensional control parameters is generated. The correlation strength distribution reflects the correlation degree between the recommended content and different priority modules. According to this distribution, the recommendation trigger condition can be dynamically set. The recommendation content display frequency threshold refers to the maximum number of times the recommended content is displayed within a certain time period. It can control the display frequency of the recommended content and avoid excessive recommendation to the user. The inter-module recommendation interval period refers to the time interval between different functional modules for recommendation, which can ensure the rationality and coherence of the recommended content. The content presentation form priority refers to the priority order of different content presentation forms (such as text, pictures, videos, etc.), which can select the most suitable content presentation form according to the user's preferences and operation scenarios. The multi-dimensional control parameters refer to a parameter set containing multiple parameters such as the recommendation content display frequency threshold, the inter-module recommendation interval period, and the content presentation form priority. The recommendation strategy adjustment coefficient is a coefficient generated according to these multi-dimensional control parameters, which can be used to adjust the recommendation strategy and improve the recommendation effect.

[0097] Finally, the recommendation strategy adjustment coefficient is associated with real-time user operation behavior data for verification. When the correlation strength between the recommended content and the current operation module is detected to be lower than the preset threshold, the dynamic redistribution mechanism of the recommendation strategy adjustment coefficient is triggered. The generation process of the recommendation strategy adjustment coefficient is synchronized with the dynamic update state of the functional module priority sequence and the incremental learning result of the user portrait, ensuring the consistency of the recommendation strategy and the evolution of user behavior. Real-time user operation behavior data refers to the behavior data of the user in the current operation process, including the functional modules visited by the user, the business data operated, the buttons clicked, and other information. Association verification refers to the association of the recommendation strategy adjustment coefficient with real-time user operation behavior data to verify the effectiveness of the recommendation strategy adjustment coefficient. The correlation strength between the recommended content and the current operation module refers to the correlation degree between the recommended content and the functional module currently operated by the user. The preset threshold is a pre-set threshold for determining whether the correlation strength is sufficient. The dynamic redistribution mechanism of the recommendation strategy adjustment coefficient refers to the recalculation and redistribution of the recommendation strategy adjustment coefficient when the correlation strength between the recommended content and the current operation module is lower than the preset threshold, in order to adjust the recommendation strategy. The dynamic update state of the functional module priority sequence refers to the state of the functional module priority sequence being updated with the change of user behavior. The incremental learning result of the user portrait refers to the updated result of the user portrait obtained by continuously learning new user behavior data.

[0098] Step S450: Three-dimensional tensor integration is performed on the interface element weight distribution, the function module priority sequence, and the recommendation strategy adjustment coefficient. Orthogonal projection is used to eliminate cross interference between dimensions, and an interactive process parameter matrix is generated. The interface element weight distribution is used to drive interface layout rendering, the function module priority sequence is used for navigation path optimization, and the recommendation strategy adjustment coefficient is used for dynamic content push strategy generation.

[0099] Three-dimensional tensor integration refers to integrating the interface element weight distribution, the function module priority sequence, and the recommendation strategy adjustment coefficient, which are data of different dimensions, into a three-dimensional tensor. A three-dimensional tensor is a multidimensional array that can represent data information of three dimensions at the same time. For example, the interface element weight distribution, the function module priority sequence, and the recommendation strategy adjustment coefficient are taken as the three dimensions of the three-dimensional tensor. For example, the interface element weight distribution is taken as the first dimension, the function module priority sequence is taken as the second dimension, and the recommendation strategy adjustment coefficient is taken as the third dimension.

[0100] Step S500: The optimized interactive process parameter matrix is called to drive the reconstruction of the business operation interface, an interactive interface adapted to the current user behavior mode is generated, and the time series analysis model is iteratively updated through the incremental feedback mechanism within a preset verification period.

[0101] The optimized interactive process parameter matrix refers to the interactive process parameter matrix obtained after processing and optimization in the previous steps, which contains information such as interface element weight distribution, function module priority sequence, and recommendation strategy adjustment coefficient. Business operation interface reconstruction refers to rearranging and designing the business operation interface according to the interactive process parameter matrix to improve the usability and user experience of the interface. The interactive interface adapted to the current user behavior mode refers to the interactive interface generated according to the current user behavior mode, which can better meet the user's needs and operation habits. The preset verification period is a pre-set time period for verifying and evaluating the generated interactive interface. The incremental feedback mechanism is a mechanism for updating the model, which collects user feedback information to update the model incrementally to improve the performance and adaptability of the model. The time series analysis model is a model for analyzing user behavior patterns and predicting potential churn risks, which is iteratively updated through the incremental feedback mechanism to better adapt to changes in user behavior.

[0102] As an implementation, the above-mentioned interactive process parameter matrix contains interface element weight distribution, function module priority sequence, and recommendation strategy adjustment coefficient. Based on this, in step S500, the optimized interactive process parameter matrix is called to drive the reconstruction of the business operation interface, which can specifically include the following steps S510-S550:

[0103] Step S510: Analyze the interface element weight distribution in the interaction flow parameter matrix, extract the spatial coordinate set of the area exceeding the weight threshold and the visual rendering priority parameter, and generate the dynamic focus instruction of the core function control.

[0104] The weight threshold is a pre-set threshold for screening important interface elements. The spatial coordinate set of the area exceeding the weight threshold refers to the set of position coordinates of the area in the interface whose weight value exceeds the threshold. The visual rendering priority parameter refers to the parameter for controlling the visual rendering order and effect of the interface elements. The core function control refers to the control that provides the main business function. The dynamic focus instruction refers to the instruction for controlling the dynamic focus of the core function control, which can make the core function control more prominent in the interface and improve the user's operation efficiency.

[0105] The process of analyzing the interface element weight distribution in the interaction flow parameter matrix is as follows: analyze and process the interface element weight distribution data in the interaction flow parameter matrix, and extract the weight value of each interface element. Set a weight threshold to screen out interface elements whose weight value exceeds the threshold. Extract the spatial coordinate set of the area where these interface elements are located and the visual rendering priority parameters related to them.

[0106] For example, according to the spatial coordinate set of the area exceeding the weight threshold and the visual rendering priority parameter, the dynamic focus instruction of the core function control is generated. The dynamic focus instruction can include the position, size, color, transparency, etc. of the core function control, as well as the time and manner of focusing, etc. For example, the core function control can be set to automatically focus when the user enters the interface, or focus when the user operates. By generating the dynamic focus instruction, the core function control can be more prominent in the interface, attracting the user's attention and improving the user's operation efficiency.

[0107] Step S520: According to the sorting result of the function module priority sequence, calculate the access path weight of each module node in the navigation menu, and generate the shortest path topology structure based on the weight gradient distribution.

[0108] The sorting result of the function module priority sequence refers to the sequence obtained by sorting the function modules according to the priority from high to low. The access path weight of each module node in the navigation menu refers to the weight of the access path from the starting node of the navigation menu to each module node, which can reflect the difficulty and importance of the access path. The weight gradient distribution refers to the distribution of the access path weight in the navigation menu, which can reflect the trend of the access path weight. The shortest path topology structure refers to the topology structure composed of the shortest access path from the starting node to each module node in the navigation menu, which can optimize the user's navigation path and improve the user's operation efficiency.

