Recommendation method and device based on random walk algorithm and graph theory

Abstract

The invention provides a recommendation method and device based on a random walk algorithm and a graph theory, and the method comprises the steps: collecting the credit investigation data of a plurality of users, and the feature data and behavior data of financial products, and constructing a bipartite graph; constructing a scoring matrix based on the bipartite graph, and further constructing a user similarity matrix and a product similarity matrix; performing Laplacian regularization processing on the target user similarity matrix and the target product similarity matrix to obtain a target user similarity matrix and a target product similarity matrix; by setting the initial step length, the number of iterations and the stop rule of the random walk algorithm, the algorithm is executed to extract a user access path. And according to the behavior data and the user access path, calculating the interest probability of the target user on the target financial product, and generating a recommendation list containing a plurality of target financial products. By constructing the bipartite graph and combining the random walk algorithm, the recommendation accuracy is remarkably improved, the complex relationship between the user and the product is accurately captured, the user interest is predicted, and a high-correlation recommendation result is provided. In conclusion, the recommendation accuracy is improved, and the user experience is improved.

Description

Technical Field

[0001] The present invention relates to the field of financial technology, and in particular to a recommendation method and device based on a random walk algorithm and graph theory. Background Art

[0002] In the financial field, the application environment of recommendation systems is highly complex and changeable. First, the volatility of the financial market means that the characteristics of financial products (such as yield, risk, liquidity, etc.) may change at any time, requiring the recommendation system to be able to update and adjust the model in real time.

[0003] Existing financial product recommendation systems mainly rely on traditional algorithms such as collaborative filtering and content recommendation, but these methods are not ideal when dealing with data sparsity and cold start problems. Collaborative filtering builds a relationship matrix by analyzing user behavior, but the characteristics of financial products (such as yield, risk, etc.) are constantly changing with market fluctuations, so time series factors are particularly important in recommendations, and traditional algorithms often have difficulty coping with such dynamic changes. In addition, due to the lack of historical data for new users, traditional methods will experience cold start phenomena, which not only affects the recommendation effect, but may also reduce user retention. At the same time, due to the sensitivity of financial data and strict requirements for privacy protection, data acquisition and model application are subject to many restrictions, which further restricts the optimization and development of recommendation systems. Summary of the invention

[0004] In view of this, an embodiment of the present invention provides a recommendation method and device based on a random walk algorithm and graph theory to solve the problem of low recommendation accuracy and poor efficiency.

[0005] To achieve the above objectives, the embodiments of the present invention provide the following technical solutions:

[0006] A first aspect of an embodiment of the present invention discloses a recommendation method based on a random walk algorithm and graph theory, the method comprising:

[0007] Collecting credit data of multiple users and characteristic data of multiple financial products, as well as behavioral data of multiple users with respect to multiple financial products;

[0008] Constructing a bipartite graph based on the credit data, the feature data and the behavior data;

[0009] According to the bipartite graph, the rating data of the plurality of users on the plurality of financial products are obtained to construct a rating matrix, and a user similarity matrix and a product similarity matrix are constructed according to the rating matrix;

[0010] Performing Laplace regularization processing on the user similarity matrix and the product similarity matrix to obtain a target user similarity matrix and a target product similarity matrix;

[0011] Setting the initial step size, number of iterations and stopping rule corresponding to the random walk algorithm based on the target user similarity matrix and the target product similarity matrix, and executing the random walk algorithm to extract the user access path;

[0012] The probability of a target user's interest in a target financial product is calculated based on the behavior data and the user access path, and a recommendation list containing multiple target financial products is generated based on the interest probability; the target user is any of the users; and the target financial product is a financial product with which there is no interaction between the target user and the target user.

[0013] Preferably, it also includes:

[0014] Collect feedback data of the target user on the target financial product in the recommendation list, and optimize the parameters of the random walk algorithm.

[0015] Preferably, constructing a user similarity matrix and a product similarity matrix based on the bipartite graph includes:

[0016] Obtaining ratings of any two of the users on the plurality of financial products according to the bipartite graph;

[0017] Calculating the sum of the products of the scores of the two users for the financial product scored together;

[0018] Based on the score and the sum of the score products, calculating the cosine similarity between the two users to construct a user similarity matrix;

[0019] Obtaining, according to the bipartite graph, a rating given by any user to any two of the financial products;

[0020] The cosine similarity between the two financial products is calculated based on the scores between the two financial products to construct a product similarity matrix.

