A data recommendation method and device, electronic equipment and computer readable medium

Abstract

The application discloses a data recommendation method and device, electronic equipment and a computer readable medium, relates to the technical field of big data analysis, and the method comprises the following steps: receiving a data recommendation request, obtaining a corresponding user identifier, obtaining a corresponding product identifier and user behavior data according to the user identifier; determining a corresponding keyword based on the product identifier, determining a corresponding first keyword feature weight matrix according to the keyword; determining a time window stage corresponding to the keyword according to the user behavior data, determining a corresponding second keyword feature weight matrix of each according to the time window stage and the first keyword feature weight matrix; determining a candidate keyword according to the first keyword feature weight matrix and the second keyword feature weight matrix of each; obtaining the weight corresponding to each candidate keyword, and then determining target recommended product data according to the candidate keyword and the weight and outputting. The safety of user financial management can be improved when recommending financial product data for the user.

Description

Technical Field

[0001] This application relates to the field of big data data analysis technology, and in particular to a data recommendation method, apparatus, electronic device and computer-readable medium. Background Technology

[0002] In scenarios where data is used to recommend wealth management products to users, the platform aims to provide safe and long-term wealth management and consulting services to new users. For users lacking investment experience, with high risk tolerance, and prone to aggressive trading, the platform seeks to accurately recommend products. However, the lack of investment experience among new users and the high risk tolerance of experienced users result in potentially poor investment security. Summary of the Invention

[0003] In view of this, embodiments of this application provide a data recommendation method, apparatus, electronic device, and computer-readable medium, which can solve the problem of poor security for both new and existing users when making financial decisions on financial platforms.

[0004] To achieve the above objectives, according to one aspect of an embodiment of this application, a data recommendation method is provided, comprising:

[0005] Receive data recommendation requests, obtain the corresponding user identifiers, and then obtain the corresponding product identifiers and user behavior data based on the user identifiers;

[0006] Based on the product identifier, the corresponding keywords are determined, and then the corresponding first keyword feature weight matrix is ​​determined based on the keywords.

[0007] Based on user behavior data, determine the time window stage corresponding to the keyword, and based on the time window stage and the first keyword feature weight matrix, determine the corresponding second keyword feature weight matrix.

[0008] Candidate keywords are determined based on the feature weight matrix of the first keyword and the feature weight matrices of each second keyword;

[0009] Obtain the weights corresponding to each candidate keyword, and then determine and output the target recommended product data based on the candidate keywords and weights.

[0010] Optionally, target recommended product data is determined and output, including:

[0011] Determine the corresponding candidate recommended product data based on the candidate keywords;

[0012] Based on the weights and candidate recommended product data, determine the sum of the keyword weights corresponding to the candidate recommended product data;

[0013] Based on the sum of keyword weights, determine and output the target recommended product data from the candidate recommended product data.

[0014] Optionally, the time window phase corresponding to the keyword can be determined based on user behavior data, including:

[0015] Based on user behavior data, determine the corresponding behavior type;

[0016] Determine the corresponding behavior weight based on the behavior type;

[0017] Obtain the number of occurrences of each behavior type, and then determine the corresponding behavior score based on the weight and the number of occurrences;

[0018] Based on the behavior score, determine the time window stage corresponding to the keyword.

[0019] Optionally, determine the corresponding feature weight matrices for each second keyword, including:

[0020] In response to the decay corresponding to the time window stage, the set of keyword feature weight matrices is invoked to calculate the similarity between the first keyword feature weight matrix and each keyword feature weight matrix in the set of keyword feature weight matrices;

[0021] Based on the similarity, determine the corresponding second keyword feature weight matrices in each keyword feature weight matrix that correspond to the first keyword feature weight matrix.

[0022] Optionally, candidate keywords are determined, including:

[0023] The first keyword feature weight matrix and each second keyword feature weight matrix are input into the recommendation model to output the corresponding candidate keywords.

[0024] Optionally, output the corresponding candidate keywords, including:

[0025] The product association processing module in the recommendation model is invoked to calculate the similarity between the first keyword feature weight matrix and each second keyword feature weight matrix based on the similarity coefficient algorithm.

[0026] The keyword interest migration event processing module in the recommendation model is called to obtain the current keyword performance vector group of user interests in adjacent time windows, and then output short-term migration status data and short-term migration behavior data.

[0027] Based on short-term migration status data and short-term migration behavior data, generate an interest migration event matrix of product keywords corresponding to adjacent time windows and user identifiers;

[0028] Candidate keywords are determined based on similarity and the interest migration event matrix of product keywords.

[0029] Optionally, candidate keywords are determined based on similarity and the interest migration event matrix of product keywords, including:

[0030] Based on similarity, determine the target second keyword feature weight matrix in each second keyword feature weight matrix;

[0031] Based on the feature weight matrix of the target second keyword, determine the corresponding keyword set;

[0032] Based on the interest migration event matrix of product keywords, each keyword in the keyword set is filtered to obtain candidate keywords.

[0033] In addition, this application also provides a data recommendation device, including:

[0034] The receiving unit is configured to receive data recommendation requests, obtain the corresponding user identifier, and then obtain the corresponding product identifier and user behavior data based on the user identifier.

[0035] The first keyword feature weight matrix determination unit is configured to determine the corresponding keyword based on the product identifier, and then determine the corresponding first keyword feature weight matrix based on the keyword.

[0036] The second keyword feature weight matrix determination unit is configured to determine the time window stage corresponding to the keyword based on user behavior data, and to determine the corresponding second keyword feature weight matrices based on the time window stage and the first keyword feature weight matrix.

[0037] The candidate keyword determination unit is configured to determine candidate keywords based on the first keyword feature weight matrix and each second keyword feature weight matrix;

[0038] The data recommendation unit is configured to obtain the weights corresponding to each candidate keyword, and then determine and output the target recommended product data based on the candidate keywords and weights.

[0039] Optionally, the data recommendation unit is further configured to:

[0040] Determine the corresponding candidate recommended product data based on the candidate keywords;

[0041] Based on the weights and candidate recommended product data, determine the sum of the keyword weights corresponding to the candidate recommended product data;

[0042] Based on the sum of keyword weights, determine and output the target recommended product data from the candidate recommended product data.

[0043] Optionally, the second keyword feature weight matrix determination unit is further configured as follows:

[0044] Based on user behavior data, determine the corresponding behavior type;

[0045] Determine the corresponding behavior weight based on the behavior type;

[0046] Obtain the number of occurrences of each behavior type, and then determine the corresponding behavior score based on the weight and the number of occurrences;

[0047] Based on the behavior score, determine the time window stage corresponding to the keyword.

