A payment label intelligent processing method and system based on multi-dimensional user portrait

Abstract

The application provides a payment label intelligent processing method and system based on multi-dimensional user portrait, relates to the technical field of big data analysis, constructs a multi-dimensional user portrait matrix, and extracts time and space self-attention of user payment behavior, obtains joint time and space characteristics, and then screens a candidate payment label set; semantic matching and rearrangement are performed on the candidate payment label set through a semantic correlation mapping and a contrast learning mechanism to obtain a semantic label sequence, the feature contribution degree is determined according to the semantic label sequence and the joint time and space characteristics; the payment label of the user is predicted and error feedback is fed back based on the feature contribution degree and the historical payment record, a semantic verification result is obtained, and then the payment label of the user is risk-identified and behavior-recommended to obtain a payment decision label used for adaptive recommendation, and the application can realize payment label self-learning and semantic dynamic optimization based on multi-source behavior characteristics to improve the payment label identification accuracy.

Description

Technical Field

[0001] This application relates to the field of big data analytics technology, and more specifically, to a method and system for intelligent processing of payment tags based on multi-dimensional user profiles. Background Technology

[0002] With the rapid development of internet technology and mobile payment, users have generated massive amounts of payment behavior data across different terminals and scenarios. Big data analytics, as an important foundation for supporting intelligent decision-making and precise services, can collect, clean, model, and extract features from multi-source heterogeneous data, thereby uncovering user behavior patterns and potential intentions.

[0003] In the payment sector, big data analytics typically combines machine learning and artificial intelligence algorithms to identify payment risks, predict payment trends, and provide personalized recommendations, thereby improving the security and user experience of payment systems. Among these, intelligent payment tag processing based on multi-dimensional user profiles can generate corresponding payment tags by analyzing multi-dimensional features such as users' historical payment data, terminal information, geographical environment, and behavioral preferences to support risk control and behavioral recommendations. However, existing intelligent payment tag processing technologies often construct user profiles based on a single dimension, failing to fully integrate spatiotemporal features and multi-source behavioral data. This results in incomplete user profile representation, reliance on static classification algorithms, a lack of semantic relevance modeling and context awareness, and an inability to accurately depict the semantic features of user payment behavior. Furthermore, the lack of data interaction for risk identification and behavioral recommendation easily leads to recognition delays and low accuracy. Therefore, how to achieve self-learning and dynamic semantic optimization of payment tags based on multi-source behavioral features to improve payment tag recognition accuracy is a challenge facing the industry. Summary of the Invention

[0004] This application provides a payment tag intelligent processing method and system based on multi-dimensional user profiles, which can realize payment tag self-learning and semantic dynamic optimization based on multi-source behavioral features, so as to improve the accuracy of payment tag recognition.

[0005] Firstly, this application provides a payment tag intelligent processing method based on multi-dimensional user profiles, the intelligent processing method comprising the following steps:

[0006] Construct a multi-dimensional user profile matrix based on multi-source behavioral features;

[0007] Based on the multi-dimensional user profile matrix, spatiotemporal self-attention joint extraction of user payment behavior is performed to obtain joint spatiotemporal features. Then, the joint spatiotemporal features are filtered through dual channels to obtain a candidate payment tag set.

[0008] The candidate payment tag set is semantically matched and rearranged using semantic relevance mapping and contrastive learning mechanisms to obtain a semantic tag sequence. Then, the feature contribution is determined based on the semantic tag sequence and the joint spatiotemporal features.

[0009] Obtain the user's historical payment records, and based on the feature contribution and the historical payment records, perform prediction error feedback on the user's payment tags to obtain semantic verification results;

[0010] Based on the semantic verification results, risk identification and behavior recommendation are performed on the user's payment tags to obtain payment decision tags for adaptive recommendation.

[0011] In this embodiment, constructing a multi-dimensional user profile matrix based on multi-source behavioral features specifically includes:

[0012] The multi-source behavioral features are time-stamp aligned and normalized to obtain a behavioral feature set;

[0013] The behavioral feature set is subjected to feature hierarchical encoding to obtain the relevance of each feature layer;

[0014] A multi-dimensional user profile matrix is ​​constructed based on all the relevances and the behavioral feature set.

[0015] In this embodiment, the spatiotemporal self-attention joint extraction of user payment behavior based on the multi-dimensional user profile matrix yields joint spatiotemporal features, specifically including:

[0016] Determine the dependency weights of user payment behavior;

[0017] Based on the dependency weights, temporal attention dependency modeling and spatial attention dependency modeling are performed on the multi-dimensional user profile matrix to obtain the temporal attention features and spatial attention features of user payment behavior.

[0018] The temporal attention features and the spatial attention features are fused to obtain joint spatiotemporal features.

[0019] In this embodiment, the dual-channel filtering of the joint spatiotemporal features to obtain the candidate payment tag set specifically includes:

[0020] A semantic feature mapping channel is constructed based on a semantic embedding network, and a rule constraint channel is constructed based on a preset rule set;

[0021] The joint spatiotemporal features are input into the semantic feature mapping channel, and the joint spatiotemporal features are mapped to the label semantic space through the semantic embedding network, thereby obtaining a set of candidate labels.

