Financial content pushing method and device, equipment and storage medium
By combining the deep multimodal learning model and the reinforcement learning model, dynamic update of user preference information is solved, and the problem of low accuracy in financial product recommendations in the existing technology is solved, achieving higher user experience and satisfaction.
Patent Information
- Application Number
- CN202411490838.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-10-24
- Publication Date
- 2025-02-07
AI Technical Summary
The existing technology cannot accurately control users' current preferences, resulting in low accuracy in financial product recommendations, affecting user experience and satisfaction.
Through the deep multimodal learning model, users' historical data are analyzed, real-time operational data and traffic data are combined, user preference information is dynamically updated, and reinforcement learning model is used to output financial products to be pushed.
It improves the accuracy of financial product recommendations, makes the pushed financial products more in line with users' real-time preferences, and improves user experience and satisfaction.
Smart Images
Figure CN119398852A_ABST
Abstract
Description
Technical Field
[0001] The present application relates to the field of data processing technology, and in particular to a financial content push method, device, equipment and storage medium. Background Art
[0002] With the development of Internet finance, the amount of data generated has exploded, and traffic has become one of the key factors in measuring the success of Internet products and services. In this context, how to push suitable financial products to potential users has become an important challenge facing financial technology companies.
[0003] At present, in the prior art, financial products are pushed to users mainly through the historical operation data of users, for example, through the historical consumption records of financial products by users, to achieve the purpose of recommending similar financial products to users.
[0004] However, the existing technology of analyzing user historical data cannot accurately control the user's current preferences, resulting in low accuracy of recommended content, thereby affecting user experience and satisfaction. Summary of the invention
[0005] The present application provides a financial content push method, device, equipment and storage medium to improve the accuracy of recommended content.
[0006] In a first aspect, the present application provides a financial content push method, which is applied to a server, and the method includes:
[0007] Acquire historical data of the front-end to be pushed, wherein the front-end to be pushed belongs to a user who has been authorized to collect data, and the historical data includes historical operation data and historical traffic data;
[0008] Inputting the historical operation data and historical traffic data of the front end to be pushed into a deep multimodal learning model to obtain user preference information;
[0009] Extract the real-time operation data and real-time traffic data of the front end to be pushed through a preset real-time data processing framework;
[0010] Adjusting the user preference information according to the real-time operation data and the real-time traffic data to obtain adjusted user preference information;
[0011] Inputting the adjusted user preference information into a reinforcement learning model to output the financial products to be pushed;
[0012] The financial product to be pushed is pushed to the front end to be pushed, so that the front end to be pushed displays the financial product to be pushed.
[0013] In a possible design, the training process of the deep multimodal learning model includes:
[0014] Receiving model parameters sent by multiple front ends, wherein the model parameters are model parameters of local models obtained by each front end training an initial global model according to local historical data;
[0015] Aggregation algorithms are used to aggregate the parameters of each model to generate a new global model;
[0016] Sending the generated new global model to each front end so that each front end trains the new global model according to local historical data to update the model parameters of the local model;
[0017] The steps from receiving model parameters sent by multiple front ends to sending the generated new global model to each front end are executed cyclically until the global model converges or reaches a preset number of cycles to obtain a trained deep multimodal learning model.
[0018] In one possible design, the method further includes:
[0019] Obtaining recommendation logic information corresponding to the financial product to be pushed;
[0020] The recommendation logic information corresponding to the financial product to be pushed is pushed to the front end to be pushed, so that the front end to be pushed displays the recommendation logic information at the corresponding position of the financial product to be pushed.
[0021] In one possible design, the method further includes:
[0022] Receiving feedback sent by the front end to be pushed, wherein the feedback is generated by the front end to be pushed in response to a user's feedback operation on the recommendation logic information;
[0023] The recommendation logic information is updated according to the feedback.
[0024] In one possible design, the method further includes:
[0025] Receiving at least one prediction request sent by the front end to be pushed, wherein the prediction request is generated by the front end to be pushed in response to the prediction task issued by the user;
[0026] Responding to each prediction request to collect the historical data to be pushed to the front end;
[0027] Input the collected historical data into the multi-task learning model to output the prediction result corresponding to each prediction request;
[0028] The prediction result corresponding to each prediction request is sent to the front end to be pushed, so that the front end to be pushed displays the prediction result corresponding to each prediction request.
[0029] In a possible design, the training process of the multi-task learning model includes:
[0030] determining at least one prediction request;
[0031] Obtain a data set corresponding to each prediction request, wherein the data set includes prediction content and actual results;
[0032] Set the initial weight for each prediction request in the multi-task learning model;
[0033] Inputting the data set corresponding to each prediction request into the multi-task learning model so that the multi-task learning model outputs the prediction result corresponding to each prediction request;
[0034] Compare the predicted results with the actual results to obtain the loss value corresponding to each prediction request;
[0035] If the loss value of at least one prediction request does not meet the preset range, adjust the weight corresponding to each prediction request;
[0036] Repeat the steps of inputting the data set corresponding to each prediction request into the multi-task learning model and adjusting the weight corresponding to each prediction request until the loss value between each prediction result output by the multi-task learning model and the actual result meets the preset range, then stop training the multi-task learning model.