[0048] Optionally, the second keyword feature weight matrix determination unit is further configured as follows:

[0049] In response to the decay corresponding to the time window stage, the set of keyword feature weight matrices is invoked to calculate the similarity between the first keyword feature weight matrix and each keyword feature weight matrix in the set of keyword feature weight matrices;

[0050] Based on the similarity, determine the corresponding second keyword feature weight matrices in each keyword feature weight matrix that correspond to the first keyword feature weight matrix.

[0051] Optionally, the candidate keyword determination unit is further configured to:

[0052] The first keyword feature weight matrix and each second keyword feature weight matrix are input into the recommendation model to output the corresponding candidate keywords.

[0053] Optionally, the candidate keyword determination unit is further configured to:

[0054] The product association processing module in the recommendation model is invoked to calculate the similarity between the first keyword feature weight matrix and each second keyword feature weight matrix based on the similarity coefficient algorithm.

[0055] The keyword interest migration event processing module in the recommendation model is called to obtain the current keyword performance vector group of user interests in adjacent time windows, and then output short-term migration status data and short-term migration behavior data.

[0056] Based on short-term migration status data and short-term migration behavior data, generate an interest migration event matrix of product keywords corresponding to adjacent time windows and user identifiers;

[0057] Candidate keywords are determined based on similarity and the interest migration event matrix of product keywords.

[0058] Optionally, the candidate keyword determination unit is further configured to:

[0059] Based on similarity, determine the target second keyword feature weight matrix in each second keyword feature weight matrix;

[0060] Based on the feature weight matrix of the target second keyword, determine the corresponding keyword set;

[0061] Based on the interest migration event matrix of product keywords, each keyword in the keyword set is filtered to obtain candidate keywords.

[0062] In addition, this application also provides a data recommendation electronic device, including: one or more processors; and a storage device for storing one or more programs, which, when executed by one or more processors, cause the one or more processors to implement the data recommendation method as described above.

[0063] In addition, this application also provides a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the data recommendation method as described above.

[0064] To achieve the above objectives, according to another aspect of the embodiments of this application, a computer program product is provided.

[0065] A computer program product according to an embodiment of this application includes a computer program that, when executed by a processor, implements the data recommendation method provided in an embodiment of this application.

[0066] One embodiment of the above invention has the following advantages or beneficial effects: This application receives a data recommendation request, obtains the corresponding user identifier, and then obtains the corresponding product identifier and user behavior data based on the user identifier; based on the product identifier, it determines the corresponding keywords, and then determines the corresponding first keyword feature weight matrix based on the keywords; it determines the time window stage corresponding to the keywords based on the user behavior data, and determines the corresponding second keyword feature weight matrices based on the time window stage and the first keyword feature weight matrix; it determines candidate keywords based on the first keyword feature weight matrix and the second keyword feature weight matrices; it obtains the weights corresponding to each candidate keyword, and then determines and outputs the target recommended product data based on the candidate keywords and their weights. This can improve the security of user financial management when recommending financial product data to users, and can achieve more effective data recommendation or incentives, promoting platform conversion effectiveness.

[0067] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0068] The accompanying drawings are provided to better understand this application and do not constitute an undue limitation thereof. Wherein:

[0069] Figure 1 This is a schematic diagram of the main flow of a data recommendation method according to an embodiment of this application;

[0070] Figure 2This is a schematic diagram of the main flow of a data recommendation method according to an embodiment of this application;

[0071] Figure 3 This is a schematic diagram of the main flow of a data recommendation method according to an embodiment of this application;

[0072] Figure 4 This is a schematic diagram of the main units of a data recommendation device according to an embodiment of this application;

[0073] Figure 5 This is an exemplary system architecture diagram to which embodiments of this application can be applied;

[0074] Figure 6 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers in the embodiments of this application. Detailed Implementation

[0075] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of this application, including various details to aid understanding. These embodiments should be considered merely exemplary. Therefore, those skilled in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this application. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description. It should be noted that the collection, analysis, use, transmission, and storage of user personal information involved in the technical solutions of this application comply with relevant laws and regulations, are used for legitimate and reasonable purposes, are not shared, disclosed, or sold outside of these legitimate uses, and are subject to supervision and management by regulatory authorities. Necessary measures should be taken to prevent unauthorized access to such personal information data, ensure that personnel authorized to access personal information data comply with relevant laws and regulations, and ensure the security of user personal information. Once this user personal information data is no longer needed, the risk should be minimized by restricting or even prohibiting data collection and / or deleting the data.

[0076] When used, including in certain relevant applications, data is deidentified to protect user privacy, for example by removing specific identifiers, controlling the amount or specificity of stored data, controlling how data is stored, and / or other methods.

[0077] Figure 1 This is a schematic diagram of the main flow of a data recommendation method according to an embodiment of this application, as shown below. Figure 1 As shown, data recommendation methods include:

[0078] Step S101: Receive a data recommendation request, obtain the corresponding user identifier, and then obtain the corresponding product identifier and user behavior data based on the user identifier.

[0079] In this embodiment, the executing entity of the data recommendation method (e.g., a server) can receive data recommendation requests via wired or wireless connections. Specifically, the data recommendation request may be a request to recommend financial products or related benefits. Benefits can be represented as a specific form of discount or discount equivalent (service). A user's preference for discounts or discount equivalents constitutes their benefit interest. Upon receiving the data recommendation request, the executing entity can obtain the corresponding user identifier. The user identifier can be used to represent the name or number of the user accessing the financial platform; however, this embodiment does not specifically limit the user identifier.

[0080] The executing entity can obtain corresponding product identifiers and user behavior data based on the user identifier. The product identifier corresponding to the user identifier could be, for example, the financial product number under the username corresponding to the user identifier. The executing entity can also obtain corresponding user behavior data on the platform based on the user identifier. Specifically, user behavior data may include user's collection data of financial products, browsing time of financial products, click data of financial products, forwarding data of financial products, etc. This application embodiment does not specifically limit the content of user behavior data.

[0081] Step S102: Based on the product identifier, determine the corresponding keywords, and then determine the corresponding first keyword feature weight matrix based on the keywords.

[0082] After obtaining the product identifier corresponding to the user identifier, the executing entity can determine the keywords corresponding to the product identifier. For example, the keywords may be the product name or product type corresponding to the product identifier. This application embodiment does not specifically limit the keywords corresponding to the product identifier.

[0083] For example, the first keyword feature weight matrix could be a financial product keyword feature weight matrix PP. The financial product keyword feature weight matrix PP can be an N*20 numerical matrix. The feature weights corresponding to the keywords can be set by business personnel when listing financial products on the financial platform, based on the keywords corresponding to the financial products. The selectable values ​​are 1-20 points. When generating the feature weight matrix, the overall weight of each row is normalized, i.e., the weight of each keyword = the keyword score / the sum of the scores of the keywords for the financial product, rounded to 4 decimal places. After normalization, the sum of the weights in each row is 1.