[0022] The candidate tag set is logically constrained and filtered using the rule matching matrix in the rule constraint channel to obtain the candidate payment tag set.

[0023] In this embodiment, the semantic matching and rearrangement of the candidate payment tag set are performed through semantic relevance mapping and contrastive learning mechanisms, respectively, to obtain a semantic tag sequence, specifically including:

[0024] Semantic embedding is performed on the candidate payment tag set to obtain a semantic embedding matrix;

[0025] The semantic relevance between the semantic embedding matrix and the label semantic space is determined, and then the relevance ranking result is determined based on the semantic relevance.

[0026] The relevance ranking results are compared and optimized to obtain a semantic label sequence.

[0027] In this embodiment, determining the feature contribution based on the semantic label sequence and the joint spatiotemporal features specifically includes:

[0028] Based on the correlation, the feature mapping weights of the semantic label sequence and the joint spatiotemporal features are calculated, and then the contribution matrix is ​​obtained;

[0029] The weight distribution vector is determined by the contribution matrix, and then the weighted response coefficient of each semantic tag and the joint spatiotemporal feature in the semantic tag sequence is determined based on the weight distribution vector.

[0030] All weighted response coefficients are aggregated according to the time and spatial dimensions to obtain the feature contribution.

[0031] In this embodiment, the user's historical payment records are obtained through a payment data management database.

[0032] In this embodiment, the process of predicting the user's payment tag based on the feature contribution and the historical payment records to obtain the semantic verification result specifically includes:

[0033] Construct a tag entity set based on the user's payment tags;

[0034] Based on the feature contribution and the historical payment records, the tag entity set is predicted and paired to obtain the prediction deviation value;

[0035] The prediction pairing process is updated by feeding back the prediction deviation value to obtain the semantic verification vector.

[0036] The semantic verification result is determined based on the semantic verification vector and the preset confidence threshold.

[0037] In this embodiment, the multi-source behavioral features refer to a multi-dimensional data set composed of user payment behavior attributes under different time, scenario and device conditions. The multi-source behavioral features include: payment features, terminal features and environmental features.

[0038] Secondly, this application provides a payment tag intelligent processing system based on multi-dimensional user profiles, used to execute a payment tag intelligent processing method based on multi-dimensional user profiles, the intelligent processing system comprising:

[0039] The user profile building module is used to construct a multi-dimensional user profile matrix based on multi-source behavioral features.

[0040] The spatiotemporal feature extraction module is used to perform spatiotemporal self-attention joint extraction of user payment behavior based on the multi-dimensional user profile matrix to obtain joint spatiotemporal features, and then perform dual-channel filtering on the joint spatiotemporal features to obtain a candidate payment tag set.

[0041] The semantic reordering module is used to perform semantic matching and reordering on the candidate payment tag set through semantic relevance mapping and contrastive learning mechanisms respectively to obtain a semantic tag sequence, and then determine the feature contribution based on the semantic tag sequence and the joint spatiotemporal features.

[0042] The feedback verification module is used to obtain the user's historical payment records, and based on the feature contribution and the historical payment records, to provide prediction error feedback for the user's payment tags, thereby obtaining semantic verification results.

[0043] The risk identification and recommendation module is used to identify risks and recommend behaviors for users' payment tags based on the semantic verification results, and obtain payment decision tags for adaptive recommendation.

[0044] The technical solutions provided by the embodiments disclosed in this application have the following beneficial effects:

[0045] A multi-dimensional user profile matrix is ​​constructed based on multi-source behavioral features. Based on this matrix, spatiotemporal self-attention joint extraction of user payment behavior is performed to obtain joint spatiotemporal features. These features are then subjected to dual-channel filtering to obtain a candidate payment tag set. The candidate payment tag set is semantically matched and rearranged using semantic relevance mapping and contrastive learning mechanisms to obtain a semantic tag sequence. The feature contribution is then determined based on the semantic tag sequence and the joint spatiotemporal features. The user's historical payment records are obtained, and prediction error feedback is performed on the user's payment tags based on the feature contribution and the historical payment records to obtain a semantic verification result. Based on the semantic verification result, risk identification and behavior recommendation are performed on the user's payment tags to obtain payment decision tags for adaptive recommendation.

[0046] Therefore, this application demonstrates that payment tag self-learning and semantic dynamic optimization based on multi-source behavioral features can be achieved. First, by constructing a multi-dimensional user profile matrix, the payment behavior features of users under different time, scenario, and device conditions can be uniformly modeled, enabling the user profile to possess high-dimensional and computable behavioral semantic expression. This facilitates the formation of a complete spatiotemporal description of individual behavioral features, providing a unified data input foundation for tag recognition. Second, spatiotemporal self-attention joint extraction of user payment behavior based on the multi-dimensional user profile matrix can capture the dynamic evolution of user payment behavior from both time series and spatial distribution dimensions. By performing dual-channel filtering on the joint spatiotemporal features, combining semantic feature mapping with rule-constrained filtering, precise compression of the tag candidate space can be achieved, reducing interference from invalid tags and thus providing high-confidence input for the semantic matching stage. Next, the candidate payment tag set is semantically matched and rearranged through semantic relevance mapping and contrastive learning mechanisms. This enables dynamic optimization of the tag sequence based on the relevance distribution in the semantic vector space. The contrastive learning mechanism strengthens semantic recognition capabilities through feature alignment of positive and negative samples. Furthermore, feature contribution is calculated based on the semantic tag sequence and joint spatiotemporal features, which helps improve the interpretability and controllability of semantic matching. Then, a self-supervised closed loop for semantic verification is established through historical payment records and feature contribution, which can correct errors in the prediction results. This facilitates continuous optimization and stable convergence in actual payment scenarios, thereby reducing recognition bias. Finally, the generation of adaptive payment decision tags can dynamically adjust the recommendation strategy based on user behavior trends and risk patterns, thereby realizing a closed-loop control mechanism from semantic recognition to intelligent recommendation, which improves the accuracy of payment tag recognition.