[0021] Preferably, executing the random walk algorithm to extract the user access path includes:

[0022] Execute the random walk algorithm according to the initial step size, the number of iterations and the stopping rule;

[0023] For each iteration, a vector is randomly generated; the user's access position is determined according to the vector, and the user's new access position is calculated according to the initial step length;

[0024] A transfer matrix is ​​constructed based on all access locations to obtain the user access path.

[0025] Preferably, the calculating the probability of the target user's interest in the target financial product according to the behavior data and the user access path includes:

[0026] Acquire a plurality of the financial products corresponding to the user access path and the quantity of the financial products;

[0027] Calculating the cosine similarity between each of the financial products and the target financial product based on the quantity;

[0028] Acquire target behavior data of a target user for a plurality of the financial products from the behavior data;

[0029] A plurality of financial products corresponding to the target behavior data are obtained, and a weighted average calculation is performed based on the cosine similarity between each of the financial products and the target financial product to obtain the probability of the target user's interest in the target financial product.

[0030] A second aspect of an embodiment of the present invention discloses a recommendation device based on a random walk algorithm and graph theory, the device comprising:

[0031] A collection unit, used to collect credit data of multiple users and feature data of multiple financial products, as well as behavior data of multiple users with respect to multiple financial products;

[0032] A first construction unit, configured to construct a bipartite graph according to the credit data, the feature data and the behavior data;

[0033] A second construction unit is used to obtain rating data of the plurality of users on the plurality of financial products according to the bipartite graph, construct a rating matrix, and construct a user similarity matrix and a product similarity matrix according to the rating matrix;

[0034] A processing unit, used for performing Laplace regularization processing on the user similarity matrix and the product similarity matrix to obtain a target user similarity matrix and a target product similarity matrix;

[0035] An extraction unit, used to set an initial step size, a number of iterations and a stop rule corresponding to a random walk algorithm based on the target user similarity matrix and the target product similarity matrix, and execute the random walk algorithm to extract user access paths;

[0036] A generating unit is used to calculate the probability of a target user's interest in a target financial product based on the behavior data and the user access path, and to generate a recommendation list containing multiple target financial products based on the interest probability; the target user is any of the users; and the target financial product is a financial product with which there is no interaction between the target user and the target user.

[0037] Preferably, it also includes:

[0038] An optimization unit is used to collect feedback data from the target user on the target financial product in the recommendation list, and optimize the parameters of the random walk algorithm.

[0039] Preferably, the second building block comprises:

[0040] A first scoring acquisition module is used to obtain scores of any two of the users on the plurality of financial products according to the bipartite graph;

[0041] A calculation module, used for calculating the sum of the products of the scores of the two users for the financial product scored together;

[0042] A first cosine similarity calculation module, used for calculating the cosine similarity between two users based on the score and the sum of the score products, so as to construct a user similarity matrix;

[0043] A second scoring acquisition module is used to obtain a score between any two of the financial products given by any one of the users according to the bipartite graph;

[0044] The second cosine similarity calculation module is used to calculate the cosine similarity between the two financial products based on the scores between the two financial products to construct a product similarity matrix.

[0045] Preferably, the extraction unit is specifically used for:

[0046] The random walk algorithm is executed according to the initial step size, the number of iterations and the stopping rule; a vector is randomly generated for each iteration; the user's access position is determined according to the vector, and the user's new access position is calculated according to the initial step size; a transfer matrix is ​​constructed according to all access positions to obtain the user's access path.

[0047] Preferably, the generating unit comprises:

[0048] A first acquisition module, configured to acquire a plurality of the financial products corresponding to the user access path and the quantity of the financial products;

[0049] A first calculation module, used to calculate the cosine similarity between each of the financial products and the target financial product based on the quantity;

[0050] A second acquisition module is used to acquire target behavior data of a target user for a plurality of the financial products from the behavior data;

[0051] The second calculation module is used to obtain multiple financial products corresponding to the target behavior data, and perform weighted average calculation based on the cosine similarity between each of the financial products and the target financial product to obtain the probability of the target user's interest in the target financial product.