[0084] Step S103: Determine the time window stage corresponding to the keyword based on user behavior data, and determine the corresponding second keyword feature weight matrix based on the time window stage and the first keyword feature weight matrix.

[0085] Specifically, determining the time window stage corresponding to the keyword based on user behavior data includes: determining the corresponding behavior type based on user behavior data; determining the corresponding behavior weight based on the behavior type; obtaining the occurrence frequency of the behavior type, and then determining the corresponding behavior score based on the weight and occurrence frequency; and determining the time window stage corresponding to the keyword based on the behavior score.

[0086] For example, user behavior types may include browsing, placing orders, and forwarding. Each behavior type has a corresponding weight based on the strength of the user's purchase intention, with behaviors indicating strong purchase intention receiving higher weights. For example, browsing might have a weight of 0.1, placing an order might have a weight of 0.6, and forwarding might have a weight of 0.3. This application embodiment does not specifically limit the weight of each behavior and can adjust it according to actual circumstances. After determining the user's behavior type, the executing entity can obtain the number of times the behavior corresponding to that type occurs, and multiply the number of occurrences by the corresponding weight to obtain the behavior score. For example, if the behavior type is browsing, the corresponding weight is 0.1, and the browsing frequency is 3 times, then the corresponding behavior score could be 0.1 * 3 = 0.3. Furthermore, the obtained behavior root count can be determined as the time window stage corresponding to the keyword. The time window stage is used to characterize the probability that the user's interest in the browsed product will shift.

[0087] Specifically, determining the corresponding second keyword feature weight matrices includes: responding to the decay corresponding to the time window stage, calling the keyword feature weight matrix set to calculate the similarity between the first keyword feature weight matrix and each keyword feature weight matrix in the keyword feature weight matrix set; and determining each second keyword feature weight matrix in each keyword feature weight matrix that corresponds to the first keyword feature weight matrix based on the similarity.

[0088] For example, in the financial product relevance processing module, the input consists of N financial product keyword feature weight matrices, including a first keyword feature weight matrix. A similarity coefficient algorithm is used, for instance, based on the calculated similarity to each element in the first keyword feature weight matrix, to determine and output various second keyword feature weight matrices that are similar to the first keyword feature weight matrix.

[0089] Step S104: Determine candidate keywords based on the first keyword feature weight matrix and each second keyword feature weight matrix.

[0090] Specifically, determining candidate keywords includes inputting the first keyword feature weight matrix and each second keyword feature weight matrix into the recommendation model to output the corresponding candidate keywords.

[0091] The executing entity can obtain the keyword prediction weight vector of user product interest in the time window corresponding to user behavior data and input it into the recommendation model. The recommendation model then uses a convolution algorithm to calculate the sum of all keyword weights for each financial product and the sum of all keyword weights for each equity based on the keyword prediction weight vector of user product interest, the first keyword feature weight matrix, and each second keyword feature weight matrix. Based on the sum of all keyword weights for each financial product and the sum of all keyword weights for each equity, the candidate keywords are determined and output.

[0092] As one implementation of this application, outputting the corresponding candidate keywords includes: calling the product relevance processing module in the recommendation model, for example, the financial product relevance processing module, to calculate the similarity between the first keyword feature weight matrix and each of the second keyword feature weight matrices based on a similarity coefficient algorithm. For example, in the financial product relevance processing module, the same time window - product relevance matrix PC(t) represents the similarity between different financial products within a certain time window, and is the window value of the t-time window of the financial product similarity association weight matrix. For example, i and j represent two different financial products, then the product relevance matrix PC(t) = {PC(i,j)}. In the financial product relevance processing module, the financial product keyword feature weight matrices of N financial products are input, and the similarity coefficient algorithm is used to calculate and output the similarity between the first keyword feature weight matrix and each of the second keyword feature weight matrices.

[0093] The recommendation model's keyword interest migration event processing module, such as the user product interest-keyword interest migration event processing module, is invoked to obtain the current keyword performance vector group of user interests in adjacent time windows (e.g., time windows t-1 and t), thereby outputting short-term migration state data and short-term migration behavior data. Based on the short-term migration state data and short-term migration behavior data, an interest migration event matrix of product keywords corresponding to adjacent time windows and user identifiers is generated. Candidate keywords are determined based on similarity and the product keyword interest migration event matrix. For example, the user product interest-keyword interest migration event processing module includes a migration derivative result processing submodule and a migration event calculation submodule. Specifically, the migration derivative result processing submodule is input with the current keyword performance vector (frequency) group UP of user product interests in adjacent time windows t-1 and t, to output the current keyword-short-term migration state-derived result vector UTT and the current keyword-short-term migration behavior-derived result vector UTB. In the migration event calculation submodule, the output of the migration derivative result processing submodule is received as input, and an algorithmic decision is made to output the interest migration event matrix UPE(t-1,t,u) for user product keywords corresponding to time windows t-1 and t. Different event types correspond to different processing methods for the user relevance matrix. Candidate keywords are determined based on similarity and the interest migration event matrix of product keywords. Specifically, the interest migration keywords in the interest migration event matrix of product keywords corresponding to the keyword feature weight matrix when the similarity exceeds a preset similarity threshold can be determined as candidate keywords.

[0094] Specifically, candidate keywords are determined based on similarity and the interest migration event matrix of product keywords. This includes: determining the target second keyword feature weight matrix in each second keyword feature weight matrix based on similarity; for example, the second keyword feature weight matrix corresponding to a similarity greater than a preset similarity threshold can be determined as the target second keyword feature weight matrix. The corresponding keyword set is determined based on the target second keyword feature weight matrix; and each keyword in the keyword set is filtered based on the interest migration event matrix of product keywords to obtain candidate keywords.

[0095] For example, the interest migration event matrix of product keywords can store key-value pairs of the original keywords and interest migration keywords of the product. The executing entity can search these key-value pairs to see if there are keywords in the keyword set. Specifically, the interest migration keywords corresponding to the keywords in the existing keyword set can be identified as candidate keywords.

[0096] Step S105: Obtain the weights corresponding to each candidate keyword, and then determine and output the target recommended product data based on the candidate keywords and weights.

[0097] Specifically, we can identify the financial products corresponding to each candidate keyword, sum the weights of the candidate keywords corresponding to the same financial product to obtain the keyword weight sum. We then sort the keyword weight sums in descending order, output the top n keyword weight sums, and determine and output the recommended financial products and benefits corresponding to the top n keyword weight sums.