[0047] In summary, the technical solution adopted in this application can realize self-learning and semantic dynamic optimization of payment tags based on multi-source behavioral features, so as to improve the accuracy of payment tag recognition. Attached Figure Description

[0048] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only for this embodiment of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0049] Figure 1 This is a flowchart of a payment tag intelligent processing method based on multi-dimensional user profiles provided in this application;

[0050] Figure 2 This is an exemplary flowchart for determining joint spatiotemporal characteristics provided in this application;

[0051] Figure 3 This is an exemplary flowchart of determining a semantic tag sequence according to the present application;

[0052] Figure 4 This is a module structure diagram of a payment tag intelligent processing system based on multi-dimensional user profiles, provided in this application. Detailed Implementation

[0053] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0054] This application provides a method and system for intelligent processing of payment tags based on multi-dimensional user profiles. The core of this method is to construct a multi-dimensional user profile matrix based on multi-source behavioral features; perform spatiotemporal self-attention joint extraction of user payment behavior based on the multi-dimensional user profile matrix to obtain joint spatiotemporal features; then perform dual-channel filtering on the joint spatiotemporal features to obtain a candidate payment tag set; perform semantic matching and rearrangement of the candidate payment tag set through semantic relevance mapping and contrastive learning mechanisms to obtain a semantic tag sequence; then determine the feature contribution based on the semantic tag sequence and the joint spatiotemporal features; obtain the user's historical payment records; perform prediction error feedback on the user's payment tags based on the feature contribution and the historical payment records to obtain a semantic verification result; and perform risk identification and behavior recommendation on the user's payment tags based on the semantic verification result to obtain payment decision tags for adaptive recommendation.

[0055] Example 1: To better understand the above technical solution, the following will provide a detailed description of the technical solution in conjunction with the accompanying drawings and specific implementation methods. (Refer to...) Figure 1 As shown in the figure, this is a flowchart of a payment tag intelligent processing method based on multi-dimensional user profiles according to this embodiment of the application. The intelligent processing method includes the following steps:

[0056] In step S1, a multi-dimensional user profile matrix is ​​constructed based on multi-source behavioral features.

[0057] It should be noted that, in this embodiment, the multi-source behavioral features refer to a multi-dimensional data set composed of user payment behavior attributes under different time, scenario, and device conditions. These multi-source behavioral features include: payment features, terminal features, and environmental features. Specifically, the payment features describe the amount changes, payment frequency, and merchant type distribution during the payment process; the terminal features describe the type of payment terminal used by the user, operating system version, network connection method, and unique device identifier; and the environmental features describe the geographical location information and time period characteristics corresponding to the user's payment behavior. In a specific implementation, the user's multi-source behavioral features can be obtained through the payment log database of the payment platform. The payment log database includes information such as payment amount, payment time, merchant number, and payment category. Preferably, the payment log database can adopt a distributed data storage structure to facilitate high-concurrency reading and horizontal scaling of behavioral data from different sources.

[0058] In this embodiment, the construction of a multi-dimensional user profile matrix based on multi-source behavioral features can be carried out in the following manner:

[0059] The multi-source behavioral features are time-stamp aligned and normalized to obtain a behavioral feature set;

[0060] The behavioral feature set is subjected to feature hierarchical encoding to obtain the relevance of each feature layer;

[0061] A multi-dimensional user profile matrix is ​​constructed based on all the relevances and the behavioral feature set.

[0062] In practical implementation, forward padding can be used to fill in missing time slices of multi-source behavioral features, and outlier values ​​can be replaced with median values. Then, Z-Score standardization can be used to normalize the numerical features to obtain the behavioral feature set. Next, the time-continuous multi-source behavioral features are hierarchically encoded according to attribute type. Specifically, for numerical features describing changes in payment quantity and amount, statistical aggregation can be used to generate time window statistics. For categorical features representing merchant categories, payment methods, etc., one-hot encoding can be used to convert them into vector form. For text features representing device information and environmental semantics, they can be converted into structured data based on a dictionary mapping table, and then the correlation of each feature layer can be calculated using the Pearson correlation coefficient. Finally, the correlation is used as the feature layer weight, and each feature layer is arranged and concatenated into a matrix using the feature layer weights. In this matrix, the rows represent user samples, and the columns represent different dimensions of multi-source behavioral features (such as payment features, terminal features, and environmental features). This matrix is ​​then used as a multi-dimensional user profile matrix.

