A data processing method, device, apparatus, and storage medium

Abstract

This invention discloses a data processing method, apparatus, device, and storage medium. The method includes: acquiring at least two event data sequences generated by a target user in at least two business domains; extracting features from each event data sequence corresponding to each business domain to determine event feature information corresponding to each business domain, and performing information fusion to determine first fused feature information; performing cross-attention processing on pairwise event data belonging to different business domains to determine event-related feature information corresponding to each event data combination, and performing information fusion to determine second fused feature information; and determining target fused feature information based on the first and second fused feature information. Through the technical solution of this invention, coarse-grained fusion at the business domain level and fine-grained fusion at the event level can be combined, improving the data fusion effect between multiple business domains and thus improving the accuracy of user analysis.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to computer technology, and particularly, to a data processing method and device, equipment and a storage medium. BACKGROUND

[0002] With the rapid development of computer technology, users often generate event data in multiple business fields, so that the user can be more accurately analyzed according to the event data generated by the user in multiple business fields, such as predicting whether the user will have a specific behavior, etc.

[0003] At present, the event data generated in multiple business fields is usually fused between the entire business fields to obtain fused feature information, and the user is analyzed based on the fused feature information.

[0004] However, in the process of implementing the present application, the inventors have found that at least the following problems exist in the prior art:

[0005] The existing feature fusion method is only for feature fusion between the entire business field, that is, coarse-grained fusion at the business field level, so that the fused feature information cannot fully represent the user features, reducing the data fusion effect and further reducing the accuracy of user analysis. SUMMARY

[0006] Embodiments of the present application provide a data processing method, device, equipment and storage medium to combine coarse-grained fusion at the business field level and fine-grained fusion at the event level, improve the data fusion effect between multiple business fields, and further improve the accuracy of user analysis.

[0007] In a first aspect, embodiments of the present application provide a data processing method, comprising:

[0008] obtaining at least two event data sequences generated by a target user in at least two business fields, wherein each event data sequence includes at least two event data;

[0009] performing feature extraction on each event data sequence corresponding to each business field to determine event feature information corresponding to each business field, and performing information fusion on each event feature information to determine first fused feature information after fusion;

[0010] performing cross-attention processing on two event data in each event data sequence belonging to different business fields to determine event-related feature information corresponding to each event data combination, and performing information fusion on each event-related feature information to determine second fused feature information after fusion;

[0011] determine target fusion feature information corresponding to the target user based on the first fusion feature information and the second fusion feature information.

[0012] In a second aspect, an embodiment of the present application further provides a data processing apparatus, comprising:

[0013] an event data sequence obtaining module, configured to obtain at least two event data sequences generated by a target user in at least two business fields, wherein each of the event data sequences comprises at least two event data;

[0014] a first fusion feature information determining module, configured to perform feature extraction on each of the event data sequences corresponding to each business field, determine event feature information corresponding to each business field, and perform information fusion on the event feature information to determine first fusion feature information after fusion;

[0015] a second fusion feature information determining module, configured to perform cross-attention processing on two-by-two event data belonging to different business fields in each of the event data sequences, determine event related feature information corresponding to each event data combination, and perform information fusion on the event related feature information to determine second fusion feature information after fusion;

[0016] a target fusion feature information determining module, configured to determine target fusion feature information corresponding to the target user based on the first fusion feature information and the second fusion feature information.

[0017] In a third aspect, an embodiment of the present application further provides an electronic device, comprising:

[0018] one or more processors;

[0019] a memory, configured to store one or more programs;

[0020] when the one or more programs are executed by the one or more processors, the one or more processors implement the data processing method provided by any of the embodiments of the present application.

[0021] In a fourth aspect, an embodiment of the present application further provides a computer readable storage medium, having a computer program stored thereon, which is executed by a processor to implement the data processing method provided by any of the embodiments of the present application.

[0022] An embodiment of the above application has the following advantages or beneficial effects:

[0023] By acquiring at least two event data sequences generated by the target user in at least two business fields, a plurality of event data generated by the target user in each business field can be obtained. Feature extraction is performed on each event data sequence corresponding to each business field, event feature information corresponding to each business field is determined, and information fusion is performed on each event feature information to determine first fusion feature information after fusion, so as to obtain first fusion feature information after fusion at a business field level and with a coarse granularity. Cross-attention processing is performed on two event data belonging to different business fields in each event data sequence, event-related feature information corresponding to each event data combination is determined, and information fusion is performed on each event-related feature information to determine second fusion feature information after fusion, so as to obtain second fusion feature information after fusion at an event level and with a fine granularity. Therefore, the target fusion feature information that more fully represents the user features can be obtained by combining the first fusion feature information and the second fusion feature information, the data fusion effect between the plurality of business fields is improved, and the accuracy of the user analysis is also improved. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0025] Figure 1 is a flowchart of a data processing method provided by an embodiment of the present application;