[0098] This embodiment receives data recommendation requests, obtains corresponding user identifiers, and then obtains corresponding product identifiers and user behavior data based on the user identifiers. Based on the product identifiers, it determines corresponding keywords and then determines a first keyword feature weight matrix. Based on the user behavior data, it determines the time window stage corresponding to the keywords, and based on the time window stage and the first keyword feature weight matrix, it determines corresponding second keyword feature weight matrices. Based on the first keyword feature weight matrix and the second keyword feature weight matrices, it determines candidate keywords. It obtains the weights corresponding to each candidate keyword, and then determines and outputs target recommended product data based on the candidate keywords and their weights. This improves the security of user financial management when recommending financial product data and enables more effective data recommendation or incentives, thus promoting platform conversion rates.

[0099] Figure 2 This is a schematic diagram of the main flow of a data recommendation method according to an embodiment of this application, such as... Figure 2 As shown, data recommendation methods include:

[0100] Step S201: Receive a data recommendation request, obtain the corresponding user identifier, and then obtain the corresponding product identifier and user behavior data based on the user identifier.

[0101] A data recommendation request can be a request to recommend a product or product benefits. Upon receiving a data recommendation request, the executing entity can obtain the corresponding user identifier. The user identifier can be, for example, a member identifier, a member level identifier, etc., but this embodiment does not specifically limit the user identifier. User behavior data can be user search data, browsing data, forwarding data, etc., and this embodiment does not specifically limit the user behavior data.

[0102] Step S202: Based on the product identifier, determine the corresponding keywords, and then determine the corresponding first keyword feature weight matrix based on the keywords.

[0103] The first keyword feature weight matrix could be, for example, a financial product keyword feature weight matrix PP. The financial product keyword feature weight matrix PP can be an N*20 numerical matrix. The feature weights corresponding to keywords can be set by business personnel when listing financial products on the financial platform, based on the keywords corresponding to the financial products. The selectable values ​​are 1-20 points. When generating the feature weight matrix, the overall weight of each row is normalized, i.e., the weight of each keyword = the keyword score / the sum of the scores of all keywords for the financial product, rounded to 4 decimal places. After normalization, the sum of the weights in each row is 1.

[0104] Step S203: Determine the time window stage corresponding to the keyword based on user behavior data, and determine the corresponding second keyword feature weight matrix based on the time window stage and the first keyword feature weight matrix.

[0105] For example, the first keyword feature weight matrix may include the recommended rights keyword feature weight matrix UPP corresponding to the time window stage and the first keyword feature weight matrix (i.e., the financial product keyword feature weight matrix PP). The recommended rights keyword feature weight matrix UPP can be an N*20 numerical matrix; the feature weights corresponding to the keywords can be set by business personnel when listing financial products and recommending rights on the rights platform, with selectable values ​​from 1 to 20 points; when generating the feature weight matrix, the overall weight of each row is normalized, i.e., the weight of each keyword = the sum of the keyword score and the keyword scores of the rights, rounded to 4 decimal places; after normalization, the sum of the weights of each row is 1; for simplicity, in this embodiment, a separate UPP matrix is ​​not established, but the UPP matrix is ​​directly replaced by the PP matrix, that is, financial products and recommended rights are equivalent, mapped one-to-one.

[0106] Specifically, the implementing entity can input the keyword feature weight matrices of N financial products into the financial product relevance processing module, including the first keyword feature weight matrix. Using a similarity coefficient algorithm, for example, based on the calculated similarity to each element in the first keyword feature weight matrix, the entity determines and outputs each second keyword feature weight matrix that is similar to the first keyword feature weight matrix.

[0107] Step S204: Determine candidate keywords based on the first keyword feature weight matrix and each second keyword feature weight matrix.

[0108] Step S205: Obtain the weight of each candidate keyword, and then determine the corresponding candidate recommended product data based on the candidate keywords.

[0109] For example, candidate recommended product data could be candidate wealth management products. Each candidate wealth management product can correspond to one or more candidate keywords. By matching the candidate keywords with the various wealth management products based on their similarity, the candidate wealth management products corresponding to each candidate keyword can be determined.

[0110] Step S206: Determine the sum of keyword weights corresponding to the candidate recommended product data based on the weights and candidate recommended product data.

[0111] Determine the weight corresponding to each candidate recommended product data, and sum the corresponding weights to obtain the sum of the keyword weights corresponding to each candidate recommended product.

[0112] Step S207: Based on the sum of keyword weights, determine and output the target recommended product data in the candidate recommended product data.

[0113] The sum of the weights of each keyword is sorted in descending order. The candidate recommended product data corresponding to the sum of the weights of the top n keywords are determined as the target recommended product data, which is then output to the user. This improves the security of users' financial management when recommending financial products and enables more effective data recommendation or incentives, thereby promoting platform conversion rates.

[0114] Figure 3 This is a schematic diagram illustrating an application scenario of a data recommendation method according to an embodiment of this application. The data recommendation method in this embodiment can be applied to scenarios involving the recommendation of financial products and related rights.

[0115] The relevant definitions involved in the data recommendation model (e.g., the periodic guided migration model) used in this application embodiment are as follows:

[0116] User interest migration cycle for financial products: The time it takes for the recurring pattern of interest shifts in financial product keywords to take place.

[0117] Migration Status: The user's interest migration status for keywords related to financial products. The short-term status set is: S(short) = {start, active, disappear}.

[0118] Migration behavior: User interest migration behavior of financial product keywords. The short-term migration behavior set is: B(short) = {product migration, activation, decline}.

[0119] Time window: The concept of a time window is used to measure the continuity of users' interest in financial products within a given window, as well as the probability of a shift in that interest. By using a time window, changes in users' financial product interests can be captured in a timely manner, allowing for feedback and updates to the user's existing financial product interest model. This, in turn, enables more accurate financial product and equity recommendations based on the user's individual circumstances.

[0120] Prediction accuracy: A concept that corresponds to a time window, indicating the precision with which changes in user interest in financial products are predicted within that time window. Decreasing accuracy indicates a shift in user interest in financial products.

[0121] Investment products: N groups of investment products.

[0122] Keywords: In this embodiment of the application, keyword design and keyword statistics are required; in this embodiment of the application, keywords for financial products and keywords for user interests are both related to keywords.

[0123] Keyword Matrix Comparison for Wealth Management Products: N*20 keyword matrix; model initialization parameter matrix; significant keyword configuration parameters for each group of wealth management products, 20 in total; if a product has fewer than 20 keywords, it will be filled with an empty string; keywords are marked by business personnel when listing wealth management products.