[0063] It should be noted that the multi-dimensional user profile matrix in this application refers to a matrix structure with user identifiers as indexes and multi-layer coded features as columns. This multi-dimensional user profile matrix can provide input data for empty feature extraction and semantic optimization processing. In this embodiment, feature hierarchical coding refers to dividing behavioral features into payment layer, terminal layer and environment layer according to feature semantics, and encoding each layer of data using common techniques such as numerical aggregation, one-hot coding or embedding mapping to eliminate feature heterogeneity.

[0064] In step S2, spatiotemporal self-attention joint extraction of user payment behavior is performed based on the multi-dimensional user profile matrix to obtain joint spatiotemporal features. Then, the joint spatiotemporal features are filtered in two channels to obtain a candidate payment tag set.

[0065] Preferably, in this embodiment, reference Figure 2 As shown, this diagram is an exemplary flowchart for determining joint spatiotemporal features according to the present application. In this embodiment, the joint spatiotemporal features are obtained by performing spatiotemporal self-attention joint extraction on user payment behavior based on the multi-dimensional user profile matrix, which can be achieved through the following steps:

[0066] First, in step S21, the dependency weights of user payment behavior are determined;

[0067] Then, in step S22, the multi-dimensional user profile matrix is ​​modeled with temporal attention dependency and spatial attention dependency respectively according to the dependency weight to obtain the temporal attention features and spatial attention features of user payment behavior.

[0068] Finally, in step S23, the temporal attention features and the spatial attention features are fused to obtain joint spatiotemporal features.

[0069] It should be noted that the user payment behavior refers to the behavioral characteristics describing payment records, including payment amount, number of payments, and average payment interval.

[0070] In specific implementation, firstly, a sliding time window (e.g., the past 7 days) is divided according to the time dimension. A linear transformation is then performed on the multi-dimensional user profile matrix using the sliding time window to obtain a time query vector, a time key vector, and a time value vector. Subsequently, Softmax is used to normalize the time query vector, time key vector, and time value vector to obtain dependency weights. Then, the dependency weights are used as the convolution kernels of a spatiotemporal convolutional network. The spatiotemporal convolutional network with pre-defined kernels is then used to perform spatiotemporal convolution on the multi-dimensional user profile matrix. The temporal attention features are obtained through the temporal channel output of this spatiotemporal convolutional network, and the spatial attention features are obtained through the spatial channel output of the same network. Preferably, in both types of attention calculations in this embodiment, a flexible... The linear transformation of the learning process adjusts the dimensions of the time query vector, time key vector, and time value vector. Residual connections can be added after attention calculation, which is beneficial for stable training. Finally, the temporal attention features and spatial attention features can be fused through a fully connected layer, and layer normalization is used for regularization to prevent overfitting. Preferably, a multi-head cross-attention mechanism can be used to calculate the interaction attention weights between the temporal attention features and spatial attention features to enhance the coupled expression of spatiotemporal features, thereby obtaining a high-dimensional vector after fusion and normalization. This high-dimensional vector can then be used as a joint spatiotemporal feature. The multi-head cross-attention mechanism can be implemented by constructing query, key, and value vectors and performing steps such as scaling dot product attention and linear transformation.

[0071] It should be noted that the joint spatiotemporal features in this application are high-dimensional vectors used to simultaneously represent temporal and spatial coupling relationships. These high-dimensional vectors can be directly used for semantic filtering and label prediction. In this embodiment, dependency weight refers to a numerical measure that quantifies the importance of user historical behavior in the current label recognition task. It can extract the spatiotemporal segments with the most influence on the current payment semantics during the spatiotemporal modeling process. Temporal attention features refer to vector representations describing the temporal evolution of user behavior obtained after weighted aggregation based on temporal attention. They can characterize the dynamic patterns of payment behavior over time. Spatial attention features refer to vector representations weighted based on geographical location, terminal type, and neighborhood similarity. They can characterize the clustering characteristics of payment behavior in spatial distribution and terminal differences.

[0072] In this embodiment, the candidate payment tag set is obtained by performing dual-channel filtering on the joint spatiotemporal features, which can be done in the following way:

[0073] A semantic feature mapping channel is constructed based on a semantic embedding network, and a rule constraint channel is constructed based on a preset rule set;

[0074] The joint spatiotemporal features are input into the semantic feature mapping channel, and the joint spatiotemporal features are mapped to the label semantic space through the semantic embedding network, thereby obtaining a set of candidate labels.

[0075] The candidate tag set is logically constrained and filtered using the rule matching matrix in the rule constraint channel to obtain the candidate payment tag set.

[0076] In specific implementation, firstly, the joint spatiotemporal features are forward encoded using a semantic embedding network based on a Transformer structure to extract the joint spatiotemporal semantic vector. This semantic vector is then mapped to the label semantic space through a semantic embedding layer, forming a semantic feature mapping channel. Based on empirically obtained payment rules, regular expressions are used to obtain constraints including payment amount ranges, payment frequency thresholds, device type restrictions, and regional whitelists. The constraints are then loaded into a rule matching matrix using a Transformer structure, and this rule matching matrix is ​​used to construct a rule constraint channel. Next, this semantic feature mapping channel uses cosine similarity to measure and filter the label semantics, thereby obtaining a candidate label set. Finally, the rule matching matrix in the rule constraint channel is used to perform rule matching operations on each candidate label in the set. That is, if a label does not meet the corresponding rule, it is removed from the set; if a label meets multiple rules, a weight score is added to it, thereby filtering to obtain a candidate payment set. The rule matching operation process can be implemented using the Pandas library.