[0052] Based on the above-mentioned embodiment of the present invention, a recommendation method and device based on a random walk algorithm and graph theory are provided. The credit data of multiple users, the characteristic data and behavior data of financial products are collected to construct a bipartite graph; a rating matrix is ​​constructed based on the bipartite graph, and a user similarity matrix and a product similarity matrix are further constructed; Laplace regularization is performed on them to obtain a target user similarity matrix and a target product similarity matrix; the algorithm is executed to extract user access paths by setting the initial step size, number of iterations and stopping rules of the random walk algorithm. The probability of the target user's interest in the target financial product is calculated based on the behavior data and the user access path, and a recommendation list containing multiple target financial products is generated. By constructing a bipartite graph and combining it with a random walk algorithm, the recommendation accuracy is significantly improved, the complex relationship between users and products is accurately captured, user interests are predicted, and highly relevant recommendation results are provided. In summary, the recommendation accuracy is improved and the user experience is enhanced. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0054] Figure 1 A flowchart of a recommendation method based on a random walk algorithm and graph theory provided by an embodiment of the present invention;

[0055] Figure 2 A flowchart of constructing a user similarity matrix and a product similarity matrix provided by an embodiment of the present invention;

[0056] Figure 3 A flowchart of calculating the probability of a target user's interest in a target financial product provided by an embodiment of the present invention;

[0057] Figure 4 A structural block diagram of a recommendation device based on a random walk algorithm and graph theory provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0058] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0059] In this application, the terms "comprises", "comprising" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or device comprising the element.

[0060] From the background technology, we can see that the existing financial product recommendation system faces challenges such as data sparsity, cold start problem and market dynamic changes. At the same time, due to privacy protection requirements, data acquisition and model application are limited, which affects the recommendation effect and system development.

[0061] Therefore, the embodiment of the present invention provides a recommendation method and device based on a random walk algorithm and graph theory, which collects credit data of multiple users, characteristic data and behavioral data of financial products, and constructs a bipartite graph; constructs a scoring matrix based on the bipartite graph, and further constructs a user similarity matrix and a product similarity matrix; performs Laplace regularization processing on them to obtain a target user similarity matrix and a target product similarity matrix; and executes the algorithm to extract user access paths by setting the initial step size, number of iterations and stopping rules of the random walk algorithm. The probability of the target user's interest in the target financial product is calculated based on the behavioral data and the user access path, and a recommendation list containing multiple target financial products is generated. By constructing a bipartite graph and combining it with a random walk algorithm, the accuracy of recommendations is significantly improved, and the complex relationship between users and products can be accurately captured, user interests can be predicted, and highly relevant recommendation results can be provided. Overall, the accuracy of recommendations is improved and the user experience is enhanced.

[0062] See also Figure 1 , shows a flowchart of a recommendation method based on a random walk algorithm and graph theory provided by an embodiment of the present invention.

[0063] It should be noted that the information (including but not limited to user device information, user personal information) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0064] Step S101: Collect credit data of multiple users and feature data of multiple financial products, as well as behavior data of multiple users for multiple financial products.

[0065] In the specific implementation of step S101, the credit data of multiple users are collected; at the same time, the characteristic data of multiple financial products are collected, including but not limited to product type, risk level, rate of return, etc.; and the behavioral data of multiple users on multiple financial products are collected, including but not limited to voice information input by users, clicks, exposure, browsing, order placement, order cancellation, etc.

[0066] Step S102: construct a bipartite graph based on the credit data, feature data and behavior data.

[0067] In the process of implementing step S102, the user set U, the financial product set P and the interaction set E are obtained based on the collected credit data, feature data and behavior data. Based on this, users and financial products are regarded as two different types of node sets, and the user's interaction with the product is regarded as the edge to construct a bipartite graph G. Specifically, G=(U∪P,E).

[0068] Among them, the user set U: consists of all users, and the attributes of the nodes include the basic information of the users, such as age, income, credit score, interest preference, etc. The financial product set P: consists of all financial products, and the attributes of the nodes include product type, risk level, rate of return, etc. The interaction set E: Each edge represents the interaction between a user and a financial product. The interaction can include browsing, clicking, purchasing, rating, etc. The weight of each edge can reflect the strength of the interaction.

[0069] Step S103: obtaining rating data of multiple users on multiple financial products according to the bipartite graph, constructing a rating matrix, and constructing a user similarity matrix and a product similarity matrix according to the rating matrix.

[0070] In the specific implementation of step S103, based on the bipartite graph, the rating data of multiple users on multiple financial products are obtained to construct a rating matrix R. Specifically, the rating matrix is ​​shown in formula (1).

[0071] R=[r ij ] (1)

[0072] Among them, rij represents the rating of financial product j by user i.

[0073] And, using methods such as cosine similarity, a user similarity matrix and a product similarity matrix are constructed according to the rating matrix. For the specific implementation process, see the embodiment of the present invention. Figure 2 As shown, Figure 2 Included:

[0074] Step S201: Obtain ratings of any two users on multiple financial products according to the bipartite graph.

[0075] For example: based on the bipartite graph, obtain the ratings of user 1 and user 2 on financial product 01, the ratings of user 1 and user 2 on financial product 02, etc.