[0026] Figure 2 is a flowchart of another data processing method provided by an embodiment of the present application;

[0027] Figure 3 is a structural example diagram of a preset cross-attention processing model provided by an embodiment of the present application;

[0028] Figure 4 is a structural schematic diagram of a data processing apparatus provided by an embodiment of the present application;

[0029] Figure 5 is a structural schematic diagram of an electronic device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0030] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0031] Figure 1 This is a flowchart illustrating a data processing method according to an embodiment of the present invention. This embodiment is applicable to the fusion of data from multiple business domains of a user, particularly in situations where data from multiple business domains of a user is fused to predict specific user behaviors. The method can be executed by a data processing device, which can be implemented in software and / or hardware and integrated into an electronic device. Figure 1 As shown, the method specifically includes the following steps:

[0032] S110. Obtain at least two event data sequences generated by the target user in at least two business domains, wherein each event data sequence includes at least two event data.

[0033] In this context, "target user" refers to any user requiring data processing across multiple business domains. A business domain refers to the business area in which the target user participates. For example, a business domain could include, but is not limited to, the shopping domain where a user purchases items, and the financial domain where a user borrows money. An event data sequence refers to multiple event data points generated within the same business domain for the same user's same behavior. The event data points in an event data sequence can be arranged in chronological order of their occurrence. Each business domain corresponds to one event data sequence. Each event data point in an event data sequence can include multiple event characteristic information. For example, in the shopping domain, the corresponding event data sequence could include shopping order data for each purchase by the user. In the financial domain, the corresponding event data sequence could include loan data for each borrowing by the user.

[0034] Specifically, based on the target user identifier, event data sequences generated by the target user within a preset time period are collected from the big data of each business domain. For example, the shopping event data sequence X generated by the target user in the shopping business domain is obtained. jdmall And the data sequence X of loan events generated in the financial field jrapp Among them, the shopping event data sequence X generated by the target user in the shopping business domain. jdmall It consists of a series of shopping behavior events of the target user over a period of time (e.g., from January 1, 2021 to April 1, 2022). Assume the shopping event data sequence X... jdmallIncluding four-dimensional features: the time elapsed since the last behavioral event (f1), the product price (f2), the product quantity (f3), and the product category (f4), the data for the j-th event is vectorized as e. jdmall,j =[f 1,j ,f 2,j ,f 3,j ,f 4,j Assume the target user's shopping event data sequence X. jdmall If the length of the data is 100 event data points, then the shopping event data sequence X generated by the target user in the shopping business domain is... jdmall ={e jdmall,1 ,e jdmall,2 ,…,e jdmall,j ,…,e jdmall,100}

[0035] Data sequence X of loan events generated by the target user in the financial field jrapp It consists of borrowing behavior event data of target users within the same time period (January 1, 2021 to April 1, 2022). Assume the borrowing event data sequence X... jrapp Including the three-dimensional features of the time length k1 since the last behavioral event, the loan amount k2, and the number of installments k3, the data of the j-th event is vectorized into e. jrapp,j =[k 1,j ,k 2,j ,k 3,j Assume the target user's loan event data sequence X. jrapp If the length of the data is 80 events, then the loan event data sequence X of this target user in the financial field is... jrapp ={e jrapp,1 ,e jrapp,2 ,…,e jrapp,j ,…,e jrapp,80}

[0036] S120. Extract features from each event data sequence corresponding to each business domain, determine the event feature information corresponding to each business domain, and fuse the event feature information to determine the first fused feature information.

[0037] The event feature information corresponding to each business domain can refer to the event feature information of the entire business domain. For example, the event feature information can be intermediate-layer feature information extracted from the event data sequence. The first fused feature information can refer to the event feature information fused at the business domain level, that is, coarse-grained fused feature information.

[0038] Specifically, based on each event data sequence in each business domain, intermediate layer feature information is extracted using the corresponding encoder for each business domain. This intermediate layer feature information is then identified as the event feature information corresponding to that business domain. The encoder can be a feature extraction device such as LSTM (Long Short-Term Memory) or GRU (Gated Recurrent Unit). The event feature information corresponding to each business domain is then fused at the business domain level. For example, the event feature information corresponding to each business domain is concatenated or weighted and summed to obtain the first fused feature information after coarse-grained fusion.