[0109] Specifically, first, the sorting result of the function module priority sequence is parsed, the in-degree weight of the module node and the historical statistical value of the transition frequency between adjacent modules are extracted, and the initial access path weight of each module node is generated. The sorting result of the function module priority sequence refers to the sequence obtained by sorting the function modules from high to low according to the priority. The in-degree weight of the module node refers to the sum of the weights of the edges pointing to the module node, which can reflect the importance of the module node. The historical statistical value of the transition frequency between adjacent modules refers to the statistical value of the number of transitions between adjacent modules by the user, which can reflect the degree of association between modules. The initial access path weight refers to the access path weight of each module node calculated according to the in-degree weight of the module node and the historical statistical value of the transition frequency between adjacent modules, which provides a basis for subsequent path weight correction and shortest path topology generation.

[0110] Then, the initial access path weight is corrected according to the weight difference gradient of adjacent modules in the weight gradient distribution, wherein the path weight of the module node in the area exceeding the preset gradient change is suppressed by a decay factor to inhibit abnormal growth of the weight. The weight gradient distribution refers to the distribution of the access path weight in the navigation menu, which can be obtained by calculating the weight difference between adjacent module nodes. The weight difference gradient of adjacent modules refers to the degree of difference between the access path weights of adjacent module nodes, which can reflect the trend of weight change. The initial access path weight refers to the access path weight of each module node generated in the above steps. The preset gradient change is a threshold value set in advance to determine whether the weight difference of adjacent modules is too large. The decay factor is a coefficient less than 1, which is used to suppress the abnormal growth of the path weight of the module node in the area exceeding the preset gradient change.

[0111] For example, first, the weight difference gradient of adjacent modules in the weight gradient distribution is calculated, and the weight difference gradient is compared with the preset gradient change. If the weight difference gradient exceeds the preset gradient change, it indicates that the path weight of the module node in this area has an abnormal growth. For the module node in the area exceeding the preset gradient change, the path weight is corrected using a decay factor (which can be set according to actual conditions). In this way, the abnormal growth of the path weight of the module node can be suppressed, and the access path weight can be more reasonable and stable.

[0112] Next, a directed weighted graph structure is constructed based on the corrected access path weight to generate an initial shortest path topology structure, with the module nodes as vertices and the path weight as edge weight. For example, each module node is abstracted as a vertex in the graph. Then, for each non-zero element in the corrected access path weight matrix, a directed edge from the starting module node to the target module node is added in the graph, and the value of the element is taken as the weight of the edge. In this way, a directed weighted graph structure can be constructed, and an initial shortest path topology structure can be obtained.

[0113] After that, the initial shortest path topology is verified for redundant paths, identifying jump nodes that violate the monotonicity of the weight gradient change in the initial shortest path topology, and obtaining a reconstructed path topology by inserting intermediate module nodes or adjusting edge weight distribution. Redundant path verification refers to checking the paths in the initial shortest path topology, removing unnecessary paths to simplify the topology. The monotonicity of the weight gradient change refers to the trend that the access path weight should show a monotonic increasing or decreasing trend on the path. The jump node refers to a node in the initial shortest path topology that violates the monotonicity of the weight gradient change, which may cause unreasonable paths. The reconstructed path topology refers to the path topology obtained after redundant path verification and node adjustment, which is more reasonable and efficient.

[0114] For example, all paths in the initial shortest path topology are traversed to check whether the access path weight on the path meets the monotonicity of the weight gradient change. If a jump node is found that violates the monotonicity of the weight gradient change, the jump node is identified. For the jump node, it can be processed by inserting an intermediate module node or adjusting the edge weight distribution. Through such processing, a reconstructed path topology can be obtained to improve the rationality and efficiency of the path.

[0115] After that, the reconstructed path topology is mapped to the guide sequence in the navigation menu, and the visual parameters of the guide sequence match the weight gradient distribution trend of each node in the reconstructed path topology. The reconstructed path topology refers to the path topology obtained after redundant path verification and node adjustment, which shows the shortest access path from the starting node of the navigation menu to each module node. The guide sequence in the navigation menu refers to the sequence used to guide the user's operation in the navigation menu, which can indicate the user's operation path through visual effects such as color, size, transparency, etc. The visual parameter refers to the parameter used to control the visual effect of the guide sequence, such as color, size, transparency, etc. The weight gradient distribution trend refers to the change trend of the access path weight of each node in the reconstructed path topology.

[0116] For example, first determine the position and layout of each module node in the navigation menu. Then, according to the reconstructed path topology, assign a position in the guide sequence to each module node. According to the weight gradient distribution trend of each node in the reconstructed path topology, set the visual parameters of the guide sequence. In this way, the reconstructed path topology is mapped to the guide sequence in the navigation menu, and the visual parameters of the guide sequence match the weight gradient distribution trend of each node in the reconstructed path topology, allowing users to more intuitively understand the navigation path and improve operation efficiency.

[0117] Finally, when the user performs a module switching operation, the path deviation index is collected in real time, the access path weight is updated based on the current weight gradient distribution, and the update result is fed back to the sorting logic of the function module priority sequence; wherein the calculation process of the access path weight realizes the dynamic balance of the path weight between the module nodes through the weight gradient distribution, ensuring the consistency of the shortest path topology structure and the evolution of user behavior patterns. The path deviation index refers to the deviation between the actual access path and the path in the shortest path topology structure when the user performs a module switching operation. It can be obtained by calculating the difference between the length of the actual access path and the length of the shortest path, or calculating the difference between the nodes on the actual access path and the nodes on the shortest path. The current weight gradient distribution refers to the gradient distribution of the access path weight of each module node in the navigation menu when the user performs a module switching operation. Updating the access path weight refers to adjusting the access path weight of each module node in the navigation menu according to the path deviation index and the current weight gradient distribution. The sorting logic of the function module priority sequence refers to the logic of sorting the function modules according to their importance and the user's operation habits.

[0118] For example, when the user performs a module switching operation, the user's actual access path is recorded. The actual access path is compared with the path in the shortest path topology structure to calculate the path deviation index. According to the path deviation index and the current weight gradient distribution, the access path weight of each module node in the navigation menu is adjusted. The updated access path weight is fed back to the sorting logic of the function module priority sequence, and the sorting logic reorders the function modules according to the new access path weight. The calculation process of the access path weight realizes the dynamic balance of the path weight between the module nodes through the weight gradient distribution, ensuring the consistency of the shortest path topology structure and the evolution of user behavior patterns.

[0119] Step S530: input the recommended strategy adjustment coefficient into the pre-trained recommendation rule generator, match the user's real-time operation scene with the historical preference features, and output the personalized recommendation content sequence associated with the current function module.

[0120] The recommendation strategy adjustment coefficient is a coefficient for adjusting the recommendation strategy, which includes multiple-dimensional regulation and control parameters such as recommended content display frequency threshold, recommended interval period between modules, and content presentation form priority. The pre-trained recommendation rule generator is a trained model that can generate recommendation rules according to the input recommendation strategy adjustment coefficient and user-related information. The user real-time operation scene refers to the scene information of the user in the current operation process, including the functional module being operated by the user, the operation time, the operation environment, etc. The historical preference feature refers to the preference information exhibited by the user in the historical operation process, which can be represented by the business preference tag set in the user portrait. The personalized recommendation content sequence refers to the recommendation content sequence associated with the current functional module generated according to the user's real-time operation scene and historical preference feature, which can meet the user's personalized needs.