[0037] In a second aspect, the present application provides a financial content push device, which is applied to a server, and the device includes:
[0038] An acquisition module is used to acquire historical data of the front end to be pushed, wherein the front end to be pushed belongs to a user who has been authorized to collect data, and the historical data includes historical operation data and historical traffic data;
[0039] An input module, used for inputting the historical operation data and historical traffic data of the front end to be pushed into the deep multimodal learning model to obtain user preference information;
[0040] An extraction module, used to extract the real-time operation data and real-time traffic data of the front end to be pushed through a preset real-time data processing framework;
[0041] An adjustment module, used for adjusting the user preference information according to the real-time operation data and the real-time traffic data to obtain adjusted user preference information;
[0042] The input module is further used to input the adjusted user preference information into the reinforcement learning model to output the financial products to be pushed;
[0043] The push module is used to push the financial product to be pushed to the front end to be pushed, so that the front end to be pushed displays the financial product to be pushed.
[0044] In a third aspect, an embodiment of the present application provides an electronic device comprising: at least one processor and a memory; the memory stores computer execution instructions; the at least one processor executes the computer execution instructions stored in the memory, so that the at least one processor executes the financial content push method described in the first aspect and various possible designs of the first aspect.
[0045] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, in which computer execution instructions are stored. When a processor executes the computer execution instructions, the financial content push method described in the first aspect and various possible designs of the first aspect is implemented.
[0046] In a fifth aspect, an embodiment of the present application provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the financial content push method described in the first aspect and various possible designs of the first aspect.
[0047] The financial content push method, device, equipment and storage medium provided by the present application obtain user preference information by using a deep multimodal learning model to analyze the historical data of the front end to be pushed, dynamically update the user preference information by collecting the real-time operation data and real-time flow data of the front end to be pushed, and analyze the user's preference information by using a reinforcement learning model to output the financial products to be pushed, and push the financial products to be pushed to the front end to be pushed for display. The present application achieves the purpose of dynamically updating user preferences by collecting real-time operation data and flow data of users, so that the launched financial products are more in line with the real-time preferences of users and the correctness of the pushed financial products is improved. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.
[0049] Figure 1 This is the application scenario diagram of this application;
[0050] Figure 2 The process of the financial content push method provided in the embodiment of the present application Figure 1 ;
[0051] Figure 3The process of the financial content push method provided in the embodiment of the present application Figure 2 ;
[0052] Figure 4 The process of the financial content push method provided in the embodiment of the present application Figure 3 ;
[0053] Figure 5 The process of the financial content push method provided in the embodiment of the present application Figure 4 ;
[0054] Figure 6 The process of the financial content push method provided in the embodiment of the present application Figure 5 ;
[0055] Figure 7 The process of the financial content push method provided in the embodiment of the present application Figure 6 ;
[0056] Figure 8 A schematic diagram of the structure of a financial content push device provided in an embodiment of the present application;
[0057] Fig. 9 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.
[0058] The above drawings have shown clear embodiments of the present application, which will be described in more detail later. These drawings and text descriptions are not intended to limit the scope of the present application in any way, but to illustrate the concept of the present application to those skilled in the art by referring to specific embodiments. DETAILED DESCRIPTION
[0059] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.
[0060] In the technical solution of this application, the collection, storage, use, processing, transmission, provision and disclosure of information such as financial data or user data involved shall comply with the provisions of relevant laws and regulations and shall not violate public order and good morals.
[0061] It should be noted that in the embodiments of the present application, certain software, components, models and other existing solutions in the industry may be mentioned, and they should be regarded as exemplary. Their purpose is only to illustrate the feasibility of implementing the technical solution of the present application, but it does not mean that the applicant has or will necessarily use the solution.
[0062] Figure 1 This is the application scenario diagram of this application. Figure 1 As shown, the application scenario includes: a front-end to be pushed and a server. The front-end to be pushed is a user end, which can be a mobile phone, tablet, computer and other electronic devices, and the server can be a server, which is used to push financial content, such as financial products, to the front-end to be pushed.
[0063] In the prior art, financial products are pushed to users mainly through historical operation data of users, for example, through historical consumption records of financial products by users, to achieve the purpose of recommending similar financial products to users. However, the prior art method of analyzing historical data of users cannot accurately control the current preferences of users, resulting in low accuracy of recommended content, thus affecting user experience and satisfaction.