[0039] For example, suppose we obtain the time series of n events for a target user across n business domains, where the time series of the event data corresponding to the i-th business domain is X. i The encoder corresponding to the i-th business domain is F. i The event feature information extracted by the encoder for the i-th business domain is F. i (X i Assuming the fusion function of event feature information is M, then the final first fused feature information can be M(F1(X1), F2(X2), ..., F...). n (X n )).

[0040] For example, S120 may include: performing a linear transformation on each event feature information to determine the first event feature information under a preset feature dimension; splicing the information of each first event feature information to obtain the spliced ​​second event feature information; and performing cross-fusion of the second event feature information based on a preset multi-layer perception model to determine the fused first fused feature information.

[0041] The first event feature information can refer to the event feature information under a preset feature dimension after linear transformation. The second event feature information can refer to the event feature information obtained by concatenating the various first event feature information. The preset multilayer perception model is a feedforward artificial neural network model that can map multiple input datasets to a single output dataset. In this embodiment, the preset multilayer perception model is obtained by pre-training based on sample data.

[0042] It should be noted that since different business domains may use different encoders, the feature dimensions output by different encoders may also differ. Different feature dimensions cannot be concatenated. However, by performing a linear transformation on the event feature information, it can be ensured that the feature dimensions that meet the concatenation requirements are obtained. For example, if the extracted event feature information meets the feature dimensions required for concatenation, then the event feature information does not need to undergo linear transformation.

[0043] Specifically, each extracted event feature is linearly transformed to obtain a preset feature dimension, and these transformed event features are designated as the first event feature. The first event features are then concatenated, and the concatenated event feature is designated as the second event feature. This second event feature is then input into a pre-trained multilayer perceptron model for cross-fusion. Based on the output of the multilayer perceptron model, the fused first feature is obtained. By utilizing the multilayer perceptron model to cross-fuse the second event feature, the fusion effect of the coarse-grained first fused feature can be further improved.

[0044] S130. Perform cross-attention processing on pairwise event data belonging to different business domains in each event data sequence, determine the event-related feature information corresponding to each event data combination, and perform information fusion on each event-related feature information to determine the second fused feature information.

[0045] Cross-Attention processing refers to a method of extracting information from two event data belonging to different business domains. Event-related feature information refers to the relevant feature information of pairwise event data from different business domains. Secondary fusion feature information refers to the event feature information fused at the event level, i.e., fine-grained fusion feature information. The event data combination can include two event data belonging to different business domains.

[0046] Specifically, based on the event data in each event data sequence corresponding to each business domain, a preset cross-attention processing model can be used to perform cross-attention processing on the pairwise event data belonging to different business domains, determine the event-related feature information corresponding to each event data combination, and fuse the event-related feature information to determine the second fused feature information after fine-grained fusion.

[0047] For example, S130 may include: concatenating the relevant feature information of each event to obtain the concatenated second fused feature information. Specifically, the relevant feature information of each event can be directly concatenated to obtain the second fused feature information that more fully represents the user's characteristics.

[0048] S140. Based on the first fusion feature information and the second fusion feature information, determine the target fusion feature information corresponding to the target user.

[0049] The target fusion feature information can refer to the final fusion feature information achieved by splicing the first fusion feature information and the second fusion feature information.

[0050] Specifically, the first fusion feature information and the second fusion feature information can be concatenated to obtain target fusion feature information that more fully represents user characteristics. This improves the data fusion effect between multiple business domains and, consequently, the accuracy of user analysis.

[0051] The technical solution of this invention obtains at least two event data sequences generated by a target user in at least two business domains, thereby acquiring multiple event data generated by the target user in each business domain. Feature extraction is performed on each event data sequence corresponding to each business domain to determine the event feature information corresponding to each business domain. Information fusion is then performed on each event feature information to determine the first fused feature information, thus obtaining the first fused feature information at the business domain level with coarse-grained fusion. Cross-attention processing is then performed on pairwise event data belonging to different business domains in each event data sequence to determine the event-related feature information corresponding to each event data combination. Information fusion is then performed on each event-related feature information to determine the second fused feature information, thus obtaining the second fused feature information at the event level with fine-grained fusion. By combining the first and second fused feature information, target fused feature information that more fully characterizes user features can be obtained, improving the data fusion effect between multiple business domains and thus improving the accuracy of user analysis.

[0052] Based on the above embodiments, after S140, the method may further include: inputting target fusion feature information into the target behavior prediction model; and determining the target behavior prediction result corresponding to the target user based on the output of the target behavior prediction model.