[0121] Step S540: Integrate the dynamic focus instruction, the shortest path topology structure, and the personalized recommendation content sequence to generate an interface layout rendering configuration file. The configuration file includes control positioning coordinates, path guide marks, and spatiotemporal distribution constraints of recommended content plugins.

[0122] The dynamic focus instruction is an instruction for controlling the dynamic focus of the core functional control, which can make the core functional control more prominent in the interface and improve the user's operation efficiency. The shortest path topology structure is a topology structure composed of the shortest access paths from the starting node to each module node in the navigation menu, which can optimize the user's navigation path. The personalized recommendation content sequence is a recommendation content sequence associated with the current functional module generated according to the user's real-time operation scene and historical preference feature, which can meet the user's personalized needs. The interface layout rendering configuration file is a file for configuring the interface layout rendering, which includes control positioning coordinates, path guide marks, and spatiotemporal distribution constraints of recommended content plugins, etc. information for guiding the rendering and display of the interface.

[0123] For example, the position, size, color, transparency, etc. of the core function control in the dynamic focus instruction are sorted. Then, the navigation path information in the shortest path topology structure, such as node position, edge weight, etc. is extracted. Next, the recommended content information in the personalized recommendation content sequence, such as the title, description, picture, etc. of the recommended content is sorted. These information is integrated to form a data set containing control positioning coordinates, path guide marks and spatiotemporal distribution constraints of recommended content plugins. According to the integrated data set, the interface layout rendering configuration file is generated. The configuration file contains control positioning coordinates, which are used to determine the position of the core function control and other interface elements in the interface; path guide marks are used to indicate the user's operation path in the navigation menu; spatiotemporal distribution constraints of recommended content plugins are used to control the display time and position of recommended content in the interface. By generating the interface layout rendering configuration file, the dynamic focus instruction, the shortest path topology structure and the personalized recommendation content sequence can be combined to realize the optimization of the interface layout and the display of the personalized recommendation content.

[0124] Step S550: After loading the interface layout rendering configuration file on the target client, the user's operation response time delay, path switching efficiency and recommended content adoption rate indicators for the core function control are collected in real time, the interface optimization feedback signal is generated and returned to the update queue of the interactive process parameter matrix; wherein the generation process of the interface layout rendering configuration file is synchronously associated with the dynamic change trend of the interface element weight distribution and the real-time sorting state of the function module priority sequence, to ensure the consistency of the interface reconstruction strategy and the user behavior pattern evolution.

[0125] The target client refers to a client device that operates a business operation interface, such as a computer, a mobile phone, etc. The interface layout rendering configuration file is a file used to configure the interface layout rendering, which contains control positioning coordinates, path guide marks and spatiotemporal distribution constraints of recommended content plugins, etc. The user's operation response time delay for the core function control refers to the time spent by the user from triggering the core function control to the control responding, which can reflect the response speed and availability of the core function control. The path switching efficiency refers to the efficiency of the user's module switching operation in the navigation menu, which can be measured by the time and path deviation degree of the user's module switching operation. The recommended content adoption rate indicator refers to the user's acceptance of the recommended content, which can be calculated by the ratio of the number of times the user clicks and uses the recommended content to the number of times the recommended content is displayed. The interface optimization feedback signal is a signal used to feedback the interface optimization situation, which contains indicators such as the user's operation response time delay for the core function control, the path switching efficiency and the recommended content adoption rate. The update queue of the interactive process parameter matrix is a queue used to store the update instructions of the interactive process parameter matrix, which can ensure that the update operation of the interactive process parameter matrix is performed in sequence.

[0126] For example, the interface element loading delay abnormal event in the user operation response time delay is analyzed first, the control coordinate set exceeding the preset delay trigger threshold and the corresponding function module identifier are identified, and the interface rendering performance bottleneck area label is generated.

[0127] The user operation response time delay refers to the time taken by the user from triggering the interface element operation to the element responding. The interface element loading delay abnormal event refers to the abnormal event that the interface element loading time is too long in the user operation response time delay. The preset delay trigger threshold is a pre-set threshold for judging whether the interface element loading delay is abnormal. The control coordinate set refers to the set of position coordinates of the interface controls exceeding the preset delay trigger threshold. The corresponding function module identifier refers to the identifier of the function module associated with the interface control exceeding the preset delay trigger threshold. The interface rendering performance bottleneck area label is a label for marking the rendering performance bottleneck area in the interface, which can provide a basis for subsequent interface optimization.

[0128] The user operation response time delay data is analyzed and processed, and the information of the interface element loading delay is extracted. The loading delay time of each interface element is compared with the preset delay trigger threshold, and when the loading delay time exceeds the preset delay trigger threshold, the loading event of the interface element is identified as an interface element loading delay abnormal event. For each interface element loading delay abnormal event, the control coordinates and the corresponding function module identifier of the interface element are extracted. These control coordinates and corresponding function module identifiers are collected respectively to form the control coordinate set and the corresponding function module identifier set. According to the control coordinate set and the corresponding function module identifier set, the interface rendering performance bottleneck area label is generated. The coordinate points in the control coordinate set can be marked on the interface, and the corresponding function module identifier is labeled.

[0129] Then, the path jump time consumption and the number of deviations in the path switching efficiency index are extracted, and the matching deviation degree of the path guidance strategy and the user's actual operation habit is calculated in combination with the node distribution characteristics of the shortest path topology structure.

[0130] The path switching efficiency index refers to an index for measuring the efficiency of the user's module switching operation in the navigation menu, which includes path jump time consumption and deviation times and the like. The path jump time consumption refers to the time spent by the user in switching from one functional module to another functional module, and the deviation times refer to the number of times the user deviates from the shortest path in the switching path. The node distribution characteristics of the shortest path topology refer to the position, connection relationship and the like of each node in the shortest path topology. The path guidance strategy is a strategy for guiding the user to perform the module switching operation in the navigation menu, which can be realized through path guidance marks and the like. The matching deviation degree refers to the difference between the path guidance strategy and the actual operation habit of the user, which can reflect the effectiveness of the path guidance strategy.

[0131] For example, the path switching efficiency index data is analyzed and processed to extract the statistical values of the path jump time consumption and the deviation times. The start time and the end time of each module switching operation of the user can be recorded to calculate the path jump time consumption. At the same time, the number of times the user deviates from the shortest path in the switching path is recorded to obtain the statistical value of the deviation times. The node distribution characteristics of the shortest path topology are analyzed to understand the direction of the shortest path and the connection relationship of the nodes. Then, the actual operation path of the user is compared with the shortest path topology to calculate the matching deviation degree between the path guidance strategy and the actual operation habit of the user. The path similarity, path deviation distance and the like can be used to measure the matching deviation degree.