[0064] In response to the above technical problems, this application proposes the following technical ideas: by using a deep multimodal learning model to analyze the historical data of the front end to be pushed to obtain user preference information, by collecting the real-time operation data and real-time traffic data of the front end to be pushed, dynamically update the user preference information, by using a reinforcement learning model to analyze the user's preference information to output the financial products to be pushed, and push the financial products to be pushed to the front end to be pushed for display. This application achieves the purpose of dynamically updating user preferences by collecting real-time operation data and traffic data of users, so that the launched financial products are more in line with the real-time preferences of users and improve the correctness of the pushed financial products.
[0065] The technical solution of the present application and how the technical solution of the present application solves the above-mentioned technical problems are described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be repeated in some embodiments. The embodiments of the present application will be described below in conjunction with the accompanying drawings.
[0066] based on Figure 1 In the application scenario shown, the embodiment of the present application also provides a financial content push method. Figure 2 The process of the financial content push method provided in the embodiment of the present application Figure 1 .like Figure 2 As shown, the financial content push method includes:
[0067] S201. Acquire historical data of a front-end to be pushed, wherein the front-end to be pushed belongs to a user who has been authorized to collect data, and the historical data includes historical operation data and historical traffic data.
[0068] The execution subject of the method of this embodiment may be Figure 2 The server shown in .
[0069] In this embodiment, the front end to be pushed can record the user's historical operation data and historical traffic data through the log system, where the historical operation data can be browsing, clicking, consumption and other operation data, and the historical traffic data represents the source information of the traffic, which can be search keywords, entry clicks, social media checks, etc.
[0070] S202: Input the historical operation data and historical traffic data of the front end to be pushed into a deep multimodal learning model to obtain user preference information.
[0071] Combined with scenario examples, the deep multimodal learning model can process different types of data at the same time. For example, the deep multimodal learning model can adopt the multi-input model architecture in the deep neural network, input the historical operation data and the historical traffic data into a separate input branch to enter the deep multimodal learning model, and fuse and analyze them within the deep multimodal learning model to obtain the user's preference information, where the user's preference information represents the financial products that the user is interested in.
[0072] S203: extracting the real-time operation data and real-time traffic data to be pushed to the front end through a preset real-time data processing framework.
[0073] In combination with the scenario example, the real-time data processing framework can be selected as Apache Kafka or Apache Flink. The real-time data processing framework can capture the real-time operation data and real-time traffic data of users on the front end to be pushed in real time.
[0074] S204: Adjust the user preference information according to the real-time operation data and the real-time traffic data to obtain adjusted user preference information.
[0075] Combined with the scenario example, the user preference information obtained by the deep multimodal learning model based on historical data can be used as the initial user preference information. The initial user preference information can be updated using the user's real-time operation data and real-time traffic data collected by the real-time processing framework to dynamically adjust the user preference information.
[0076] S205: Input the adjusted user preference information into a reinforcement learning model to output financial products to be pushed.
[0077] Combined with scenario examples, a reinforcement learning model is designed to continuously learn the optimal content recommendation strategy. The preference information adjusted in real time can be input into the reinforcement learning model, and the reinforcement learning model can obtain the financial products to be pushed based on the recommendation strategy of the preference information.
[0078] S206: Push the financial product to be pushed to the front end to be pushed, so that the front end to be pushed displays the financial product to be pushed.
[0079] In combination with the scenario example, the obtained financial products to be pushed are pushed to the front end to be pushed, so that the front end to be pushed displays the obtained financial products to be pushed on the display interface, so that the user can operate the products to be pushed.
[0080] For example, the user preference information obtained by the deep multimodal learning model based on historical data can be used as the initial user preference information, and the initial user preference information can be input into the reinforcement learning model to obtain the initial products to be pushed, and the initial products to be pushed can be pushed to the front end to be pushed, and the user can perform real-time operations on the initial products to be pushed. For example, the initial push products include: product A, product B and product C, so that products A, product B and product C are displayed on the interface of the front end to be pushed. The real-time data processing framework captures the user's real-time operations on products A, product B and product C in real time, and adjusts the initial user preference information in real time based on the user's real-time operations to obtain the adjusted user preference information, and inputs the adjusted user preference information into the reinforcement learning model, so that the reinforcement learning model re-obtains the adjusted products to be pushed according to the adjusted user preference information, such as updating products A, product B and product C to products D, product E and product F, and pushing the updated products D, product E and product F to the front end to be pushed, so that products D, product E and product F are displayed on the interface of the front end to be pushed.
[0081] This example achieves the purpose of dynamically updating user preferences by collecting real-time user operation data and traffic data, so that the launched financial products are more in line with the user's real-time preferences and the accuracy of the pushed financial products is improved.