[0053] The target behavior prediction model is a network model used to predict whether a user will engage in a target behavior. Multiple business domains are associated with the target behavior. This embodiment can determine each business domain based on the target behavior to be predicted. For example, if the target behavior refers to a target user's overdue payment, the corresponding business domains can include shopping domains (such as the target user's purchase of goods) and financial domains (such as the target user's borrowing of loans), reflecting the target user's consumption and repayment capabilities. Therefore, by performing fine-grained and coarse-grained fusion of shopping event sequences in the shopping domain and borrowing event sequences in the financial domain, more comprehensive target fusion feature information representing user characteristics can be obtained. It should be noted that the target behavior prediction model is pre-trained based on sample data to ensure its prediction accuracy.

[0054] Specifically, the target fusion feature information corresponding to the target user is input into a pre-trained target behavior prediction model to predict the target behavior. Based on the output of the target behavior prediction model, it is determined whether the target user will perform the target behavior. Thus, the target fusion feature information obtained by fine-grained fusion and coarse-grained fusion can more accurately predict the target behavior, thereby improving the prediction accuracy of the target behavior.

[0055] Figure 2 This is a flowchart illustrating another data processing method provided by an embodiment of the present invention. Based on the above embodiments, this embodiment describes in detail the specific process of determining the event-related feature information corresponding to each event data combination. Explanations of terms that are the same as or corresponding to those in the above embodiments are not repeated here.

[0056] See Figure 2 Another data processing method provided in this embodiment specifically includes the following steps:

[0057] S210. Obtain at least two event data sequences generated by the target user in at least two business domains, wherein each event data sequence includes at least two event data.

[0058] S220. Extract features from each event data sequence corresponding to each business domain, determine the event feature information corresponding to each business domain, and fuse the event feature information to determine the first fused feature information.

[0059] S230. Combine the pairwise event data belonging to different business domains in the pairwise event data sequence to determine each event data combination.

[0060] Specifically, based on the event data in each event data sequence corresponding to each business domain, every two event data belonging to different business domains are divided into an event data combination, thereby obtaining the event data combinations corresponding to all business domains.

[0061] For example, S230 may include: truncating each event data sequence based on the minimum sequence length in each event data sequence to determine each truncated target event data sequence; and combining pairs of event data belonging to different business domains in each pair of target event data sequences to determine each event data combination.

[0062] The target event data sequence can refer to event data sequences with the same minimum sequence length.

[0063] Specifically, the sequence lengths of each event data sequence corresponding to each business domain are compared to determine the minimum sequence length for each event data sequence. Based on this minimum sequence length, each event data sequence is truncated, and the event data sequences with the same minimum sequence length after truncation are determined as the target event data sequences. It should be noted that this embodiment can truncate the data according to the order of the event data in the event data sequence. For example, if the sequence length of an event data sequence is 100 (containing 100 event data), and the minimum sequence length is 50, then either the first 50 or the last 50 event data in the event data sequence can be truncated to obtain the target event data sequence composed of either the first or the last 50 event data. Combining all target event data sequences with the same sequence length in pairs can reduce the number of event data combinations without affecting the fusion effect, thereby improving data fusion efficiency. It is worth noting that the purpose of truncating each event data sequence based on the minimum sequence length is to reduce computational complexity and does not affect the final fusion result.

[0064] For example, after determining the extracted target event data sequences, a linear transformation can be performed on each target event data sequence to obtain transformed target event data sequences. These transformed target event data sequences ensure that the feature dimensions required for combining into event data sets are obtained. For example, if a particular transformed target event data sequence meets the feature dimensions required for combining into event data sets, then no linear transformation is required for that transformed target event data sequence.

[0065] S240. For each event data combination, determine the target query data, target key information, and target value information corresponding to the event data combination based on the first event data and the second event data in the event data combination.

[0066] Here, the first event data and the second event data can refer to the two event data in each event data combination. The target query data can refer to the query data across the attention processing model. The target key information can refer to the keys data across the attention processing model. The target value information can refer to the values ​​data across the attention processing model.

[0067] Specifically, for each obtained event data combination, the target query data, target key information, and target value information corresponding to each event data combination are determined based on the first event data and the second event data in each event data combination.

[0068] S250. Input the target query data, target key information and target value information into the preset cross-attention processing model, and determine the event-related feature information corresponding to the event data combination based on the output of the preset cross-attention processing model.

[0069] The pre-defined cross-attention processing model can be a network model used to perform attention processing on two event data belonging to different business domains. This pre-defined cross-attention processing model can be obtained by pre-training based on sample data.