[0132] Then, the recommended content adoption rate index is decomposed into the ratio of the exposure times of the recommended content associated with different functional modules to the number of times of the user's active interaction, to generate a module-level recommendation utility evaluation matrix. The recommended content adoption rate index refers to the acceptance degree of the user to the recommended content, which can be calculated by the ratio of the number of times of the user's clicking and using the recommended content to the number of times of the recommended content display. The recommended content associated with different functional modules refers to the recommended content associated with each functional module. The exposure times refer to the number of times of the display of the recommended content on the interface, and the number of times of the user's active interaction refers to the number of times of the user's active operation such as clicking and using the recommended content. The module-level recommendation utility evaluation matrix is a matrix that records the recommendation utility evaluation results of the recommended content associated with different functional modules, the rows of the matrix represent the functional modules, the columns represent the recommended content, and the values of the elements represent the recommendation utility evaluation values of the recommended content associated with the functional module.

[0133] For example, the recommended content adoption rate index data is analyzed and processed, and the recommended content is classified according to the function modules. For the recommended content associated with each function module, the number of exposures and the number of user active interactions are counted. The number of user active interactions is divided by the number of exposures to obtain the recommended content adoption rate of the recommended content associated with the function module. The recommended content adoption rate of the recommended content associated with each function module is taken as an element of a matrix to form a module-level recommended utility evaluation matrix. The rows of the matrix represent the function modules, the columns represent the recommended content, and the values of the elements represent the recommended utility evaluation values of the recommended content associated with the function module.

[0134] After that, the interface rendering performance bottleneck area mark, the matching deviation degree and the recommended utility evaluation matrix are subjected to multi-dimensional correlation analysis to determine the collaborative optimization direction of the interface element weight distribution, the function module priority sequence and the recommended strategy adjustment coefficient in the interaction flow parameter matrix.

[0135] Multi-dimensional correlation analysis refers to correlating and analyzing data in multiple dimensions to discover the relationships and patterns between the data. In this step, the data in three dimensions of the interface rendering performance bottleneck area mark, the matching deviation degree and the recommended utility evaluation matrix are subjected to correlation analysis. The interface rendering performance bottleneck area mark reflects the areas with poor rendering performance in the interface, the matching deviation degree reflects the difference between the path guidance strategy and the user's actual operation habits, and the recommended utility evaluation matrix reflects the recommended utility of the recommended content associated with different function modules. The interaction flow parameter matrix contains information such as the interface element weight distribution, the function module priority sequence and the recommended strategy adjustment coefficient, and the collaborative optimization direction refers to the direction of optimizing these parameters to improve the usability and user experience of the interface.

[0136] For example, the data in the interface rendering performance bottleneck area mark, the matching deviation degree and the recommended utility evaluation matrix are sorted and preprocessed to make them comparable. Then, correlation analysis methods such as correlation analysis, clustering analysis, etc. are used to analyze these data to find their correlation. For example, the relationship between the interface rendering performance bottleneck area and the matching deviation degree, and the relationship between the matching deviation degree and the recommended utility evaluation matrix can be analyzed. According to the results of multi-dimensional correlation analysis, the optimization direction of the interface element weight distribution, the function module priority sequence and the recommended strategy adjustment coefficient in the interaction flow parameter matrix is determined. For example, if it is found that the interface rendering performance bottleneck area is related to the excessive weight of certain interface elements, the weight of these interface elements can be appropriately reduced; if it is found that the large matching deviation degree is related to the unreasonable function module priority sequence, the function module priority sequence can be adjusted; if it is found that the recommended utility of certain recommended content in the recommended utility evaluation matrix is low and related to the unreasonable recommended strategy adjustment coefficient, the recommended strategy adjustment coefficient can be adjusted. By collaboratively optimizing these parameters, the usability and user experience of the interface can be improved.

[0137] Then, based on the collaborative optimization direction, a parameter matrix update instruction is generated, which includes an interface element weight decay coefficient, a module priority order correction amplitude, and a recommended strategy trigger threshold adjustment amount.

[0138] The collaborative optimization direction refers to the optimization direction of the interface element weight distribution, the function module priority sequence, and the recommended strategy adjustment coefficient in the interaction process parameter matrix determined through multi-dimensional correlation analysis of the interface rendering performance bottleneck area marking, the matching deviation degree, and the recommended utility evaluation matrix. The parameter matrix update instruction is an instruction for updating the interaction process parameter matrix, which includes specific information for adjusting the interface element weight distribution, the function module priority sequence, and the recommended strategy adjustment coefficient. The interface element weight decay coefficient is a coefficient for adjusting the weight of the interface element, which can make the weight of the interface element decay in proportion to optimize the interface layout. The module priority order correction amplitude refers to the amplitude of the correction of the function module priority sequence, which can change the priority order of the function module to improve the rationality of the navigation path. The recommended strategy trigger threshold adjustment amount refers to the amount of adjustment of the recommended strategy trigger threshold, which can change the display conditions of the recommended content to improve the relevance and effectiveness of the recommended content.

[0139] For example, according to the collaborative optimization direction, the adjustment direction and amplitude of the interface element weight distribution, the function module priority sequence, and the recommended strategy adjustment coefficient are determined. For the interface element weight distribution, according to the analysis results of the interface rendering performance bottleneck area marking and the matching deviation degree, the interface elements that need to be adjusted are determined, and the corresponding interface element weight decay coefficient is calculated. For example, if the area where a certain interface element is located is a rendering performance bottleneck area, or does not match the user's operation habits, the weight of the element can be appropriately reduced by setting an interface element weight decay coefficient less than 1. For the function module priority sequence, according to the analysis results of the matching deviation degree and the recommended utility evaluation matrix, the function modules that need to be corrected in priority order are determined, and the module priority order correction amplitude is calculated. For example, if the recommended utility of a certain function module is low, or the user often deviates from the path of the module during operation, the priority of the module can be appropriately reduced by setting a suitable module priority order correction amplitude. For the recommended strategy adjustment coefficient, according to the analysis results of the recommended utility evaluation matrix, the recommended strategy trigger threshold that needs to be adjusted is determined, and the recommended strategy trigger threshold adjustment amount is calculated. For example, if the adoption rate of the recommended content associated with a certain function module is low, the recommended strategy trigger threshold of the module can be appropriately reduced by setting a suitable recommended strategy trigger threshold adjustment amount. The interface element weight decay coefficient, the module priority order correction amplitude, and the recommended strategy trigger threshold adjustment amount are integrated to form the parameter matrix update instruction.

[0140] Next, the parameter matrix update instructions are stored in the update queue in priority order. When it is detected that the dynamic change amplitude of the interface element weight distribution exceeds the preset threshold, real-time incremental updating of the parameter matrix is triggered.

[0141] The update queue is a queue for storing parameter matrix update instructions, which can ensure that the update instructions are processed in order. Storing parameter matrix update instructions in priority order means that the update instructions are prioritized according to their importance and urgency, and stored in the update queue. This ensures that important and urgent update instructions are processed first. The dynamic change amplitude of the interface element weight distribution refers to the degree of change in the interface element weight distribution over a period of time, which can be measured by calculating the rate of change of the interface element weight, standard deviation, etc. The preset threshold is a pre-set threshold for determining whether the dynamic change amplitude of the interface element weight distribution is too large. Real-time incremental updating of the parameter matrix refers to updating the interactive process parameter matrix immediately when it is detected that the dynamic change amplitude of the interface element weight distribution exceeds the preset threshold, in order to ensure the usability and user experience of the interface.