[0077] It should be noted that the candidate payment tag set in this application is a payment tag obtained through semantic and rule-based filtering, used to describe the user's key payment information; the semantic embedding network in this embodiment is a deep encoding network used to uniformly map structured joint spatiotemporal features and textual tag descriptions to the same vector space, used to measure the semantic relevance between payment behavior features and tag descriptions; the tag semantic space is a multi-dimensional vector space generated by the semantic embedding network, used to represent the semantic differences between tags, the smaller the semantic distance, the more closely the behavior matches the tag; the rule matching matrix refers to a structured constraint table that expresses and stores business rules in matrix form, which can apply business logic constraints to the semantic candidate set; the candidate tag set refers to the tag set obtained through preliminary filtering based on semantic relevance.

[0078] In step S3, the candidate payment tag set is semantically matched and rearranged through semantic relevance mapping and contrastive learning mechanisms to obtain a semantic tag sequence, and then the feature contribution is determined based on the semantic tag sequence and the joint spatiotemporal features.

[0079] Preferably, in this embodiment, reference Figure 3As shown, this diagram is an exemplary flowchart for determining a semantic label sequence according to the present application. In this embodiment, the semantic matching and rearrangement of the candidate payment label set through semantic relevance mapping and contrastive learning mechanisms to obtain the semantic label sequence can be achieved through the following steps:

[0080] First, in step S31, semantic embedding is performed on the candidate payment tag set to obtain a semantic embedding matrix;

[0081] Then, in step S32, the semantic relevance of the semantic embedding matrix and the label semantic space is determined, and the relevance ranking result is determined through the semantic relevance.

[0082] Finally, in step S33, the relevance ranking results are compared and optimized to obtain a semantic label sequence.

[0083] In specific implementation, firstly, the text descriptions of candidate tags can be segmented, stop words removed, and lemmatization performed. Then, the processed text is input into a Transformer-based text encoder for encoding, yielding a text semantic vector for each tag. All tag semantic vectors are then combined row-wise into a matrix structure to obtain a semantic embedding matrix. Next, cosine similarity is used to calculate the semantic similarity between the text semantic vector and the semantic vectors in the tag semantic space. The mean of all semantic similarities is taken as the semantic relevance, and the semantic vectors with higher semantic similarity than semantic relevance are used to replace the semantic vectors in the tag semantic space, resulting in a relevance ranking. Finally, a contrastive loss function (i.e., cross-entropy contrastive loss) is used to calculate the loss value, and the parameters of the semantic embedding network are updated through backpropagation. The similarity score between the semantic vector and the text semantic vector is calculated and the ranking is updated until the loss value reaches Top-1 stability. Finally, the optimized tag sequence, arranged from high to low semantic matching strength, is output as a semantic tag sequence, which can be used as direct input for feature contribution calculation.

[0084] It should be noted that the semantic label sequence in this application refers to an ordered list of labels arranged from high to low semantic relevance after contrastive learning optimization, which is used to provide high-confidence label input for feature contribution calculation and semantic verification; the semantic embedding matrix refers to the matrix structure formed after mapping the candidate label text description into a continuous vector set through a language coding model, which is used to convert discrete label semantics into a computable high-dimensional vector representation; semantic relevance refers to a numerical measure of the closeness between the behavioral semantic vector and the label semantic vector in the semantic space; in this embodiment, contrastive learning optimization refers to an optimization unit with the core of minimizing the training and contrastive loss of positive and negative sample pairs, which helps to improve the discriminative ability of semantic matching by shortening the distance between positive samples and widening the distance between negative samples.

[0085] In this embodiment, the feature contribution degree is determined based on the semantic label sequence and the joint spatiotemporal features in the following manner:

[0086] Based on the correlation, the feature mapping weights of the semantic label sequence and the joint spatiotemporal features are calculated, and then the contribution matrix is ​​obtained;

[0087] The weight distribution vector is determined by the contribution matrix, and then the weighted response coefficient of each semantic tag and the joint spatiotemporal feature in the semantic tag sequence is determined based on the weight distribution vector.

[0088] All weighted response coefficients are aggregated according to the time and spatial dimensions to obtain the feature contribution.

[0089] In specific implementation, firstly, the similarity between the semantic vector of each label in the semantic label sequence and the spatiotemporal feature vector in the joint spatiotemporal features is measured. Preferably, the vector dot product can be used as the measurement method, and the obtained similarity value is used as the feature mapping weight. The matrix elements are filled according to the index position of [semantic label, joint spatiotemporal feature] to obtain the contribution matrix. Then, a normalization function (such as the Softmax function) is applied to normalize each row and each column of the contribution matrix to obtain the weight distribution vector. The weight distribution vector and the corresponding joint spatiotemporal feature vector are multiplied element-wise and summed in the dimension to obtain the weighted response coefficient of each semantic label relative to the spatiotemporal feature. Finally, the average value of the weighted response coefficient can be taken according to the set time window (such as the past 7 days) to obtain the time aggregation result. Then, the weighted response coefficient is spatially aggregated according to spatial grouping (such as by terminal type) to generate the feature contribution.