[0070] Specifically, for each event data combination, the target query data, target key information, and target value information corresponding to the event data combination are input into a preset cross-attention processing model for correlation processing. Based on the output of the preset cross-attention processing model, the event-related feature information corresponding to each event data combination is determined.

[0071] S260. Perform information fusion on the relevant feature information of each event to determine the second fused feature information.

[0072] S270. Based on the first fusion feature information and the second fusion feature information, determine the target fusion feature information corresponding to the target user.

[0073] The technical solution of this invention, for each event data combination, determines the target query data, target key information, and target value information corresponding to the event data combination based on the first event data and the second event data in the event data combination. The target query data, target key information, and target value information are input into a preset cross-attention processing model, and the event-related feature information corresponding to the event data combination is determined based on the output of the preset cross-attention processing model. Therefore, the preset cross-attention processing model can more quickly determine the event-related feature information corresponding to each event data combination, further improving the data fusion efficiency at a fine-grained level.

[0074] Based on the above embodiments, S240 may include: using the first event data in the event data combination as target query data, and using the second event data in the event data combination as target key information; performing feature dimension data concatenation on the first event data and the second event data, and determining the concatenation result as target value information.

[0075] Specifically, for each combination of event data, the first event data in the combination is identified as the target query data, and the second event data in the combination is identified as the target key information. The first and second event data in the combination are then concatenated along their feature dimensions, and the concatenated result is identified as the target value information.

[0076] For example, for each event data combination, the first event data T1 in the event data combination is taken as the target query data Q, i.e., Q = T1. The second event data T2 in the event data combination is determined as the target key information K, i.e., K = T2. The first event data T1 and the second event data T2 in the event data combination are concatenated according to the feature dimensions, and the concatenation result is determined as the target value information V, i.e., V = Concatenate(T1, T2).

[0077] Based on the above embodiments, the preset cross-attention processing model may include: a first cross-processing sub-model, a normalization sub-model, and a second cross-processing sub-model. S250 may include: inputting target query data and target key information into the first cross-processing sub-model for cross-processing to determine first relevant feature information; inputting the first relevant feature information into the normalization sub-model for information normalization to determine normalized second relevant feature information; inputting the second relevant feature information and target value information into the second cross-processing sub-model for cross-processing to determine event-related feature information.

[0078] The first cross-processing sub-model can be used to perform matrix multiplication of the target query data and target key information to obtain the correlation feature information between the target query data and target key information. The normalization sub-model can be used to normalize the correlation feature information between each target query data and each target key information. For example, the normalization sub-model can use softmax (normalization exponential function) for normalization. The second cross-processing sub-model can be used to perform a weighted summation of the target value information according to weight coefficients. The first correlation feature information can refer to the correlation feature information between the target query data and target key information. The second correlation feature information can refer to the first correlation feature information after normalization.

[0079] Specifically, Figure 3 A structural example diagram of a pre-defined cross-attention processing model is given, such as... Figure 3 As shown, the target query data and target key information are input into the first cross-processing sub-model for cross-processing, and the output of the first cross-processing sub-model is determined as the first relevant feature information. The first relevant feature information is input into the normalization sub-model for information normalization processing, and the processing result of the normalization sub-model is determined as the second relevant feature information. The second relevant feature information and the target value information are input into the second cross-processing sub-model for cross-processing, and the event-related feature information corresponding to the event data combination can be determined. By using the preset cross-attention processing model, the event-related feature information corresponding to each event data combination can be determined more quickly, further improving the data fusion efficiency at a fine-grained level.

[0080] For example, after determining the event-related feature information, a linear transformation can be performed on each event-related feature information to obtain the transformed event-related feature information. The transformed event-related feature information ensures that the feature dimensions required for fusion into the second fused feature information are obtained. For example, if the obtained transformed event-related feature information meets the feature dimensions required for fusion into the second fused feature information, then no linear transformation processing is required for that transformed event-related feature information.

[0081] The following are embodiments of the data processing apparatus provided in this invention. This apparatus and the data processing methods in the above embodiments belong to the same inventive concept. For details not described in detail in the embodiments of the data processing apparatus, please refer to the embodiments of the above data processing methods.

[0082] Figure 4 This is a schematic diagram of a data processing device provided in an embodiment of the present invention. This embodiment is applicable to situations where specific user behaviors are predicted, especially in scenarios where specific user behaviors are predicted based on event data generated in multiple business domains.Figure 4 As shown, the device specifically includes: an event data sequence acquisition module 410, a first fusion feature information determination module 420, a second fusion feature information determination module 430, and a target fusion feature information determination module 440.