[0142] For example, each parameter matrix update instruction is assigned a priority. The priority can be determined according to the degree of influence of the update instruction on the interface performance, user experience, etc. For example, adjustment instructions for the bottleneck area of the interface rendering performance can be assigned a higher priority, while adjustment instructions for some secondary interface elements can be assigned a lower priority. Then, the update instructions are stored in the update queue in order of priority from high to low. When a new update instruction arrives, it is inserted into the appropriate position in the update queue to ensure the priority order of the queue.

[0143] The dynamic change amplitude of the interface element weight distribution is monitored in real time. The rate of change of the interface element weight, standard deviation, etc. can be calculated periodically and compared with the preset threshold. When the dynamic change amplitude is detected to exceed the preset threshold, it indicates that the interface element weight distribution has changed significantly, which may affect the usability and user experience of the interface. At this time, the update instruction with the highest priority is taken out of the update queue, and the interactive process parameter matrix is updated in real time. During the updating process, the interface element weight distribution, functional module priority sequence, and recommendation strategy adjustment coefficient are adjusted according to the interface element weight decay coefficient, module priority order correction amplitude, and recommended strategy trigger threshold adjustment amount in the update instruction. Through real-time incremental updating, the changes in the interface element weight distribution can be responded to in a timely manner, ensuring the stability and user experience of the interface.

[0144] Finally, after performing the incremental update, the user operation data is re-collected to generate a new round of interface optimization feedback signals, forming a closed-loop optimization link; wherein the generation logic of the interface optimization feedback signals is synchronized with the version iteration record of the interface layout rendering configuration file and the user behavior pattern evolution trajectory, ensuring the dynamic adaptation of the parameter matrix update strategy and the system performance indicators.

[0145] For example, the real-time change trend of the interface element weight distribution and the function module priority sequence after the incremental update can be monitored, and the spatial coordinates of the abnormal fluctuation nodes in the weight gradient distribution and the associated module identifiers are extracted. The function module priority sequence is the sequence obtained by sorting each function module according to a preset rule, which reflects the importance and use priority of the function module. The weight gradient distribution describes the change of the interface element weight in space, and the abnormal fluctuation node refers to the node in the weight gradient distribution that deviates significantly from the normal change trend. The spatial coordinates are used to locate the specific position on the interface, and the associated module identifier is used to uniquely identify the function module associated with the abnormal fluctuation node.

[0146] In order to monitor the real-time change trend of the interface element weight distribution and the function module priority sequence, real-time data collection and analysis methods can be used. For example, sensors or log recording tools are used to collect user interaction data with the interface in real time, including clicking, sliding, etc. Then, through data analysis algorithms such as time series analysis algorithms, these data are processed to obtain the real-time change of the interface element weight distribution and the function module priority sequence.

[0147] In extracting the spatial coordinates of the abnormal fluctuation nodes in the weight gradient distribution and the associated module identifiers, threshold judgment method can be used. First, calculate the mean and standard deviation of the weight gradient distribution, then set a threshold, when the weight gradient value of a node exceeds the threshold, it is judged as an abnormal fluctuation node. Then, through the interface layout information and the function module association information, the spatial coordinates and the associated module identifiers of the abnormal fluctuation nodes are obtained. For example, assuming that the weight gradient value of a certain interface element is 10, and the mean of the weight gradient distribution is 2, the standard deviation is 1, and the threshold is set to 3 times the standard deviation, i.e. 6, then the node is judged as an abnormal fluctuation node. Through the interface layout file, the spatial coordinates of the node are obtained as (x, y), and through the function module association table, the associated module identifier M1 is determined.

[0148] Then, capture the user's operation trajectory data in the new version interface, which can specifically include the click frequency distribution correction amount of the core function control, the path switching efficiency improvement degree, and the recommended content adoption rate change amplitude.

[0149] The click frequency distribution correction amount refers to the change in the click frequency distribution of the core function control relative to that before the update. The path switching efficiency improvement degree refers to the degree of improvement in the efficiency of the user switching between function modules in the new version interface relative to the old version. The recommended content adoption rate change amplitude refers to the change in the user's adoption rate of recommended content after the incremental update relative to that before the update. To calculate the click frequency distribution correction amount of the core function control, the click frequency distribution of the core function control before and after the update can be first counted, and then the difference between the two is calculated. For example, the click frequency of the core function control C1 before the update is 100 times, and after the update it is 120 times, so the click frequency distribution correction amount is 20 times.

[0150] Then, the operation trajectory data is compared with the historical behavior pattern baseline to identify the behavior pattern offset caused by the parameter matrix update, and a spatial and temporal distribution heat map of the offset is generated.

[0151] The historical behavior pattern baseline refers to a relatively stable pattern formed by the user's operation behavior before the parameter matrix update, which can be used as a reference standard to measure the changes in the user's behavior pattern after the parameter matrix update. The behavior pattern offset refers to the degree of deviation of the user's operation behavior relative to the historical behavior pattern baseline after the parameter matrix update. The spatial and temporal distribution heat map of the offset is a visual representation that shows the distribution of the behavior pattern offset at different spatial locations and time points through the depth of color.

[0152] When comparing the operation trajectory data with the historical behavior pattern baseline, a similarity calculation method can be used. For example, the dynamic time warping (DTW) algorithm is used to calculate the similarity between the operation trajectory data and the historical behavior pattern baseline. In order to identify the behavior pattern offset caused by the parameter matrix update, a causal analysis method can be used. For example, Granger causality test is used to determine whether the parameter matrix update is the cause of the behavior pattern offset.

[0153] After that, the abnormal area exceeding the preset confidence threshold in the offset spatiotemporal distribution heat map is extracted, the rendering parameter adjustment record of the corresponding interface element is associated with the function module priority correction history to generate a parameter adjustment influence factor set. The abnormal area refers to the area in the offset spatiotemporal distribution heat map, in which the behavior pattern offset exceeds the preset confidence threshold. The rendering parameter adjustment record refers to the record of adjusting the rendering parameter of the interface element in the interface reconstruction process, including the adjustment of the color, size, transparency and other parameters. When the abnormal area exceeding the preset confidence threshold in the offset spatiotemporal distribution heat map is extracted, the threshold screening method can be used. The rendering parameter adjustment record of the corresponding interface element is associated with the function module priority correction history, which can be achieved by establishing an association table. In the association table, the rendering parameter adjustment record and the function module priority correction history of the interface element corresponding to each abnormal area are recorded. The parameter adjustment influence factor set can be generated by using the weighted summation method.

[0154] Then, based on the influence factor set, the optimization direction correction coefficient of the next round of parameter matrix update instruction is calculated, and the correction coefficient is used to suppress the unexpected behavior pattern offset and enhance the target optimization effect. The influence factor set is generated by the foregoing steps and reflects the influence degree of parameter adjustment on the user behavior pattern. The optimization direction correction coefficient is a coefficient for adjusting the optimization direction of the next round of parameter matrix update instruction, which can be calculated according to the influence factor set. The unexpected behavior pattern offset refers to the change of the user behavior pattern that does not conform to the expectation after the parameter matrix update. The target optimization effect refers to the optimization target expected to be achieved when the parameter matrix is updated, such as improving the user operation efficiency, improving the recommendation content adoption rate, etc.