[0124] The keyword feature weight matrix PP for wealth management products is an N*20 numerical matrix. The keyword feature weight is set by the business personnel when listing wealth management products on the wealth management platform, and the selectable value is 1-20. When generating the feature weight matrix, the overall weight of each row is normalized, that is, the weight of each keyword = the score of the keyword / the sum of the scores of the keywords of the wealth management product, and is rounded to 4 decimal places. After normalization, the sum of the weights of each row is 1.

[0125] The Recommended Benefits Keyword Feature Weight Matrix (UPP) is an N*20 numerical matrix. Keyword feature weights are set by business personnel when recommending benefits on the benefits platform, with selectable values ​​from 1 to 20. When generating the feature weight matrix, the overall weight of each row is normalized, i.e., the weight of each keyword = the sum of the keyword scores / the sum of the benefit's keyword scores, rounded to four decimal places. After normalization, the sum of the weights in each row is 1. For simplicity, this embodiment does not establish a separate UPP matrix; instead, it uses a PP matrix, making the benefits equivalent to the financial products, with a one-to-one mapping.

[0126] The similarity association weight matrix PC for financial products is an N*N matrix; it represents the similarity weight matrix between N financial products; pairwise matching is performed based on the keyword matrix of the financial products; the matching degree is determined by a similarity coefficient r, with a value ranging from -1 to 1; then, the summation of the product weights in the corresponding keyword feature weight matrix of the financial products (keeping four decimal places) is performed to obtain the similarity association weight matrix between the two financial products. In this embodiment, for the sake of simplicity in mechanism description, only keywords between financial products are used as the similarity association; in the actual model implementation, the similarity association weight matrix between the two financial products is much richer.

[0127] The current keyword performance (frequency) vector group for user product interest is UP: a vector of 20 values; generated based on financial user behavior (browsing, holding financial products, changes in financial behavior, etc.), with different weights assigned to different behaviors for frequency counting; when the keyword performance vector exceeds 20 values, the corresponding keyword performance vectors are sorted in descending order of size, and the top 20 values ​​are used to form a vector group; among them, static behaviors such as holding financial products are counted as 1 frequency; if there is sustained interest, the weight is set to 15, i.e., the frequency count is 15; changes in financial behavior are counted as 1 frequency; if it is a clearly converted financial behavior, the weight of buying is 10, and the weight of selling is 5, i.e., the final count is 10 and / or 5; behaviors such as browsing are actually counted according to the weight set to 1. In this embodiment of the application, for the sake of simplicity, a vector counted by frequency is described. In the actual model implementation, the dimensions and implementation methods of the performance vector are more abundant.

[0128] The current keyword of user product interest - short-term migration state - derived result vector UTT: a vector of 20 values; generated in the short-term migration state processing module; the vector values ​​are as described above - migration state.

[0129] User product interest current keyword - short-term migration behavior - derived result vector UTB: a vector of 20 values; generated in the short-term migration behavior processing module; for vector values, see above - migration behavior.

[0130] User Product Interest - Current Keyword Interest Migration Event Vector Matrix UPE: Each row of the matrix is ​​a vector of 20 values, representing the keyword interest migration events of user u from time window t-1 to t; each value contains three event types. Therefore, UPE = {Financial Product Keyword Interest (Short-term, Initial) -> Product Migration -> Financial Product Keyword Interest (Short-term, Active), Financial Product Keyword Interest (Short-term, Decline) - Activation -> Financial Product Keyword Interest (Short-term, Active), Financial Product Keyword Interest (Short-term, Active) - Decline -> Financial Product Keyword Interest (Short-term, Decline)}.

[0131] User Product Interest Keyword Prediction Weight Vector Group (UPI): A vector of 20 values; predicts the keyword weight vector of user product interest for the next time period, i.e., time window t.

[0132] This application embodiment identifies users' financial product keyword interests in the initial / extinct state and stimulates conversion through benefit recommendations, enabling periodic guidance and migration to achieve benefit recommendations, user product guidance, and user product conversion.

[0133] like Figure 3 The establishment of the data recommendation model (e.g., the periodic guided migration model), and an overview of the functions and processes of the important modules related to the optimization of the data recommendation algorithm are shown below:

[0134] 1) Financial Product Relevance Processing Module: The same time window - product relevance matrix PC(t) represents the similarity between different financial products within a certain time window. It is the window value of time window t in the financial product similarity association weight matrix; i and j represent two different financial products, then the product relevance matrix PC(t) = {PC(i,j)}. In the financial product relevance processing module, the input is the financial product keyword feature weight parameter matrix of N financial products, and the similarity coefficient algorithm is used to calculate and output the similarity of the financial product keyword feature weight parameter matrices of N financial products.

[0135] 2) User Product Interest - Keyword Interest Migration Event Processing Module: Includes a migration derivative result processing submodule and a migration event calculation submodule.

[0136] The migration-derived result processing submodule takes as input the current keyword performance vector (number of times) group UP of user interest in time windows t-1 and t, and outputs the current keyword-short-term migration status-derived result vector UTT and the current keyword-short-term migration behavior-derived result vector UTB of user interest.

[0137] In the migration event calculation submodule, the output of the migration derivative result processing submodule is used as the input of the migration event calculation submodule to perform algorithmic judgment and output the interest migration event matrix UPE(t-1,t,u) of user product keywords for time windows t-1 and t. Different event types correspond to different processing methods for the user association degree matrix.

[0138] 3) Keyword prediction module for user product interests: includes a user relevance processing submodule and a product interest keyword prediction submodule;

[0139] In the user relevance processing submodule, the same user relevance matrix UC is determined based on different time windows. For example, t and t-1 represent two different time windows, then the user relevance matrix UC = {UC(t, t-1)}, which represents the similarity of users' financial product interests in two different time windows. The user relevance processing module requires two model building and calculation processes.

[0140] The first model is established by creating a Kalman filter prediction model to predict the keyword representation vector at time t. The process of establishing the Kalman filter prediction model is as follows: By inputting the current keyword performance vectors of user interests at time windows t-1 and t, respectively, the initial value at t-1 and the change value / rate of change / acceleration from time window t to time t-1 are calculated. Based on this, the Kalman filter state transition matrix is ​​established. After passing through the Kalman filter prediction algorithm, the prediction data for time window t is calculated, that is, the prediction group UP(|predict) of the current keyword performance (frequency) vector group of user interests. Then, data cleaning is performed. The current keyword performance vector group UP and the prediction vector group UP(|predict) of user interests are input. Data that deviates from 3 standard deviations are removed, and data that meet the conditions are retained. If both the actual vector group and the prediction vector group deviate from 3 standard deviations, the expected value of the vector group is used as the replacement. If both the actual vector group and the prediction vector group meet 3 standard deviations, the vector group that is closer to the standard deviation is retained. Thus, the current keyword performance (frequency) vector group UP(|r) of user interests for any time window is obtained.