[0082] Optional, Figure 3 The process of the financial content push method provided in the embodiment of the present application Figure 2 .like Figure 3 As shown in Figure 1, the training process of the deep multimodal learning model includes:
[0083] S301. Receive model parameters sent by multiple front ends, wherein the model parameters are model parameters of a local model obtained by each front end training an initial global model according to local historical data.
[0084] Combined with the scenario example, a federated learning framework can be used to allow data to be processed locally on the front end before being uploaded to the server. Specifically, the federated learning framework has a global model. At the beginning of training, it can be an initial global model. There is a local model on the user's front-end device. For each front end, the local data of the front end can be used to train the initial global model with the local model to obtain the model parameters of the local model on each front end. Each front end uploads the model parameters of the local model obtained to the server, and the server receives the model parameters sent by each front end.
[0085] S302: Aggregate each model parameter using an aggregation algorithm to generate a new global model.
[0086] In combination with the scenario example, after S301, the server aggregates the received model parameters to update the initial global model to obtain an updated new global model.
[0087] S303: Send the generated new global model to each front end, so that each front end trains the new global model according to local historical data to update the model parameters of the local model.
[0088] Combined with the scenario example, after obtaining the new global model, execute the previous S301 again. For each front end, use the local data of the front end to train the new global model with the local model to update the model parameters of the local model on each front end. Each front end sends the model parameters of the updated local model to the server, and the server receives the model parameters sent by each front end.
[0089] S304, looping through the steps from receiving model parameters sent by multiple front ends to sending the generated new global model to each front end until the global model converges or reaches a preset number of cycles to obtain a trained deep multimodal learning model.
[0090] In combination with the scenario example, after executing S301 again, execute S302 again. Specifically, aggregate the received model parameters to obtain a new global model again. Continue to loop S301 and S302 until the global model reaches convergence and ends the loop. Alternatively, the maximum number of repetitions can be preset in advance, and the loop ends after the number of repeated executions reaches the preset maximum number of repetitions. At this time, the global model finally obtained in sequence is put into use as a deep multimodal learning model deployed on the server.
[0091] This example uses the federated learning framework to make the data uploaded to the global model the model parameters of the local model rather than the local data of the front end, which can protect the privacy of local data and improve the security of local data on the front end.
[0092] Optional, Figure 4 The process of the financial content push method provided in the embodiment of the present application Figure 3 .like Figure 4 As shown, the method also includes:
[0093] S401: Obtain recommendation logic information corresponding to the financial product to be pushed.
[0094] In combination with the scenario example, the server can integrate an interpretable model to analyze the products to be pushed to obtain the corresponding recommendation logic, such as Local Interpretable Model-agnostic Explanations (LIME) technology, decision tree, or Shapley additive interpretation to calculate the Shapley value of each feature to obtain the recommendation logic information of the financial products to be pushed.
[0095] S402: Push the recommendation logic information corresponding to the financial product to be pushed to the front end to be pushed, so that the front end to be pushed displays the recommendation logic information at a corresponding position of the financial product to be pushed.
[0096] Combined with the scenario example, after obtaining the recommendation logic information corresponding to each financial product to be pushed, the recommendation logic information is pushed to the front end, and the front end can display the corresponding recommendation logic information in the corresponding position of the product to be pushed in a visual or textual form. The recommendation logic can be the proportion of characteristic information such as keywords, topics or tags carried by the recommended financial products to be pushed. This example can increase users' trust and satisfaction with the pushed financial products by providing the recommendation logic information corresponding to the financial products to be pushed.
[0097] Figure 5 The process of the financial content push method provided in the embodiment of the present application Figure 4 .like Figure 5 As shown, the method also includes:
[0098] S501: Receive feedback sent by the front end to be pushed, wherein the feedback is generated by the front end to be pushed in response to a user's feedback operation on the recommendation logic information.
[0099] Combined with scenario examples, users can provide feedback on recommendation logic information. For example, if the keywords carried are incorrect, or the topic is incorrect, etc., users can input the feedback through the front end and transmit it to the server through the front end. The server monitors the feedback transmitted by the front end in real time.
[0100] S502: Update the recommendation logic information according to the feedback.
[0101] Combined with the scenario example, the server modifies the recommendation logic information in real time according to the received feedback, and after the modification, resends the modified recommendation logic information to the front end. For example, if product A is one of the financial products to be pushed, the recommendation logic for product A indicates that the keyword carried by product A is "keyword 1", but product A does not actually carry "keyword 1", but carries "keyword 2". At this time, the user can point out in the feedback that the keyword carried by product A is "keyword 2". After receiving the feedback, the server analyzes it again through the interpretable model. If the analysis result indicates that the keyword carried by product A is "keyword 2", it is the correct result, then the extra keywords carried in the recommendation logic information about product A can be modified to: "keyword 2", and the modified recommendation logic information can be resent to the front end, so that the front end can update the recommendation logic information displayed about product A.