[0090] It should be noted that the feature contribution in this application refers to the weighted response strength of semantic tags in the joint spatiotemporal feature space, which is used to comprehensively describe the overall influence of semantic tags at the spatiotemporal level and can provide a quantitative reference for prediction error correction; the contribution matrix is ​​a two-dimensional matrix structure with semantic tags as rows and each dimension or time window of the joint spatiotemporal features as columns, and its matrix elements are the values ​​between tags and features after similarity measurement, which are used to reflect the response strength of semantic tags to each spatiotemporal feature; the weight distribution vector represents the relative contribution ratio of each spatiotemporal feature under a specific semantic tag, which is used to map similarity to an interpretable weight distribution; the weighted response coefficient represents the response strength of a specific semantic tag in the current spatiotemporal context, which is used to convert statistical weights into semantic response indicators.

[0091] In step S4, the user's historical payment records are obtained, and the user's payment tag is predicted based on the feature contribution and the historical payment records to obtain the semantic verification result.

[0092] It should be noted that in this embodiment, the user's historical payment records are obtained through a payment data management database. The payment data management database is a structured data management system used to centrally store and manage user payment data, account information, and payment behavior logs. The payment data management database can be a distributed database. The historical payment records recorded in the payment data management database include fields such as user identifier, payment time, payment amount, payment type, merchant number, payment method, and payment terminal, which can completely reflect the user's payment behavior trajectory. In actual implementation, the payment data management database can obtain historical payment records by calling the payment system's log interface. In this embodiment, the historical payment records can be selected from the current payment tag acquisition time to the past 7 days, which is beneficial for providing reliable data support for prediction error feedback and semantic verification.

[0093] In this embodiment, the prediction error feedback of the user's payment tag based on the feature contribution and the historical payment record, and then the semantic verification result, can be obtained in the following way:

[0094] Construct a tag entity set based on the user's payment tags;

[0095] Based on the feature contribution and the historical payment records, the tag entity set is predicted and paired to obtain the prediction deviation value;

[0096] The prediction pairing process is updated by feeding back the prediction deviation value to obtain the semantic verification vector.

[0097] The semantic verification result is determined based on the semantic verification vector and the preset confidence threshold.

[0098] It should be noted that the user's payment tags include tags such as payment amount, number of payments, and average payment interval. In specific implementation, firstly, the user's payment tags can be preprocessed through text cleaning (such as removing spaces, unifying capitalization, and mapping synonyms) to construct a tag entity set for comparison. Secondly, a mapping relationship is established between the tag entity set, semantic tag sequence, and feature contribution, that is, a weighted score is calculated for the historical frequency and feature contribution of each tag entity, and the weighted score is compared with the historical true statistical value to obtain the prediction deviation value. Preferably, the mean squared error can be used as the difference comparison method, which is beneficial for simultaneous statistics and difference comparison.

[0099] In specific implementation, the feedback update of the prediction pairing process can adopt an error backpropagation algorithm based on gradient descent. In actual implementation, the prediction deviation value can be input into the feedback update model of the error backpropagation algorithm based on gradient descent, with the mean squared error as the loss function, and the gradient direction between the model's predicted score and the historical true statistical value can be calculated. The semantic mapping parameters and label weights are adjusted by the gradient direction so that the model's output distribution gradually approximates the statistical distribution of historical true behavior. Preferably, the Adam optimization algorithm can be used for parameter adaptive adjustment to avoid oscillations caused by excessive learning rate. The response score of each label is obtained, and all response scores are converted into a score vector between 0 and 1. This confidence score vector is used as the semantic verification vector. In this embodiment, The prediction pairing process refers to matching the model's output label prediction results with historical true labels one by one and calculating the differences between them. Finally, the semantic verification vector is judged according to a preset confidence threshold. The confidence threshold can be determined based on the statistical distribution of historical samples, for example, by calculating the mean of the historical confidence of all labels and using this mean as the confidence threshold to balance recall and precision. In actual implementation, the semantic confidence of the semantic verification vector output by the feedback update model can be compared with the confidence threshold. That is, when the semantic confidence is greater than or equal to the confidence threshold, the semantic verification vector is judged to be semantically verified; otherwise, it is recorded as a low-confidence semantic verification vector. The set of all semantic verification vectors that pass the verification is taken as the semantic verification result set.

[0100] It should be noted that the semantic verification vector in this application refers to the set of updated label confidence scores, which has the same dimension as the payment labels and can be used to transform semantic matching results into quantifiable confidence indicators; the confidence threshold is the boundary value used to distinguish between high-confidence labels and low-confidence labels; the semantic verification result refers to the set of labels obtained by comparing the semantic verification vector with the confidence threshold, which includes labels that the model considers to be semantically reliable and consistent with the user's actual payment behavior, and can be used for risk identification and adaptive recommendation processes; the label entity set refers to the set of labels formed after standardization and deduplication, which can be used to establish a unified label benchmark for prediction comparison, and facilitate the calculation of the deviation between historical statistical features and model prediction results.