[0155] In order to calculate the optimization direction correction coefficient of the next round of parameter matrix update instruction based on the influence factor set, the regression analysis method can be used. The influence factor set is used as the independent variable, and the unexpected behavior pattern offset is used as the dependent variable to establish a regression model. The regression coefficient of the regression model is used as the optimization direction correction coefficient. When the correction coefficient is positive, it means that the parameter matrix update instruction needs to be adjusted in a certain direction to reduce the unexpected behavior pattern offset; when the correction coefficient is negative, it means that the parameter matrix update instruction needs to be adjusted in the opposite direction.

[0156] After that, a dynamic learning rate adjustment mechanism is introduced in the closed-loop optimization link to dynamically adjust the parameter matrix update amplitude and frequency according to the historical update instruction execution effect feedback.

[0157] The closed-loop optimization link refers to a cycle from collecting user operation data, generating interface optimization feedback signals, updating parameter matrices to collecting user operation data again. It can continuously optimize the interface. In the closed-loop optimization link, a dynamic learning rate adjustment mechanism is introduced. Adaptive learning rate algorithms such as Adagrad, Adadelta, Adam, etc. can be used.

[0158] Finally, when it is detected that the user behavior pattern deviation is lower than the preset stability threshold after S consecutive updates, the conservative update strategy is switched to, and only the modules with path matching deviation exceeding the limit in the interface rendering performance bottleneck area are executed for local parameter correction, S≥3; wherein the running state of the closed-loop optimization link is synchronized in real time to the incremental training process of the time series analysis model, to ensure the consistency of the collaborative evolution of the user behavior pattern prediction and the interface optimization strategy.

[0159] The preset stability threshold is a preset value for determining whether the user behavior pattern has reached a stable state. The conservative update strategy refers to a relatively conservative update method used after the user behavior pattern has reached a stable state, which only performs local parameter correction on necessary areas. The interface rendering performance bottleneck area refers to an area where performance bottlenecks occur during interface rendering, such as areas with long loading times, lag, etc. When detecting whether the user behavior pattern deviation is lower than the preset stability threshold after S consecutive updates, the user behavior pattern deviation after each update can be recorded and compared. When the user behavior pattern deviation after S consecutive updates is lower than the preset stability threshold, it means that the user behavior pattern has reached a stable state. After switching to the conservative update strategy, only the modules with path matching deviation exceeding the limit in the interface rendering performance bottleneck area are executed for local parameter correction.

[0160] In step S500, the time series analysis model is iteratively updated through the incremental feedback mechanism within the preset verification period, which can include the following steps S560-S5100:

[0161] Step S560: After the interactive interface is deployed, user behavior response data and business conversion rate indicators are continuously collected.

[0162] User behavior response data refers to a series of behavior data generated by users during interaction with the interactive interface, including click, swipe, input, and other operation data, which reflects the user's operation habits and behavior patterns. The business conversion rate indicator refers to the ratio of the number of users who achieve business goals to the total number of users during business operations, which reflects the conversion effect of the business. User behavior response data and business conversion rate indicators can be collected using data collection tools such as log recording tools, and embedded statistical tools.

[0163] Step S570: Perform difference comparison between the behavior response data and the historical behavior trajectory feature set to identify the behavior pattern deviation after interface optimization.

[0164] The historical behavior trajectory feature set refers to a series of features formed by the user's operation behavior before interface optimization, which can be used as a reference standard to measure the changes in the user's behavior pattern after interface optimization. The behavior pattern deviation refers to the degree of deviation of the user's operation behavior relative to the historical behavior trajectory feature set after interface optimization.

[0165] When performing difference comparison between the behavior response data and the historical behavior trajectory feature set, a feature matching method can be used. The behavior response data is extracted as a feature vector, which is then matched with the feature vector in the historical behavior trajectory feature set. Cosine similarity, Euclidean distance, etc. can be used to measure the similarity between the feature vectors. When the similarity is low, it means that the behavior pattern has deviated greatly.

[0166] Step S580: Based on the behavior pattern deviation, construct a model update gradient matrix, and maintain the learned key features not to be covered by an elastic weight solidification strategy.

[0167] The model update gradient matrix refers to the matrix used to update the parameters of the time series analysis model, which can be constructed according to the behavior pattern deviation. The elastic weight solidification strategy is a strategy for protecting the learned key features from being covered by new data, which introduces an elastic constraint term in the loss function to punish the deviation of key parameters.

[0168] Based on the behavior pattern deviation, the model update gradient matrix can be constructed using the gradient descent algorithm. The behavior pattern deviation is used as an error signal, and the gradient of the model parameters is calculated through the backpropagation algorithm, and then the model parameters are updated according to the gradient.

[0169] The elastic weight solidification strategy can be used to maintain the learned key features not to be covered, which can include:

[0170] First, calculate the importance score matrix of each layer parameter of the time series analysis model in the historical training process, and the score matrix is generated based on the cumulative value of the parameter gradient change amplitude. The importance score matrix refers to the matrix used to measure the importance of each layer parameter of the time series analysis model in the historical training process, which can be generated by the cumulative value of the parameter gradient change amplitude. The parameter gradient change amplitude refers to the size of the parameter gradient value in each training process. In order to calculate the importance score matrix of each layer parameter of the time series analysis model in the historical training process, the gradient value of the parameter in each training process can be recorded and accumulated.

[0171] After that, an elastic constraint term is introduced for each parameter in the incremental training stage, and the strength of the constraint term is positively correlated with the importance score of the corresponding parameter. The elastic constraint term refers to a constraint term introduced in the loss function, which is used to punish the deviation of the key parameter. The strength of the constraint term refers to the degree of punishment of the parameter deviation by the elastic constraint term, which is positively correlated with the importance score of the corresponding parameter, that is, the higher the importance score of the parameter, the greater the strength of the constraint term.

[0172] In the incremental training stage, an elastic constraint term is introduced for each parameter, which can add an additional term to the loss function.

[0173] Then, an elastic constraint regularization term is added to the loss function to punish the deviation of the key parameter caused by the new training data. The elastic constraint regularization term refers to a regularization term added to the loss function, which is used to punish the deviation of the key parameter caused by the new training data. It can protect the learned key features from being covered by limiting the range of changes of the key parameters. Adding an elastic constraint regularization term to the loss function can add a regularization term to the original loss function.

[0174] Then, the learning rate allocation strategy of the optimizer is dynamically adjusted, and the decay learning rate is used for the first parameter and the adaptive learning rate is used for the second parameter, where the importance of the first parameter is greater than that of the second parameter. The optimizer refers to an algorithm used to update model parameters, such as stochastic gradient descent (SGD), Adam, etc. The learning rate allocation strategy refers to the strategy of allocating different learning rates to different parameters. The decay learning rate refers to the strategy of gradually reducing the learning rate as the training progresses. The adaptive learning rate refers to the strategy of automatically adjusting the learning rate according to the gradient changes of the parameters. The first parameter refers to the parameter with higher importance, and the second parameter refers to the parameter with lower importance. Dynamically adjusting the learning rate allocation strategy of the optimizer can be done according to the importance score of the parameters. For the first parameter with higher importance, the decay learning rate is used to avoid over-update and cover the key features; for the second parameter with lower importance, the adaptive learning rate is used to converge to the optimal solution faster.