[0102] In this example, by receiving user feedback and making real-time adjustments to the recommendation logic information, the accuracy of the displayed recommendation logic information can be improved.
[0103] Optional, Figure 6 The process of the financial content push method provided in the embodiment of the present application Figure 5 .like Figure 6 As shown, the method also includes:
[0104] S601: Receive at least one prediction request sent by a to-be-pushed front end, wherein the prediction request is generated by the to-be-pushed front end in response to a prediction task issued by a user.
[0105] Combined with the scenario example, the prediction request can be a click-through rate prediction, a conversion rate prediction, a user satisfaction prediction, etc. The prediction request can be generated according to the actual needs of the user, mainly through the specific content of the prediction task issued by the user on the front end.
[0106] S602: Respond to each prediction request to collect the historical data to be pushed to the front end.
[0107] Combined with the scenario example, for each prediction request, the corresponding historical data can be collected from the front end. For example, if the prediction request is a click-through rate request, since the historical data includes the user's historical behavior data, including historical click data, the user's historical click data can be collected from the front end. Similarly, if the prediction request is a click-through rate request, the user's jump data between financial products can be collected from the front end.
[0108] S603: Input the collected historical data into the multi-task learning model to output the prediction result corresponding to each prediction request.
[0109] Combined with the scenario example, the multi-task learning model can process multiple prediction requests at the same time. All received prediction requests can be input into the multi-task learning model at the same time. The multi-task learning model has a shared underlying representation for multiple prediction requests, which is used to capture the common features between all prediction requests, and a top layer specific to each prediction request, which is used to make predictions for each prediction request, thereby obtaining the prediction results of each prediction request respectively, and outputting the prediction results of each prediction request.
[0110] S604: Send the prediction result corresponding to each prediction request to the front end to be pushed, so that the front end to be pushed displays the prediction result corresponding to each prediction request.
[0111] Combined with the scenario example, the prediction results of each prediction request output by the multi-task learning model are sent to the front-end to be pushed, so that the front-end to be pushed displays the prediction results. In this example, multiple prediction requests can be predicted simultaneously through the multi-task learning model, which can improve the overall prediction efficiency of the prediction request.
[0112] Optional, Figure 7 The process of the financial content push method provided in the embodiment of the present application Figure 6 .like Figure 7 As shown, the training process of the multi-task learning model includes:
[0113] S701. Determine at least one prediction request.
[0114] Combined with the scenario example, the purpose of the multi-task prediction model is to predict multiple prediction requests at the same time, so when training the task prediction model, it is necessary to complete the training of the multi-task model through multiple prediction requests.
[0115] S702: Obtain a data set corresponding to each prediction request, where the data set includes prediction content and actual results.
[0116] Combined with the scenario example, the prediction request used to train the task prediction model is a prediction request defined in advance. For example, the multi-task prediction model finally put into use is used to predict 10 prediction requests. The prediction request used to train the multi-task prediction model is the data set corresponding to the 10 prediction requests, so as to ensure that the multi-task prediction model after training can predict the predicted prediction request. The data set of each prediction request includes the prediction content and actual results corresponding to the prediction request, which can be the prediction content in text form.
[0117] S703: Setting an initial weight corresponding to each prediction request in the multi-task learning model.
[0118] Combined with the scenario example, at the beginning of training the multi-task learning model, the initial weight of each prediction request can be preset in advance. For example, the initial weights of the preset 10 prediction requests can be set to X1, X2, X3...X10 respectively.
[0119] S704: Input the data set corresponding to each prediction request into the multi-task learning model, so that the multi-task learning model outputs the prediction result corresponding to each prediction request.
[0120] Combined with the scenario example, the prediction content in the data set of each prediction request is input into the multi-task learning model, and the multi-task learning model is jointly trained so that the multi-task learning model outputs the first round of prediction results for each prediction request.
[0121] S705: Compare the predicted result with the actual result to obtain the loss value corresponding to each prediction request.
[0122] Combined with the scenario example, the first round prediction results of each prediction request output by the multi-task learning model are compared with the actual results in the dataset, and the loss value between the first round prediction results and the actual results is calculated through the loss function.
[0123] S706: If the loss value of at least one prediction request does not meet the preset range, adjust the weight corresponding to each prediction request.
[0124] Combined with the scenario example, a maximum threshold of the loss value can be preset, and the loss value corresponding to each prediction request can be compared with the preset maximum threshold. If the loss value of at least one prediction request exceeds the maximum threshold, the weight corresponding to each prediction request is adjusted to achieve performance optimization on the specific prediction request.
[0125] S707. Repeat the steps of inputting the data set corresponding to each prediction request into the multi-task learning model and adjusting the weight corresponding to each prediction request until the loss value between each prediction result output by the multi-task learning model and the actual result meets the preset range, and then stop training the multi-task learning model.