[0141] The second model building process involves establishing a logistic regression model. In the first step of model building, the input is the vector group UP(|r) representing the current keyword performance (frequency) of user interests in any time window after data cleaning. This is used to calculate the relevance vector group for each user in time windows t-1 and t, and integrate them to obtain the user relevance matrix UC(t,u) for time window t. Data cleaning is then performed again, removing data that does not meet the three standard deviations. The model regression calculation is then performed, using the cleaned user relevance matrix UC(t-1,u) for time window t, the product relevance matrix PC(t) for time window t, and the interest migration event matrix UPE(t-1,t,u) for both time windows as independent variables. These are used as inputs to the logistic regression model, and the model is recursively trained, adjusting the evidence weights WOE until the dependent variable and a single independent variable form a monotonically linear relationship. For different weights, a recognition degree judgment is made, and a set of evidence weights (WOE) that meet the business recognition degree requirements are selected as the model parameters of the logistic regression model; thus, the establishment of the logistic regression model is completed; in this logistic regression model, the formula for calculating the user relevance matrix is ​​as follows:

[0142] UC(t,u)=f{UC(t-1,u),PC(t),UPE(t-1,t,u)}

[0143] Wherein, matrices UC(t,u) and UC(t-1,u) represent the relevance matrices of users in time windows t and t-1, respectively; PC(t) represents the similarity association weight matrix of financial products in time window t; and UPE(t-1,t,u) represents the current keyword interest migration event vector matrix of user u from time window t-1 to t.

[0144] In the product interest keyword prediction submodule, a model needs to be built; the logistic regression model is selected. Specifically, the financial product keyword feature weight parameter matrix of similar users in the t-1 time window, the financial product keyword feature weight parameter matrix of users u in the t-1 time window, and the user relevance matrix UC(t,u) in the t time window are used as inputs to the logistic regression model. After calculation, the output is the keyword prediction weight vector group UPI(t,u) of user product interests in the t time window.

[0145] 4) Decision-making module for guiding users' product interests in financial products and recommending benefits: This includes a decision-making sub-module for guiding users in financial products and a decision-making sub-module for recommending benefits.

[0146] For ease of description, wealth management products and recommended benefits are defined as one-to-one mappings, meaning the PP and UPP matrices are the same matrix. Therefore, the two sub-modules mentioned above are combined into one. In actual implementation, the algorithms for the wealth management product guidance decision-making sub-module and the benefit recommendation sub-module are similar but have differences. The decision-making processes of the two sub-modules are decoupled, resulting in a richer decision-making algorithm.

[0147] In the decision-making module, the input consists of the keyword prediction weight vector UPI(t,u) for user product interests within a time window t and the keyword feature weight parameter matrix PP (or UPP) for financial products. The calculation process uses a convolution algorithm to calculate the sum of all keyword weights for each financial product and the sum of all keyword weights for each equity. The decision-making process involves sorting the keyword weight sums in descending order and outputting the top three financial products as recommended financial products, as well as the top three equity interests as recommended equity interests.

[0148] exist Figure 3 The wealth management platform associated with the wealth management product recommendation application and the rights recommendation application includes a basic data area for inputting the required data into the data recommendation model. This basic data area may include the following data: wealth management products, wealth management accounts, wealth management transaction history, wealth management product keyword matrix PK, wealth management product keyword feature weight matrix PP, wealth management product rights transaction history, wealth management ability evaluation, user behavior (browsing, sharing), recommended rights keyword feature weight matrix UPP, and other necessary data. Figure 3 The equity pigtail can include a model-derived data area, which may contain the following data: the similarity association weight matrix PC of financial products, the current keyword performance (frequency) vector group UP of user product interests, the short-term migration state vector group UTT of user product interests, the short-term migration behavior vector group UTB of user product interests, the current keyword interest migration event vector matrix UPE of user product interests, the user product relevance matrix UC, and the keyword prediction weight vector group UPI of user product interests. The equity platform also includes a basic data area, specifically containing equity account data and equity product data, which can be used by the data recommendation model.

[0149] In this embodiment, the following are examples of cyclical migration: users' content access, interests, and attention exhibit cyclical characteristics. Content is used to guide and stimulate or enhance the cyclical migration of users' content access, interests, and attention. User product conversion: This is manifested in a financial product that introduces new users and achieves user conversion, including situations such as new users purchasing the product and old users purchasing the new product. User rights and interests: Rights and interests are manifested in a specific form of discount or discount equivalent (service). Users' preferences for discounts or discount equivalents are user rights and interests.

[0150] This application embodiment establishes a data recommendation model based on product interests. This model includes a financial product relevance processing module, a keyword interest migration event processing module, and a keyword prediction module. It can periodically guide users' product interest migration. Through this data recommendation model, new users can be recommended benefits when purchasing products, and existing users can be recommended benefits when purchasing new products. This provides benefit incentives based on guided product interests, and dynamic benefit recommendations that conform to user psychology and platform operation needs. It solves the problems of guiding the safety margin of new financial users, guiding the profit margin of new products for existing users, and the conversion problem of recommending high-quality financial products, thereby stimulating the conversion of financial users to products and increasing the scale of product sales.

[0151] This application's embodiments define keywords for financial products and rights by defining elements of user interest, state transitions, behavioral elements, and events. In daily operations, user product interest is guided through content guidance and analysis of user behavior (financial holding behavior, financial buying / selling transaction behavior, browsing, sharing, etc.). By establishing a data recommendation model (e.g., a periodic guided migration model), including setting periodic (time window) guided migration mechanisms (update mechanism, forgetting mechanism, etc.), user interest in financial products is guided and predicted. For scenarios where new users purchase financial products and existing users purchase new products, decisions on financial product recommendations and rights recommendations are output.

[0152] This application's data recommendation method combines product-guided benefit recommendation methods and collaborative filtering-based benefit recommendation methods, and proposes a data-based recommendation model, such as a periodic guided migration model, to analyze users' interest in financial products, proactively guide migration, and predict users' interest in financial products. It provides accurate financial product and benefit recommendations that meet the specific user psychology and platform operation needs for two specific business scenarios: product guidance and conversion for new users and new product guidance and conversion for existing users.

[0153] By establishing data recommendation models, such as periodic guided migration models, and implementing data recommendation algorithms, we can guide and predict users' interest in financial products, recommend and guide users' rights and interests, provide more accurate and effective rights and interests recommendations or incentives, and improve the conversion rate of financial platforms.