[0175] Finally, the importance score matrix is updated after each training period, and the strength coefficient of the elastic constraint term is recalculated. After each training period, since the model parameters have been updated, the importance score of the parameters will also change. Therefore, the importance score matrix needs to be updated, and the strength coefficient of the elastic constraint term needs to be recalculated.

[0176] Step S590: Adjust the parameters of the time series analysis model step by step using the mini-batch incremental training method, and calculate the model prediction stability index through the online validation set after each update.

[0177] The mini-batch incremental training mode refers to a training mode in which training data is divided into a plurality of mini-batches, and each time, only data in one mini-batch is used to update the model parameters. The online validation set refers to a data set used to verify the model performance in real time during the training process. The model prediction stability index refers to an index used to measure the stability of the model prediction results, such as mean square error (MSE), mean absolute error (MAE), etc. Gradually adjusting the parameters of the time series analysis model by using the mini-batch incremental training mode can reduce the amount of calculation and improve the training efficiency. After each update, the model prediction stability index is calculated through the online validation set, so that the performance change of the model can be understood in a timely manner.

[0178] Step S5100: When the stability index fluctuation range of the continuous K validation periods is lower than the preset threshold, terminate the incremental update process and lock the current model version, K>3.

[0179] The preset threshold is a pre-set value used to determine whether the stability index fluctuation range of the model is within an acceptable range. The continuous K validation periods refer to a process of performing K validations continuously. The stability index fluctuation range refers to the difference between the maximum value and the minimum value of the model prediction stability index in the continuous K validation periods.

[0180] When the stability index fluctuation range of the continuous K validation periods is lower than the preset threshold, it indicates that the performance of the model has tended to be stable, and at this time, the incremental update process can be terminated and the current model version can be locked. In this way, the stability and reliability of the model can be guaranteed, and the performance of the model can be prevented from being degraded due to excessive updates.

[0181] Please refer to Figure 2 , Figure 2 A structural schematic diagram of a computer system provided by the embodiment of the present application is provided, which at least includes a processor 101, a communication interface 102 and a memory 103. The processor 101, the communication interface 102 and the memory 103 can be connected through a bus or other means. The processor 101 can analyze various instructions in the computer system and process various data of the computer system. The communication interface 102 can optionally include a standard wired interface, a wireless interface (such as WI-FI, a mobile communication interface, etc.), and can be used for transmitting and receiving data under the control of the processor 101; the communication interface 102 can also be used for the transmission and interaction of data within the computer system. The memory 103 is a memory device in the computer system, used for storing programs and data. The memory 103 provides a storage space, which stores an operating system of the computer system. The processor 101 executes the computer program in the memory 103 to implement the user behavior data mining method applied to digital enterprise management provided by the embodiment of the present application.

Claims

1. A user behavior data mining method applied to digital enterprise management, characterized in that, The method includes: Collect multi-dimensional behavioral data of target users on the business operation interface. The multi-dimensional behavioral data includes interface click trajectory, operation duration distribution, function module switching path and related business data request records. Multimodal data parsing is performed on the multidimensional behavioral data to generate a behavioral trajectory feature set with temporal correlation. Specifically, this includes: extracting the spatial coordinate sequence from the interface click trajectory and calculating the interface area attention heatmap based on the coordinate distribution density; parsing the dwell interval features in the operation duration distribution to identify operation clusters composed of continuous operation events exceeding a frequency threshold and abnormal stagnation nodes, with high-frequency operation clusters composed of continuous events with operation frequency exceeding a preset threshold within the sliding window; constructing the transition probability matrix of the functional module switching path, calculating the state transition weights between each functional module, and generating a path prediction sequence based on a hidden Markov model; mapping the associated business data request records into request type encoding vectors, and constructing a request load evaluation index through request response latency and result data volume; performing multimodal feature fusion on the attention heatmap, dwell interval features, transition probability matrix, and request load evaluation index, and using a time axis alignment strategy to generate a temporal embedding vector containing global context information as the behavioral trajectory feature set; wherein different modal data are synchronized with timestamps and semantically correlated through a cross-modal alignment strategy. An adaptive time-series analysis model is trained based on the behavioral trajectory feature set. The time-series analysis model captures the long-term and short-term dependencies in user behavior patterns through a dynamic window partitioning strategy and generates potential churn risk prediction indicators. An interaction flow parameter matrix is ​​constructed based on the potential churn risk prediction indicators. This matrix includes the weight distribution of interface elements, the priority sequence of functional modules, and the recommendation strategy adjustment coefficient. Specifically, it includes: extracting user behavior deviation features from the potential churn risk prediction indicators, whereby the deviation features include the matching difference results between the interface operation frequency distribution and historical behavior trajectories, and the abnormal fluctuation coefficient of functional module dwell time; calculating the dynamic weight allocation value of each interface element based on the interface operation frequency distribution and matching difference results, generating an interface element weight distribution, wherein the weight values ​​of regions with matching difference results below a preset difference threshold show a positive correlation increase; and combining the abnormal fluctuation coefficient of functional module dwell time with the user's historical operation trajectory. The module switching path frequency is analyzed to construct a module transition probability graph. A functional module priority sequence is generated by sorting the in-degree weights of nodes in the graph. A set of business preference tags is extracted from the user profile, and semantic matching is performed between these tags and the functional module priority sequence to calculate the association strength between recommended content and target modules, generating recommendation strategy adjustment coefficients. The interface element weight distribution, functional module priority sequence, and recommendation strategy adjustment coefficients are integrated using a three-dimensional tensor, and orthogonal projection is used to eliminate cross-dimensional interference, generating the interaction flow parameter matrix. The interface element weight distribution drives the interface layout rendering, the functional module priority sequence optimizes navigation paths, and the recommendation strategy adjustment coefficients generate dynamic content push strategies. The optimized interaction flow parameter matrix drives the reconstruction of the business operation interface, generating an interaction interface adapted to the current user behavior pattern. Within a preset verification period, the time-series analysis model is iteratively updated through an incremental feedback mechanism. Specifically, this includes: parsing the weight distribution of interface elements in the interaction flow parameter matrix, extracting the spatial coordinate set and visual rendering priority parameters of regions exceeding weight thresholds, and generating dynamic focusing instructions for core functional controls; calculating the access path weights of each module node in the navigation menu based on the sorting results of the functional module priority sequence, and generating the shortest path topology based on the weight gradient distribution; inputting the recommendation strategy adjustment coefficients into a pre-trained recommendation rule generator, matching the user's real-time operation scenario and historical preference features, and outputting a personalized recommendation content sequence associated with the current functional module; integrating the dynamic focusing instructions, the shortest path topology, and the personalized recommendation content sequence to generate an interface layout rendering configuration file; and after the interface layout rendering configuration file is loaded on the target client, collecting in real-time user operation response latency, path switching efficiency, and recommended content adoption rate indicators for core functional controls, generating interface optimization feedback signals, and sending them back to the update queue of the interaction flow parameter matrix.