[0083] The event data sequence acquisition module 410 is used to acquire at least two event data sequences generated by the target user in at least two business domains, wherein each event data sequence includes at least two event data; the first fusion feature information determination module 420 is used to extract features from each event data sequence corresponding to each business domain, determine the event feature information corresponding to each business domain, and perform information fusion on each event feature information to determine the fused first fusion feature information; the second fusion feature information determination module 430 is used to perform cross-attention processing on pairwise event data belonging to different business domains in each event data sequence, determine the event-related feature information corresponding to each event data combination, and perform information fusion on each event-related feature information to determine the fused second fusion feature information; the target fusion feature information determination module 440 is used to determine the target fusion feature information corresponding to the target user based on the first fusion feature information and the second fusion feature information.

[0084] By acquiring at least two event data sequences generated by the target user in at least two business domains, multiple event data points generated by the target user in each business domain can be obtained. Feature extraction is performed on each event data sequence corresponding to each business domain to determine the event feature information corresponding to each business domain. Information fusion is then performed on these event feature information points to determine the first fused feature information, thus obtaining the first fused feature information at the business domain level (coarse-grained fusion). Cross-attention processing is then applied to pairwise event data points belonging to different business domains within each event data sequence to determine the event-related feature information corresponding to each event data combination. Information fusion is then performed on these event-related feature information points to determine the second fused feature information, thus obtaining the second fused feature information at the event level (fine-grained fusion). By combining the first and second fused feature information, a more comprehensive target fused feature information representing user characteristics can be obtained, improving the data fusion effect across multiple business domains and consequently enhancing the accuracy of user analysis.

[0085] Optionally, the second fusion feature information determination module 430 may include:

[0086] The event data combination unit is used to combine pairs of event data belonging to different business domains in a pairwise event data sequence to determine each event data combination;

[0087] The information determination unit is used to determine the target query data, target key information and target value information corresponding to each event data combination based on the first event data and the second event data in the event data combination.

[0088] The event-related feature information determination unit is used to input target query data, target key information and target value information into a preset cross-attention processing model, and determine the event-related feature information corresponding to the event data combination based on the output of the preset cross-attention processing model.

[0089] Optionally, the event data combination unit is specifically used for:

[0090] Based on the minimum sequence length in each event data sequence, data is truncated from each event data sequence to determine the target event data sequences after truncation; then, event data from each pair of target event data sequences belonging to different business domains are combined to determine the event data combinations.

[0091] Optionally, the information determining unit is specifically used for:

[0092] The first event data in the event data combination is used as the target query data, and the second event data in the event data combination is used as the target key information; the first event data and the second event data are concatenated according to the feature dimensions, and the concatenation result is determined as the target value information.

[0093] Optionally, the preset cross-attention processing model includes: a first cross-processing sub-model, a normalization sub-model, and a second cross-processing sub-model. The event-related feature information determination unit is specifically used for:

[0094] The target query data and target key information are input into the first cross-processing sub-model for cross-processing to determine the first relevant feature information; the first relevant feature information is input into the normalization sub-model for information normalization to determine the normalized second relevant feature information; the second relevant feature information and target value information are input into the second cross-processing sub-model for cross-processing to determine the event-related feature information.

[0095] Optionally, the second fusion feature information determination module 430 is specifically used to: splice the relevant feature information of each event to obtain the spliced ​​second fusion feature information.

[0096] Optionally, the first fusion feature information determination module 420 is specifically used for:

[0097] A linear transformation is performed on each event feature information to determine the first event feature information under a preset feature dimension; the first event feature information is spliced ​​together to obtain the spliced ​​second event feature information; based on a preset multi-layer perception model, the second event feature information is cross-fused to determine the fused first fused feature information.

[0098] Optionally, the data processing apparatus further includes:

[0099] The target behavior prediction result determination module is used to: input target fusion feature information into the target behavior prediction model, wherein the target behavior prediction model is a network model used to predict whether a user will perform a target behavior, and at least two business domains are associated with the target behavior; and determine the target behavior prediction result corresponding to the target user based on the output of the target behavior prediction model.

[0100] The data processing apparatus provided in the embodiments of the present invention can execute the data processing method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects for executing the data processing method.

[0101] It is worth noting that in the embodiments of the above data processing device, the various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of the present invention.

[0102] Figure 5 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Figure 5 A block diagram is shown of an exemplary electronic device 12 suitable for implementing embodiments of the present invention. Figure 5 The electronic device 12 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0103] like Figure 5 As shown, the electronic device 12 is represented in the form of a general-purpose computing device. The components of the electronic device 12 may include, but are not limited to: one or more processors or processing units 16, system memory 28, and bus 18 connecting different system components (including system memory 28 and processing unit 16).