[0101] In step S5, risk identification and behavior recommendation are performed on the user's payment tags based on the semantic verification results to obtain payment decision tags for adaptive recommendation.

[0102] In this embodiment, risk identification and behavior recommendation are performed on the user's payment tags based on the semantic verification results to obtain payment decision tags for adaptive recommendation. Specifically, this can be achieved in the following manner:

[0103] The semantic verification results are subjected to risk factor deconstruction analysis to obtain a risk feature matrix;

[0104] The user's payment tags are filtered using the risk feature matrix to obtain the filtered payment tags.

[0105] An adaptive optimization strategy network is used to rank the filtered payment tags to obtain payment decision tags for adaptive recommendation.

[0106] In practice, firstly, historical payment records corresponding to the semantic verification results are retrieved from the payment data management database. Structured query statements are used to extract the data into the Pandas analysis environment. Risk factors are calculated for the extracted data according to the label dimension (risk factors include: mean semantic confidence score, historical frequency of label occurrence, mean and variance of payment amount, expected deviation rate, abnormal payment ratio, etc.). Multiple risk factors for each label are then organized into a matrix of rows in a fixed column order to obtain the risk feature matrix. Next, risk filtering is performed on the user's payment labels based on the risk feature matrix. Specifically, the risk feature matrix is ​​input into a Gaussian mixture model, and filtering is performed using this model to obtain the filtered payment labels. In other words, the risk confidence score for each payment label is calculated using the Gaussian mixture model and compared with a preset threshold to determine the risk level. Tags with confidence scores exceeding a preset threshold are marked and removed, while tags below the preset threshold that meet the business whitelist are retained. For tags in the intermediate risk range, a secondary judgment can be made based on rules (e.g., priority is given to those with small amount fluctuations and high historical frequency). The preset threshold can be set according to the actual number and amount requirements of payment tags. Finally, the adaptive optimization strategy network is a multilayer perceptron network. The input of the multilayer perceptron network includes user profile vectors (read from a multi-dimensional user profile matrix), semantic embedding vectors of each candidate tag, behavior matching scores and corresponding risk confidence scores, and is trained in batches using cross-entropy and the Adam optimizer. The filtered payment tags are input into this adaptive optimization strategy network, and the payment decision tags used for adaptive recommendation are obtained through the output of this adaptive optimization strategy network.

[0107] It should be noted that, in this application, the payment decision label refers to a set of labels with confidence scores and risk indicators output after risk filtering and adaptive sorting, which can be used as the decision basis for the system front end for personalized display or real-time push; the risk feature matrix refers to a two-dimensional numerical matrix organized by labels as rows and various risk factors as columns, which is used to express the risk performance of each candidate label in a structured form in dimensions such as semantic confidence, frequency, amount stability and anomaly.

[0108] In summary, the technical solution adopted in this application can realize self-learning and semantic dynamic optimization of payment tags based on multi-source behavioral features, so as to improve the accuracy of payment tag recognition.

[0109] Example 2: This application provides a payment tag intelligent processing system based on multi-dimensional user profiles, referencing... Figure 4 As shown, this figure is a module structure diagram of a payment tag intelligent processing system based on multi-dimensional user profiles provided in this application. The intelligent processing system includes:

[0110] User profile building module 100 is used to build a multi-dimensional user profile matrix based on multi-source behavioral features;

[0111] The spatiotemporal feature extraction module 200 is used to perform spatiotemporal self-attention joint extraction of user payment behavior based on the multi-dimensional user profile matrix to obtain joint spatiotemporal features, and then perform dual-channel filtering on the joint spatiotemporal features to obtain a candidate payment tag set.

[0112] The semantic reordering module 300 is used to perform semantic matching and reordering on the candidate payment tag set through semantic relevance mapping and contrastive learning mechanisms respectively to obtain a semantic tag sequence, and then determine the feature contribution based on the semantic tag sequence and the joint spatiotemporal features.

[0113] The feedback verification module 400 is used to obtain the user's historical payment records, and to perform prediction error feedback on the user's payment tags based on the feature contribution and the historical payment records, thereby obtaining the semantic verification result;

[0114] The risk identification and recommendation module 500 is used to identify risks and recommend behaviors for the user's payment tags based on the semantic verification results, and obtain payment decision tags for adaptive recommendation.

[0115] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0116] Those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, including read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-Erasable Programmable Read-Only Memory (EEPROM), compactdisc read-only memory (CD-ROM) or other optical disc storage, disk storage, magnetic tape storage, or any other computer-readable medium capable of carrying or storing data.