[0126] In combination with the scenario example, the prediction content in the data set of each prediction request is re-input into the multi-task learning model that has adjusted the corresponding weight of each prediction request, and the multi-task learning model is iteratively trained until the loss value between the prediction result of each prediction request output by the multi-task learning model and the standard result does not exceed the preset threshold value, then it is determined that the multi-task learning model has been trained. Or when the number of training iterations reaches the preset maximum number, stop training the task learning model. In this example, the multi-task learning model can be jointly trained with the data set of multiple prediction requests to obtain a multi-task learning model that can simultaneously predict multiple prediction requests.
[0127] This embodiment achieves the purpose of dynamically updating user preferences by collecting real-time user operation data and traffic data, thereby making the launched financial products more in line with the user's real-time preferences and improving the accuracy of the pushed financial products.
[0128] Figure 8 A schematic diagram of the structure of the financial content push device provided in the embodiment of the present application, such as Figure 8 As shown, the financial content push device includes: an acquisition module 81, an input module 82, an extraction module 83, an adjustment module 84 and a push module 85, wherein:
[0129] An acquisition module 81 is used to acquire historical data of a front-end to be pushed, wherein the front-end to be pushed belongs to a user who has been authorized to collect data, and the historical data includes historical operation data and historical traffic data;
[0130] An input module 82, used to input the historical operation data and historical traffic data of the front end to be pushed into the deep multimodal learning model to obtain user preference information;
[0131] An extraction module 83 is used to extract the real-time operation data and real-time traffic data of the front end to be pushed through a preset real-time data processing framework;
[0132] An adjustment module 84, configured to adjust the user preference information according to the real-time operation data and the real-time traffic data to obtain adjusted user preference information;
[0133] The input module 82 is further used to input the adjusted user preference information into the reinforcement learning model to output the financial products to be pushed;
[0134] The push module 85 is used to push the financial product to be pushed to the front end to be pushed, so that the front end to be pushed displays the financial product to be pushed.
[0135] Optionally, the financial content push device includes: a training module 86;
[0136] A training module 86, configured to receive model parameters sent by a plurality of front ends, wherein the model parameters are model parameters of a local model obtained by each front end training an initial global model according to local historical data;
[0137] The training module 86 is also used to aggregate the model parameters using an aggregation algorithm to generate a new global model;
[0138] The training module 86 is further used to send the generated new global model to each front end, so that each front end trains the new global model according to the local historical data to update the model parameters of the local model;
[0139] The training module 86 is also used to cyclically execute the steps from receiving model parameters sent by multiple front ends to sending the generated new global model to each front end until the global model converges or reaches a preset number of cycles to obtain a trained deep multimodal learning model.
[0140] Optionally, an acquisition module 81 is specifically used to acquire recommendation logic information corresponding to the financial product to be pushed;
[0141] The acquisition module 82 is further specifically used to push the recommendation logic information corresponding to the financial product to be pushed to the front end to be pushed, so that the front end to be pushed displays the recommendation logic information at the corresponding position of the financial product to be pushed.
[0142] Optionally, a push module 85 is specifically configured to receive feedback sent by the front end to be pushed, wherein the feedback is generated by the front end to be pushed in response to a user's feedback operation on the recommendation logic information;
[0143] The push module 85 is further configured to update the recommendation logic information according to the feedback.
[0144] Optionally, the training module 86 is further configured to receive at least one prediction request sent by the front end to be pushed, wherein the prediction request is generated by the front end to be pushed in response to the prediction task issued by the user;
[0145] The training module 86 is further used to respond to each prediction request to collect the historical data to be pushed to the front end;
[0146] The training module 86 is further used to input the collected historical data into the multi-task learning model to output a prediction result corresponding to each prediction request;
[0147] The training module 86 is further used to send the prediction result corresponding to each prediction request to the front end to be pushed, so that the front end to be pushed displays the prediction result corresponding to each prediction request.
[0148] Optionally, the training module 86 is further configured to determine at least one prediction request;
[0149] The training module 86 is further used to obtain a data set corresponding to each prediction request, wherein the data set includes prediction content and actual results;
[0150] The training module 86 is also used to set the initial weight corresponding to each prediction request in the multi-task learning model;
[0151] The training module 86 is further used to input the data set corresponding to each prediction request into the multi-task learning model, so that the multi-task learning model outputs the prediction result corresponding to each prediction request;
[0152] The training module 86 is also used to compare the predicted result with the actual result to obtain the loss value corresponding to each prediction request;
[0153] The training module 86 is further configured to adjust the weight corresponding to each prediction request if the loss value of at least one prediction request does not meet the preset range;
[0154] The training module 86 is also used to repeatedly execute the steps from inputting the data set corresponding to each prediction request into the multi-task learning model to adjusting the weight corresponding to each prediction request, until the loss value between each prediction result output by the multi-task learning model and the actual result meets the preset range, and then stop training the multi-task learning model.