[0104] Bus 18 represents one or more of several bus architectures, including a memory bus or memory controller, a peripheral bus, a graphics acceleration port, a processor, or a local bus using any of the various bus architectures. For example, these architectures include, but are not limited to, the Industry Standard Architecture (ISA) bus, the Micro Channel Architecture (MAC) bus, the Enhanced ISA bus, the Video Electronics Standards Association (VESA) local bus, and the Peripheral Component Interconnect (PCI) bus.

[0105] Electronic device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by electronic device 12, including volatile and non-volatile media, removable and non-removable media.

[0106] System memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Electronic device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 may be used to read and write non-removable, non-volatile magnetic media (… Figure 5 Not shown; usually referred to as a "hard drive"). Although Figure 5 Not shown, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk") and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to bus 18 via one or more data media interfaces. System memory 28 may include at least one program product having a set (e.g., at least one) of program modules configured to perform the functions of the embodiments of the present invention.

[0107] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in system memory 28. Such program modules 42 include, but are not limited to, an operating system, one or more application programs, other program modules, and program data. Each or some combination of these examples may include an implementation of a network environment. Program modules 42 typically perform the functions and / or methods described in the embodiments of the present invention.

[0108] Electronic device 12 can also communicate with one or more external devices 14 (e.g., keyboard, pointing device, display 24, etc.), and with one or more devices that enable a user to interact with electronic device 12, and / or with any device that enables electronic device 12 to communicate with one or more other computing devices (e.g., network card, modem, etc.). This communication can be performed via input / output (I / O) interface 22. Furthermore, electronic device 12 can also communicate with one or more networks (e.g., local area network (LAN), wide area network (WAN), and / or public networks, such as the Internet) via network adapter 20. As shown, network adapter 20 communicates with other modules of electronic device 12 via bus 18. It should be understood that, although not shown in the figures, other hardware and / or software modules can be used in conjunction with electronic device 12, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0109] Processing unit 16 executes various functional applications and data processing by running programs stored in system memory 28, such as implementing the steps of a data processing method provided in this embodiment, the method including:

[0110] Acquire at least two event data sequences generated by the target user in at least two business domains, wherein each event data sequence includes at least two event data;

[0111] Feature extraction is performed on each event data sequence corresponding to each business domain to determine the event feature information corresponding to each business domain, and information fusion is performed on each event feature information to determine the first fused feature information.

[0112] Cross-attention processing is performed on pairs of event data belonging to different business domains in each event data sequence to determine the event-related feature information corresponding to each event data combination, and information fusion is performed on each event-related feature information to determine the fused second fused feature information;

[0113] Based on the first fusion feature information and the second fusion feature information, the target fusion feature information corresponding to the target user is determined.

[0114] Of course, those skilled in the art will understand that the processor can also implement the technical solutions of the data processing methods provided in any embodiment of the present invention.

[0115] This embodiment provides a computer-readable storage medium storing a computer program thereon. When executed by a processor, the program implements the data processing method steps provided in any embodiment of the present invention, the method comprising:

[0116] Acquire at least two event data sequences generated by the target user in at least two business domains, wherein each event data sequence includes at least two event data;

[0117] Feature extraction is performed on each event data sequence corresponding to each business domain to determine the event feature information corresponding to each business domain, and information fusion is performed on each event feature information to determine the first fused feature information.

[0118] Cross-attention processing is performed on pairs of event data belonging to different business domains in each event data sequence to determine the event-related feature information corresponding to each event data combination, and information fusion is performed on each event-related feature information to determine the fused second fused feature information;

[0119] Based on the first fusion feature information and the second fusion feature information, the target fusion feature information corresponding to the target user is determined.

[0120] The computer storage medium of this invention can be any combination of one or more computer-readable media. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this document, a computer-readable storage medium can be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, apparatus, or device.

[0121] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, capable of sending, propagating, or transmitting programs for use by or in connection with an instruction execution system, apparatus, or device.

[0122] Program code contained on a computer-readable medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0123] Computer program code for performing the operations of this invention can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or it can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0124] Those skilled in the art will understand that the modules or steps of the present invention described above can be implemented using general-purpose computing devices. They can be centralized on a single computing device or distributed across a network of multiple computing devices. Optionally, they can be implemented using computer-executable program code, thereby allowing them to be stored in a storage device for execution by a computing device, or they can be fabricated as separate integrated circuit modules, or multiple modules or steps can be fabricated as a single integrated circuit module. Thus, the present invention is not limited to any particular combination of hardware and software.