[0154] Figure 4 This is a schematic diagram of the main units of a data recommendation device according to an embodiment of this application. For example... Figure 4 As shown, the data recommendation device 400 includes a receiving unit 401, a first keyword feature weight matrix determination unit 402, a second keyword feature weight matrix determination unit 403, and a data recommendation unit 405.

[0155] The receiving unit 401 is configured to receive data recommendation requests, obtain the corresponding user identifier, and then obtain the corresponding product identifier and user behavior data based on the user identifier.

[0156] The first keyword feature weight matrix determination unit 402 is configured to determine the corresponding keyword based on the product identifier, and then determine the corresponding first keyword feature weight matrix based on the keyword.

[0157] The second keyword feature weight matrix determination unit 403 is configured to determine the time window stage corresponding to the keyword based on user behavior data, and to determine the corresponding second keyword feature weight matrices based on the time window stage and the first keyword feature weight matrix.

[0158] The candidate keyword determination unit 404 is configured to determine candidate keywords based on the first keyword feature weight matrix and each second keyword feature weight matrix;

[0159] The data recommendation unit 405 is configured to obtain the weights corresponding to each candidate keyword, and then determine and output the target recommended product data based on the candidate keywords and weights.

[0160] In some embodiments, the data recommendation unit 405 is further configured to: determine the corresponding candidate recommended product data based on the candidate keywords; determine the sum of keyword weights corresponding to the candidate recommended product data based on the weights and the candidate recommended product data; and determine and output the target recommended product data in the candidate recommended product data based on the sum of keyword weights.

[0161] In some embodiments, the second keyword feature weight matrix determination unit 403 is further configured to: determine the corresponding behavior type based on user behavior data; determine the corresponding behavior weight according to the behavior type; obtain the occurrence frequency of the behavior type, and then determine the corresponding behavior score according to the weight and the occurrence frequency; and determine the time window stage corresponding to the keyword according to the behavior score.

[0162] In some embodiments, the second keyword feature weight matrix determination unit 403 is further configured to: in response to a time window stage corresponding to decay, invoke the keyword feature weight matrix set to calculate the similarity between the first keyword feature weight matrix and each keyword feature weight matrix in the keyword feature weight matrix set; and determine each second keyword feature weight matrix corresponding to the first keyword feature weight matrix in each keyword feature weight matrix based on the similarity.

[0163] In some embodiments, the candidate keyword determination unit 404 is further configured to input the first keyword feature weight matrix and each of the second keyword feature weight matrices into the recommendation model to output the corresponding candidate keywords.

[0164] In some embodiments, the candidate keyword determination unit 404 is further configured to: invoke the product association processing module in the recommendation model to calculate the similarity between the first keyword feature weight matrix and each second keyword feature weight matrix based on a similarity coefficient algorithm; invoke the keyword interest migration event processing module in the recommendation model to obtain the current keyword performance vector group of user interests in adjacent time windows, and then output short-term migration status data and short-term migration behavior data; generate an interest migration event matrix of product keywords corresponding to adjacent time windows and user identifiers based on the short-term migration status data and short-term migration behavior data; and determine candidate keywords based on the similarity and the interest migration event matrix of product keywords.

[0165] In some embodiments, the candidate keyword determination unit 404 is further configured to: determine the target second keyword feature weight matrix in each second keyword feature weight matrix based on similarity; determine the corresponding keyword set based on the target second keyword feature weight matrix; and filter each keyword in the keyword set based on the product keyword interest migration event matrix to obtain candidate keywords.

[0166] It should be noted that the data recommendation method and data recommendation device in this application are related in terms of specific implementation, so repeated content will not be described again.

[0167] Figure 5 An exemplary system architecture 500 is shown that can be applied to the data recommendation method or data recommendation apparatus of the embodiments of this application.

[0168] like Figure 5 As shown, system architecture 500 may include terminal devices 501, 502, and 503, a network 504, and a server 505. Network 504 serves as the medium for providing communication links between terminal devices 501, 502, and 503 and server 505. Network 504 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0169] Users can use terminal devices 501, 502, and 503 to interact with server 505 via network 504 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 501, 502, and 503, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0170] Terminal devices 501, 502, and 503 can be various electronic devices with data recommendation processing screens and support web browsing, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0171] Server 505 can be a server providing various services, such as a backend management server supporting data recommendation requests submitted by users using terminal devices 501, 502, and 503 (this is just an example). The backend management server can receive data recommendation requests, obtain the corresponding user identifier, and then obtain the corresponding product identifier and user behavior data based on the user identifier; based on the product identifier, determine the corresponding keywords, and then determine the corresponding first keyword feature weight matrix; determine the time window stage corresponding to the keywords based on the user behavior data, and determine the corresponding second keyword feature weight matrices based on the time window stage and the first keyword feature weight matrix; determine candidate keywords based on the first keyword feature weight matrix and the second keyword feature weight matrices; obtain the weights corresponding to each candidate keyword, and then determine and output the target recommended product data based on the candidate keywords and their weights. This can improve the security of user financial management when recommending financial product data, and can achieve more effective data recommendation or incentives, promoting platform conversion effectiveness.

[0172] It should be noted that the data recommendation method provided in this application embodiment is generally executed by server 505, and correspondingly, the data recommendation device is generally set in server 505.

[0173] It should be understood that Figure 5 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0174] The following is for reference. Figure 6 It shows a schematic diagram of the structure of a computer system 600 suitable for implementing a terminal device according to the embodiments of this application. Figure 6 The terminal device shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0175] like Figure 6 As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 602 or programs loaded from storage section 608 into random access memory (RAM) 603. The RAM 603 also stores various programs and data required for the operation of the computer system 600. The CPU 601, ROM 602, and RAM 603 are interconnected via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0176] The following components are connected to I / O interface 605: an input section 606 including a keyboard, mouse, etc.; an output section 607 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to I / O interface 605 as needed. A removable medium 611, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 610 as needed so that computer programs read from it can be installed into storage section 608 as needed.

[0177] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 609, and / or installed from removable medium 611. When the computer program is executed by central processing unit (CPU) 601, it performs the functions defined above in the system of this application.

[0178] It should be noted that the computer-readable medium shown in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. Computer-readable storage media can be, for example, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, 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 devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0179] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0180] The units described in the embodiments of this application can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor can be described as including a receiving unit, a first keyword feature weight matrix determination unit, a second keyword feature weight matrix determination unit, a candidate keyword determination unit, and a data recommendation unit. The names of these units do not necessarily constitute a limitation on the unit itself.