2. The method according to claim 1, characterized in that, The training of the adaptive time-series analysis model based on the behavioral trajectory feature set includes: The temporal embedding vector is input into a bidirectional long short-term memory network, and the forward and reverse temporal dependencies are captured through a gating mechanism to generate an initial hidden state sequence. A multi-head self-attention mechanism layer is connected to the back end of the bidirectional long short-term memory network to calculate the correlation weights between features at different time steps, and the initial hidden state sequence is aggregated with attention weights to obtain the aggregated feature sequence. A dynamic convolutional kernel group is used to perform multi-scale feature extraction on the aggregated feature sequence to obtain multi-scale features. The dynamic convolutional kernel group adaptively adjusts the convolutional kernel size and dilation coefficient according to the current input feature dimension. The multi-scale features are residually connected to the original temporal embedding vectors and mapped through a fully connected layer to a latent vector space containing abstract representations of user behavior patterns. A joint training framework for contrastive learning and prediction tasks is constructed in the latent vector space. The contrastive learning task enhances the model's ability to identify differences in behavioral patterns through positive and negative sample pairs, while the prediction task optimizes the accuracy of potential churn risk prediction indicators through the cross-entropy loss function.

3. The method according to claim 2, characterized in that, The method employs dynamic convolutional kernel groups to perform multi-scale feature extraction on the aggregated feature sequence, resulting in multi-scale features, including: The kernel size adjustment factor is calculated based on the time step and channel dimension of the input feature sequence. The adjustment factor is negatively correlated with the local variance of the feature sequence. Based on the adjustment factor, a set of basic convolutional kernel sizes is generated, and a learnable dilation coefficient matrix is ​​assigned within each convolutional kernel. The dilation coefficient matrix dynamically adjusts the receptive field range according to the feature differences between adjacent time steps. Multiple dynamic convolution kernels are applied in parallel to the same input feature sequence to generate feature maps with different scale characteristics; Channel attention weighting is applied to the feature map, the importance score of each channel feature is calculated, and the channel weights are redistributed. The weighted feature map is concatenated along the channel dimension and then dimensionality is reduced by a separable convolutional layer to generate the multi-scale features.

4. The method according to claim 3, characterized in that, The step of calculating the access path weight of each module node in the navigation menu based on the sorting result of the priority sequence of the functional modules, and generating the shortest path topology based on the weight gradient distribution, includes: The sorting results of the priority sequence of the functional modules are analyzed, and the historical statistical values ​​of the in-degree weight of the module nodes and the transfer frequency between adjacent modules are extracted to generate the initial access path weight of each module node. The initial access path weight is corrected based on the weight difference gradient between adjacent modules in the weight gradient distribution. The module node path weights in regions exceeding the preset gradient change are suppressed by a decay factor to prevent abnormal weight growth. Based on the corrected access path weights, a directed weighted graph structure is constructed, with module nodes as vertices and path weights as edge weights, to generate the initial shortest path topology. Redundancy path verification is performed on the initial shortest path topology, and jump nodes that violate the monotonicity of weight gradient change in the initial shortest path topology are identified. The reconstructed path topology is obtained by inserting intermediate module nodes or adjusting edge weight allocation. The reconstructed path topology is mapped to a guide sequence in the navigation menu, and the visual parameters of the guide sequence are matched with the weight gradient distribution trend of each node on the reconstructed path topology. When a user performs a module switching operation, the path deviation index is collected in real time, the access path weight is updated based on the current weight gradient distribution, and the update result is synchronously fed back to the sorting logic of the priority sequence of the functional modules.

5. The method according to claim 1, characterized in that, The step of iteratively updating the time series analysis model through an incremental feedback mechanism within a preset verification period includes: After the interactive interface is deployed, user behavior response data and business conversion rate indicators are continuously collected. The behavioral response data is compared with the historical behavioral trajectory feature set to identify the behavioral pattern offset after interface optimization. The model update gradient matrix is ​​constructed based on the behavioral pattern offset, and the learned key features are kept from being covered by the elastic weight solidification strategy. The parameters of the time series analysis model are gradually adjusted using a small-batch incremental training method, and the model's predicted stability index is calculated using an online validation set after each update. When the stability index fluctuation range is lower than the preset threshold for K consecutive verification cycles, the incremental update process is terminated and the current model version is locked, where K≥3.

6. The method according to claim 5, characterized in that, The strategy of maintaining learned key features from being covered through elastic weight solidification includes: Calculate the importance score matrix of each layer parameter of the time series analysis model during the historical training process. The score matrix is ​​generated based on the cumulative value of the parameter gradient change magnitude. During the incremental training phase, an elastic constraint term is introduced for each parameter, and the strength of the constraint term is positively correlated with the importance score of the corresponding parameter. Add a resilience constraint regularization term to the loss function to penalize the shift in key parameters caused by new training data; The learning rate allocation strategy of the optimizer is dynamically adjusted. A decaying learning rate is used for the first parameter and an adaptive learning rate is used for the second parameter. The first parameter is more important than the second parameter. The importance score matrix is ​​updated after each training cycle, and the strength coefficients of the elastic constraint terms are recalculated.

7. The method according to claim 1, characterized in that, The generation of interface optimization feedback signals and their transmission back to the update queue of the interaction process parameter matrix includes: Analyze UI element loading delay exception events in user operation response latency, identify control coordinate sets and corresponding functional module identifiers that exceed the preset delay trigger threshold, and generate UI rendering performance bottleneck area markers. Extract the path jump time and deviation count statistics from the path switching efficiency index, and combine them with the node distribution characteristics of the shortest path topology to calculate the matching deviation between the path guidance strategy and the user's actual operating habits. The recommended content adoption rate metric is decomposed into the ratio of the number of times the recommended content associated with different functional modules is exposed to the number of times users actively interact with it, and a module-level recommendation utility evaluation matrix is ​​generated. Multidimensional correlation analysis is performed on the interface rendering performance bottleneck area marking, matching deviation degree and recommendation utility evaluation matrix to determine the collaborative optimization direction of interface element weight distribution, functional module priority sequence and recommendation strategy adjustment coefficient in the interaction process parameter matrix; The parameter matrix update instruction is generated based on the collaborative optimization direction. The instruction includes the interface element weight decay coefficient, the module priority ranking correction magnitude, and the recommendation strategy trigger threshold adjustment amount. The parameter matrix update instructions are stored in the update queue in order of priority. When the dynamic change of the weight distribution of interface elements is detected to exceed the preset threshold, the real-time incremental update of the parameter matrix is ​​triggered. After performing the incremental update, user operation data is collected again to generate a new round of interface optimization feedback signals, forming a closed-loop optimization chain; The logic for generating the interface optimization feedback signal is synchronously linked to the version iteration record of the interface layout rendering configuration file and the evolution trajectory of user behavior patterns, ensuring dynamic adaptation between the parameter matrix update strategy and system performance indicators.

8. A computer system, characterized in that, include: A memory, wherein a computer program is stored; A processor for loading the computer program to implement the user behavior data mining method for digital enterprise management as described in any one of claims 1-7.

Images