[0125] Note that the above description is merely a preferred embodiment of the present invention and the technical principles employed. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and various obvious changes, readjustments, and substitutions can be made without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in detail through the above embodiments, the present invention is not limited to the above embodiments, and may include many other equivalent embodiments without departing from the concept of the present invention, the scope of which is determined by the scope of the appended claims.

Claims

1. A data processing method, characterized by, The method comprises: obtaining at least two event data sequences generated by a target user in at least two business fields, wherein each event data sequence comprises at least two event data; performing feature extraction on each event data sequence corresponding to each business field, determining event feature information corresponding to each business field, and performing information fusion on each event feature information to determine first fused feature information after fusion; performing cross-attention processing on two event data in each event data sequence that belong to different business fields, determining event-related feature information corresponding to each event data combination, and performing information fusion on each event-related feature information to determine second fused feature information after fusion; determining target fused feature information corresponding to the target user based on the first fused feature information and the second fused feature information.

2. The method of claim 1, wherein, The cross-attention processing on two event data in each event data sequence that belong to different business fields, the determination of event-related feature information corresponding to each event data combination, comprises: combining two event data in each event data sequence that belong to different business fields to determine each event data combination; for each event data combination, determining target query data, target key information and target value information corresponding to the event data combination according to first event data and second event data in the event data combination; inputting the target query data, the target key information and the target value information into a preset cross-attention processing model, and determining event-related feature information corresponding to the event data combination according to the output of the preset cross-attention processing model.

3. The method of claim 2, wherein, The combination of two event data in each event data sequence that belong to different business fields to determine each event data combination comprises: based on the minimum sequence length in each event data sequence, data truncation is performed on each event data sequence to determine each target event data sequence after truncation; combining two event data in each target event data sequence that belong to different business fields to determine each event data combination.

4. The method of claim 2, wherein, The determination of target query data, target key information and target value information corresponding to the event data combination according to first event data and second event data in the event data combination comprises: taking the first event data in the event data combination as the target query data, and taking the second event data in the event data combination as the target key information; performing feature dimension data splicing on the first event data and the second event data, and determining the splicing result as the target value information.

5. The method of claim 2, wherein, The preset cross-attention processing model comprises: a first cross processing sub-model, a normalization sub-model and a second cross processing sub-model, the inputting of the target query data, the target key information and the target value information into the preset cross-attention processing model comprises: inputting the target query data and the target key information into the first cross processing sub-model for cross processing to determine first related feature information; inputting the first related feature information into the normalization sub-model to perform information normalization, and determining second related feature information after normalization; inputting the second related feature information and the target value information into the second cross processing sub-model to perform cross processing, and determining event related feature information.

6. The method of claim 1, wherein, The information fusion of each event related feature information to determine the second fusion feature information after fusion includes: performing information splicing on each event related feature information to obtain second fusion feature information after splicing.

7. The method of claim 1, wherein, The information fusion of each event feature information to determine the first fusion feature information after fusion includes: performing linear transformation on each event feature information to determine first event feature information in a preset feature dimension; performing information splicing on each first event feature information to obtain second event feature information after splicing; based on a preset multi-layer perception model, performing information cross fusion on the second event feature information to determine the first fusion feature information after fusion.

8. The method according to any one of claims 1 to 7, characterized in that, After determining the target fusion feature information corresponding to the target user, further comprising: inputting the target fusion feature information into a target behavior prediction model, wherein the target behavior prediction model is a network model for predicting whether a user has a target behavior, and the at least two business fields are associated with the target behavior; based on the output of the target behavior prediction model, determining a target behavior prediction result corresponding to the target user.

9. A data processing apparatus, characterized by, including: an event data sequence acquisition module configured to acquire at least two event data sequences generated by a target user in at least two business fields, wherein each event data sequence includes at least two event data; a first fusion feature information determination module configured to extract features from each event data sequence corresponding to each business field, determine event feature information corresponding to each business field, and fuse information of each event feature information to determine first fusion feature information after fusion; a second fusion feature information determination module configured to perform cross attention processing on two events data belonging to different business fields in each event data sequence to determine event related feature information corresponding to each event data combination, and fuse information of each event related feature information to determine second fusion feature information after fusion; a target fusion feature information determination module configured to determine target fusion feature information corresponding to the target user based on the first fusion feature information and the second fusion feature information.

10. An electronic device, comprising: The electronic device includes: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the data processing method of any one of claims 1-8.

11. A computer readable storage medium having stored thereon a computer program, characterized in that The program is executed by the processor to implement the data processing method of any one of claims 1-8.

Images