[0117] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

Claims

1. A payment tag intelligent processing method based on multi-dimensional user profiles, characterized in that, The intelligent processing method includes the following steps: Construct a multi-dimensional user profile matrix based on multi-source behavioral features; Based on the multi-dimensional user profile matrix, spatiotemporal self-attention joint extraction of user payment behavior is performed to obtain joint spatiotemporal features. Then, the joint spatiotemporal features are filtered through dual channels to obtain a candidate payment tag set. The candidate payment tag set is semantically matched and rearranged using semantic relevance mapping and contrastive learning mechanisms to obtain a semantic tag sequence. Then, the feature contribution is determined based on the semantic tag sequence and the joint spatiotemporal features. Obtain the user's historical payment records, and based on the feature contribution and the historical payment records, perform prediction error feedback on the user's payment tags to obtain semantic verification results; Based on the semantic verification results, risk identification and behavior recommendation are performed on the user's payment tags to obtain payment decision tags for adaptive recommendation; Specifically, the spatiotemporal self-attention joint extraction of user payment behavior based on the multi-dimensional user profile matrix yields the following joint spatiotemporal features: Determine the dependency weights of user payment behavior; Based on the dependency weights, temporal attention dependency modeling and spatial attention dependency modeling are performed on the multi-dimensional user profile matrix to obtain the temporal attention features and spatial attention features of user payment behavior. The temporal attention features and the spatial attention features are fused to obtain joint spatiotemporal features; Specifically, the dual-channel screening of the joint spatiotemporal features to obtain the candidate payment tag set includes: A semantic feature mapping channel is constructed based on a semantic embedding network, and a rule constraint channel is constructed based on a preset rule set; The joint spatiotemporal features are input into the semantic feature mapping channel, and the joint spatiotemporal features are mapped to the label semantic space through the semantic embedding network, thereby obtaining a set of candidate labels. The candidate tag set is logically constrained and filtered using the rule matching matrix in the rule constraint channel to obtain the candidate payment tag set; Specifically, the semantic tag sequence obtained by semantically matching and rearranging the candidate payment tag set through semantic relevance mapping and contrastive learning mechanisms includes: Semantic embedding is performed on the candidate payment tag set to obtain a semantic embedding matrix; The semantic relevance between the semantic embedding matrix and the label semantic space is determined, and then the relevance ranking result is determined based on the semantic relevance. The relevance ranking results are compared and optimized to obtain a semantic label sequence.

2. The intelligent processing method for payment tags based on multi-dimensional user profiles as described in claim 1, characterized in that, Constructing a multi-dimensional user profile matrix based on multi-source behavioral features specifically includes: The multi-source behavioral features are time-stamp aligned and normalized to obtain a behavioral feature set; The behavioral feature set is subjected to feature hierarchical encoding to obtain the relevance of each feature layer; A multi-dimensional user profile matrix is ​​constructed based on all the relevances and the behavioral feature set.

3. The intelligent processing method for payment tags based on multi-dimensional user profiles as described in claim 1, characterized in that, Determining the feature contribution based on the semantic label sequence and the joint spatiotemporal features specifically includes: Based on the correlation, the feature mapping weights of the semantic label sequence and the joint spatiotemporal features are calculated, and then the contribution matrix is ​​obtained; The weight distribution vector is determined by the contribution matrix, and then the weighted response coefficient of each semantic tag and the joint spatiotemporal feature in the semantic tag sequence is determined based on the weight distribution vector. All weighted response coefficients are aggregated according to the time and spatial dimensions to obtain the feature contribution.

4. The intelligent processing method for payment tags based on multi-dimensional user profiles as described in claim 1, characterized in that, The payment data management database retrieves users' historical payment records.

5. The intelligent processing method for payment tags based on multi-dimensional user profiles as described in claim 1, characterized in that, Based on the feature contribution and the historical payment records, the user's payment tag is predicted with error feedback, and the semantic verification result is obtained, specifically including: Construct a tag entity set based on the user's payment tags; Based on the feature contribution and the historical payment records, the tag entity set is predicted and paired to obtain the prediction deviation value; The prediction pairing process is updated by feeding back the prediction deviation value to obtain the semantic verification vector. The semantic verification result is determined based on the semantic verification vector and the preset confidence threshold.

6. The intelligent processing method for payment tags based on multi-dimensional user profiles as described in claim 1, characterized in that, The multi-source behavioral features refer to a multi-dimensional data set composed of user payment behavior attributes under different time, scenario and device conditions. The multi-source behavioral features include: payment features, terminal features and environmental features.

7. A payment tag intelligent processing system based on multi-dimensional user profiles, used to execute the payment tag intelligent processing method based on multi-dimensional user profiles as described in any one of claims 1 to 6, characterized in that, The intelligent processing system includes: The user profile building module is used to construct a multi-dimensional user profile matrix based on multi-source behavioral features. The spatiotemporal feature extraction module is used to perform spatiotemporal self-attention joint extraction of user payment behavior based on the multi-dimensional user profile matrix to obtain joint spatiotemporal features, and then perform dual-channel filtering on the joint spatiotemporal features to obtain a candidate payment tag set. The semantic reordering module is used to perform semantic matching and reordering on the candidate payment tag set through semantic relevance mapping and contrastive learning mechanisms respectively to obtain a semantic tag sequence, and then determine the feature contribution based on the semantic tag sequence and the joint spatiotemporal features. The feedback verification module is used to obtain the user's historical payment records, and based on the feature contribution and the historical payment records, to provide prediction error feedback for the user's payment tags, thereby obtaining semantic verification results. The risk identification and recommendation module is used to identify risks and recommend behaviors for users' payment tags based on the semantic verification results, and obtain payment decision tags for adaptive recommendation.

Images