[0155] The financial content push device provided in the embodiment of the present application can be used to execute the technical solution of the financial content push method in the above embodiment. Its implementation principle and technical effect are similar and will not be repeated here.
[0156] It should be noted that it should be understood that the division of the various modules of the above device is only a division of logical functions. In actual implementation, they can be fully or partially integrated into one physical entity, or they can be physically separated. And these modules can all be implemented in the form of software called by processing elements; they can also be all implemented in the form of hardware; some modules can also be implemented in the form of software called by processing elements, and some modules can be implemented in the form of hardware. For example, the training module 86 can be a separately established processing element, or it can be integrated in a chip of the above device. In addition, it can also be stored in the memory of the above device in the form of program code, and called and executed by a processing element of the above device. The implementation of other modules is similar. In addition, these modules can be fully or partially integrated together, or they can be implemented independently. The processing element here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each module above can be completed by an integrated logic circuit of hardware in the processor element or instructions in the form of software.
[0157] Fig. 9 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Fig. 9 As shown, the electronic device may include: a transceiver 121 , a processor 122 , and a memory 123 .
[0158] The processor 122 executes the computer execution instructions stored in the memory, so that the processor 122 executes the scheme in the above embodiment. The processor 122 can be a general-purpose processor, including a central processing unit CPU, a network processor (network processor, NP), etc.; it can also be a digital signal processor DSP, an application-specific integrated circuit ASIC, a field programmable gate array FPGA or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components.
[0159] The memory 123 is connected to the processor 122 via a system bus and completes communication between them. The memory 123 is used to store computer program instructions.
[0160] The transceiver 121 may be used to obtain tasks to be executed and configuration information of the tasks to be executed.
[0161] The system bus can be a peripheral component interconnect (PCI) bus or an extended industry standard architecture (EISA) bus, etc. The system bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus. The transceiver is used to realize the communication between the database access device and other computers (such as clients, read-write libraries, and read-only libraries). The memory may include random access memory (RAM) and may also include non-volatile memory.
[0162] The electronic device provided in the embodiment of the present application may be the terminal device of the above embodiment.
[0163] An embodiment of the present application also provides a chip for running instructions, which is used to execute the technical solution of the financial content push method in the above embodiment.
[0164] An embodiment of the present application also provides a computer-readable storage medium, in which computer instructions are stored. When the computer instructions are executed on a computer, the computer executes the technical solution of the financial content push method of the above embodiment.
[0165] An embodiment of the present application also provides a computer program product, which includes a computer program stored in a computer-readable storage medium. At least one processor can read the computer program from the computer-readable storage medium. When at least one processor executes the computer program, it can implement the technical solution of the financial content push method in the above embodiment.
[0166] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of modules is only a logical function division, and there may be other division methods in actual implementation, such as multiple modules can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or modules, which can be electrical, mechanical or other forms.
[0167] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to implement the solution of this embodiment.
[0168] In addition, each functional module in each embodiment of the present application can be integrated into one processing unit, or each module can exist physically separately, or two or more modules can be integrated into one unit. The above-mentioned module-composed unit can be implemented in the form of hardware or in the form of hardware plus software functional units.
[0169] The above-mentioned integrated module implemented in the form of a software function module can be stored in a computer-readable storage medium. The above-mentioned software function module is stored in a storage medium, including a number of instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) or a processor to perform some steps of the methods of various embodiments of the present application.
[0170] It should be understood that the processor may be a central processing unit (CPU), or other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), etc. A general-purpose processor may be a microprocessor or any conventional processor. The steps of the method disclosed in the invention may be directly implemented as being executed by a hardware processor, or may be implemented by a combination of hardware and software modules in the processor.
[0171] The memory may include a high-speed RAM memory, and may also include a non-volatile storage NVM, such as at least one disk memory, and may also be a USB flash drive, a mobile hard disk, a read-only memory, a magnetic disk or an optical disk, etc.
[0172] The bus may be an Industry Standard Architecture (ISA) bus, a Peripheral Component Interconnect (PCI) bus, or an Extended Industry Standard Architecture (EISA) bus, etc. The bus may be divided into an address bus, a data bus, a control bus, etc. For ease of representation, the bus in the drawings of the present application is not limited to only one bus or one type of bus.
[0173] The above storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk. The storage medium can be any available medium that can be accessed by a general or special purpose computer.
[0174] An exemplary storage medium is coupled to a processor so that the processor can read information from the storage medium and write information to the storage medium. Of course, the storage medium can also be a component of the processor. The processor and the storage medium can be located in an application specific integrated circuit (ASIC). Of course, the processor and the storage medium can also exist as discrete components in an electronic control unit or a main control device.