[0181] In another aspect, this application also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to receive a data recommendation request, obtain a corresponding user identifier, and then obtain a corresponding product identifier and user behavior data based on the user identifier; determine corresponding keywords based on the product identifier, and then determine a corresponding first keyword feature weight matrix based on the keywords; determine the time window stage corresponding to the keywords based on the user behavior data, and determine corresponding second keyword feature weight matrices based on the time window stage and the first keyword feature weight matrix; determine candidate keywords based on the first keyword feature weight matrix and the second keyword feature weight matrices; obtain the weights corresponding to each candidate keyword, and then determine and output target recommended product data based on the candidate keywords and their weights.

[0182] The computer program product of this application includes a computer program that, when executed by a processor, implements the data recommendation method in the embodiments of this application.

[0183] According to the technical solution of the embodiments of this application, the security of users' financial management can be improved when recommending financial product data to users.

[0184] The specific embodiments described above do not constitute a limitation on the scope of protection of this application. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the scope of protection of this application.

Claims

1. A data recommendation method, characterized in that, include: Receive data recommendation requests, obtain the corresponding user identifiers, and then obtain the corresponding product identifiers and user behavior data based on the user identifiers; Based on the product identifier, the corresponding keywords are determined, and then the corresponding first keyword feature weight matrix is ​​determined based on the keywords. Based on the user behavior data, a time window stage corresponding to the keyword is determined. Based on the time window stage and the first keyword feature weight matrix, corresponding second keyword feature weight matrices are determined, including: in response to the time window stage corresponding to decay, a set of keyword feature weight matrices is invoked to calculate the similarity between the first keyword feature weight matrix and each keyword feature weight matrix in the set of keyword feature weight matrices; based on the similarity, each second keyword feature weight matrix corresponding to the first keyword feature weight matrix is ​​determined. Candidate keywords are determined based on the first keyword feature weight matrix and each of the second keyword feature weight matrices; Obtain the weights corresponding to each candidate keyword, and then determine and output the target recommended product data based on the candidate keywords and the weights. The step of determining candidate keywords includes: inputting the first keyword feature weight matrix and each of the second keyword feature weight matrices into a recommendation model to output corresponding candidate keywords, including: calling the product association processing module in the recommendation model to calculate the similarity between the first keyword feature weight matrix and each of the second keyword feature weight matrices based on a similarity coefficient algorithm; calling the keyword interest migration event processing module in the recommendation model to obtain the current keyword performance vector group of user interests in adjacent time windows, and then outputting short-term migration status data and short-term migration behavior data; generating an interest migration event matrix of product keywords corresponding to the adjacent time windows and the user identifier based on the short-term migration status data and the short-term migration behavior data; and determining candidate keywords based on the similarity and the interest migration event matrix of the product keywords.

2. The method according to claim 1, characterized in that, The process of determining and outputting target recommended product data includes: Based on the candidate keywords, determine the corresponding candidate recommended product data; Based on the weights and the candidate recommended product data, determine the sum of the keyword weights corresponding to the candidate recommended product data; Based on the sum of the keyword weights, the target recommended product data in the candidate recommended product data is determined and output.

3. The method according to claim 1, characterized in that, The step of determining the time window stage corresponding to the keyword based on the user behavior data includes: Based on the user behavior data, the corresponding behavior type is determined; Based on the behavior type, determine the corresponding behavior weight; The number of occurrences of the behavior type is obtained, and then the corresponding behavior score is determined based on the weight and the number of occurrences; Based on the behavior score, the time window stage corresponding to the keyword is determined.

4. The method according to claim 1, characterized in that, The step of determining candidate keywords based on the similarity and the interest migration event matrix of the product keywords includes: Based on the similarity, determine the target second keyword feature weight matrix in each of the second keyword feature weight matrices; Based on the target second keyword feature weight matrix, determine the corresponding keyword set; Candidate keywords are obtained by filtering each keyword in the keyword set based on the interest migration event matrix of the product keywords.

5. A data recommendation device, characterized in that, include: The receiving unit is configured to receive data recommendation requests, obtain the corresponding user identifier, and then obtain the corresponding product identifier and user behavior data based on the user identifier. The first keyword feature weight matrix determination unit is configured to determine the corresponding keyword based on the product identifier, and then determine the corresponding first keyword feature weight matrix based on the keyword. The second keyword feature weight matrix determination unit is configured to determine the time window stage corresponding to the keyword based on the user behavior data, and to determine the corresponding second keyword feature weight matrices based on the time window stage and the first keyword feature weight matrix. The candidate keyword determination unit is configured to determine candidate keywords based on the first keyword feature weight matrix and each of the second keyword feature weight matrices; The data recommendation unit is configured to obtain the weights corresponding to each of the candidate keywords, and then determine and output the target recommended product data based on the candidate keywords and the weights. The second keyword feature weight matrix determination unit is further configured to: in response to the time window stage corresponding to decay, invoke the keyword feature weight matrix set to calculate the similarity between the first keyword feature weight matrix and each keyword feature weight matrix in the keyword feature weight matrix set; and determine each second keyword feature weight matrix corresponding to the first keyword feature weight matrix in each keyword feature weight matrix based on the similarity. The candidate keyword determination unit is further configured to: input the first keyword feature weight matrix and each of the second keyword feature weight matrices into the recommendation model to output the corresponding candidate keywords, including: calling the product association processing module in the recommendation model to calculate the similarity between the first keyword feature weight matrix and each of the second keyword feature weight matrices based on the similarity coefficient algorithm; The keyword interest migration event processing module in the recommendation model is invoked to obtain the current keyword performance vector group of user interests in adjacent time windows, and then output short-term migration status data and short-term migration behavior data; based on the short-term migration status data and the short-term migration behavior data, an interest migration event matrix of product keywords corresponding to the adjacent time windows and the user identifier is generated; based on the similarity and the interest migration event matrix of the product keywords, candidate keywords are determined.

6. The apparatus according to claim 5, characterized in that, The data recommendation unit is further configured to: Based on the candidate keywords, determine the corresponding candidate recommended product data; Based on the weights and the candidate recommended product data, determine the sum of the keyword weights corresponding to the candidate recommended product data; Based on the sum of the keyword weights, the target recommended product data in the candidate recommended product data is determined and output.

7. The apparatus according to claim 5, characterized in that, The second keyword feature weight matrix determination unit is further configured as follows: Based on the user behavior data, the corresponding behavior type is determined; Based on the behavior type, determine the corresponding behavior weight; The number of occurrences of the behavior type is obtained, and then the corresponding behavior score is determined based on the weight and the number of occurrences; Based on the behavior score, the time window stage corresponding to the keyword is determined.

8. A data recommendation electronic device, characterized in that, include: One or more processors; Storage device 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 method as described in any one of claims 1-4.

9. A computer-readable medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method as described in any one of claims 1-4.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-4.

Images