[0175] Those skilled in the art can understand that all or part of the steps of implementing the above-mentioned method embodiments can be completed by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, the steps of the above-mentioned method embodiments are executed; and the aforementioned storage medium includes: ROM, RAM, disk or optical disk and other media that can store program codes.
[0176] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present application.
Claims
1. A method for pushing financial content, characterized in that: Applied to the server, the method includes: Acquire historical data of the front-end to be pushed, wherein the front-end to be pushed belongs to a user who has been authorized to collect data, and the historical data includes historical operation data and historical traffic data; Inputting the historical operation data and historical traffic data of the front end to be pushed into a deep multimodal learning model to obtain user preference information; Extract the real-time operation data and real-time traffic data of the front end to be pushed through a preset real-time data processing framework; Adjusting the user preference information according to the real-time operation data and the real-time traffic data to obtain adjusted user preference information; Inputting the adjusted user preference information into a reinforcement learning model to output the financial products to be pushed; The financial product to be pushed is pushed to the front end to be pushed, so that the front end to be pushed displays the financial product to be pushed.
2. The method according to claim 1, characterized in that The training process of the deep multimodal learning model includes: Receiving model parameters sent by multiple front ends, wherein the model parameters are model parameters of local models obtained by each front end training an initial global model according to local historical data; Aggregation algorithms are used to aggregate the parameters of each model to generate a new global model; Sending the generated new global model to each front end so that each front end trains the new global model according to local historical data to update the model parameters of the local model; The steps from receiving model parameters sent by multiple front ends to sending the generated new global model to each front end are executed cyclically until the global model converges or reaches a preset number of cycles to obtain a trained deep multimodal learning model.
3. The method according to claim 1, characterized in that The method further comprises: Obtaining recommendation logic information corresponding to the financial product to be pushed; The recommendation logic information corresponding to the financial product to be pushed is pushed to the front end to be pushed, so that the front end to be pushed displays the recommendation logic information at the corresponding position of the financial product to be pushed.
4. The method according to claim 3, characterized in that The method further comprises: Receiving feedback sent by the front end to be pushed, wherein the feedback is generated by the front end to be pushed in response to a user's feedback operation on the recommendation logic information; The recommendation logic information is updated according to the feedback.
5. The method according to any one of claims 1 to 4, characterized in that: The method further comprises: Receiving at least one prediction request sent by the front end to be pushed, wherein the prediction request is generated by the front end to be pushed in response to the prediction task issued by the user; Responding to each prediction request to collect the historical data to be pushed to the front end; Input the collected historical data into the multi-task learning model to output the prediction result corresponding to each prediction request; The prediction result corresponding to each prediction request is sent to the front end to be pushed, so that the front end to be pushed displays the prediction result corresponding to each prediction request.
6. The method according to claim 5, characterized in that The training process of the multi-task learning model includes: determining at least one prediction request; Obtain a data set corresponding to each prediction request, wherein the data set includes prediction content and actual results; Set the initial weight for each prediction request in the multi-task learning model; Inputting the data set corresponding to each prediction request into the multi-task learning model so that the multi-task learning model outputs the prediction result corresponding to each prediction request; Compare the predicted results with the actual results to obtain the loss value corresponding to each prediction request; If the loss value of at least one prediction request does not meet the preset range, adjust the weight corresponding to each prediction request; Repeat the steps of inputting the data set corresponding to each prediction request into the multi-task learning model and adjusting the weight corresponding to each prediction request until the loss value between each prediction result output by the multi-task learning model and the actual result meets the preset range, then stop training the multi-task learning model.
7. A financial content push device, applied to a server, characterized in that: The device comprises: An acquisition module is used to acquire historical data of the front end to be pushed, wherein the front end to be pushed belongs to a user who has been authorized to collect data, and the historical data includes historical operation data and historical traffic data; An input module, used for inputting the historical operation data and historical traffic data of the front end to be pushed into the deep multimodal learning model to obtain user preference information; An extraction module, used to extract the real-time operation data and real-time traffic data of the front end to be pushed through a preset real-time data processing framework; An adjustment module, used for adjusting the user preference information according to the real-time operation data and the real-time traffic data to obtain adjusted user preference information; The input module is further used to input the adjusted user preference information into the reinforcement learning model to output the financial products to be pushed; The push module is used to push the financial product to be pushed to the front end to be pushed, so that the front end to be pushed displays the financial product to be pushed.
8. An electronic device, characterized in that: include: A processor, and a memory communicatively connected to the processor; The memory stores computer-executable instructions; The processor executes the computer-executable instructions stored in the memory to implement the method according to any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer-executable instructions, which are used to implement the method according to any one of claims 1 to 6 when executed by a processor.
10. A computer program product, characterized in that The invention comprises a computer program, which implements the method according to any one of claims 1 to 6 when being executed by a processor.