[0076] Step S202: Calculate the sum of the products of the scores of the two users for the jointly rated financial product.

[0077] For example, the sum of the products of user 1's first rating of financial product 01, user 1's second rating of financial product 02, user 2's third rating of financial product 01, and user 2's fourth rating of financial product 02 is calculated.

[0078] Step S203: Based on the sum of the scores and the products of the scores, the cosine similarity between the two users is calculated to construct a user similarity matrix.

[0079] The specific calculation process is shown in formula (2).

[0080] (2)

[0081] Among them, ui and uj represent two users; r ik represents the rating of user ui on the kth financial product; r jk represents the rating of user uj on the k-th financial product; Represents the cosine similarity between user ui and user uj.

[0082] It represents the sum of the products of the ratings of all products rated by user ui and user uj.

[0083] Represents the product of the Euclidean norms of the rating vectors of user ui and user uj.

[0084] It can be understood that the similarity between two users calculated by formula (2) is used to measure the similarity of their interests. A similarity value close to 1 indicates that the interests of the two users are very similar, close to -1 indicates that their interests are very different, and close to 0 indicates that there is no significant similarity or difference between the interests of the two users.

[0085] Step S204: Obtain any user's rating between any two financial products according to the bipartite graph.

[0086] For example, user 1's rating of financial product 01 and financial product 02 are obtained based on the bipartite graph.

[0087] Step S205: Calculate the cosine similarity between the two financial products based on the ratings between the two financial products to construct a product similarity matrix.

[0088] The specific calculation process is shown in formula (3).

[0089] (3)

[0090] Among them, r ki represents the rating of user k on financial product i; r kj represents the rating of user k on financial product j; Represents the cosine similarity between financial product i and financial product j.

[0091] It can be understood that the user similarity matrix is ​​used to represent the similarity relationship between different users, for example, based on their behaviors, interests or ratings, etc. Product similarity matrix: used to represent the similarity relationship between different financial products, for example, based on their characteristics, user evaluations, etc.

[0092] Step S104: Perform Laplace regularization processing on the user similarity matrix and the product similarity matrix to obtain a target user similarity matrix and a target product similarity matrix.

[0093] It should be noted that in order to enhance the generalization ability of the model and avoid overfitting, Laplace regularization can be applied to the similarity matrix. Laplace regularization usually adjusts the adjacency relationship of the similarity matrix to make it smoother, thereby controlling excessive dependence.

[0094] In the specific implementation of step S104, Laplace regularization is performed on the user similarity matrix and the product similarity matrix using formula (4) to obtain the target user similarity matrix and the target product similarity matrix.

[0095] Specifically, formula (4) is as follows:

[0096] (4)

[0097] Where: A is the original similarity matrix (or adjacency matrix); A represents the relationship between users and financial products. ij =1 means user i has rated financial product j (or there is some interaction); if there is no rating, then Aij =0.

[0098] D is the degree matrix, which is a diagonal matrix. Each element on the diagonal D ii Represents the degree of user i or financial product i (i.e., the number of connections of the user or financial product, that is, how many financial products or users they have rated or interacted with, respectively).

[0099] L is the Laplacian matrix, defined as L=D−A. is the similarity matrix after Laplace regularization.

[0100] Step S105: setting the initial step size, number of iterations and stopping rule corresponding to the random walk algorithm based on the target user similarity matrix and the target product similarity matrix, and executing the random walk algorithm to extract the user access path.

[0101] It can be understood that, first of all, according to the principle of random walk algorithm, assuming that X(t) represents the current state of the ant (simulating user behavior), at each step, the ant will select a new financial product node from the current financial product node according to a certain strategy. Each step of the transfer is represented by a random variable r i To determine, this random variable reflects the behavior choice of the ants. Specifically, as shown in formula (5).

[0102] X(t)={r 1, r2,…,r t} (5)

[0103] Among them, X(t) represents the number of steps of the ant's random walk, r i is the random step length of the i-th step.

[0104] It should be noted that a transfer matrix is ​​constructed, combining the user's rating of financial products and the similarity between financial products to set the path and depth of the random walk algorithm, that is, to set the initial step size and number of iterations. Specifically, the transfer matrix is ​​shown in formula (6).

[0105] P=[p ij ] (6)

[0106] Among them, p ij Represents the probability of transition from financial product i to financial product j.

[0107] It is understandable that in order to ensure that the random walk stops at the appropriate time, the stopping rules are set. Common stopping rules include: step limit: the walk stops after reaching a certain number of steps. Usually the number of steps is related to the depth of the walk. The more steps, the deeper the walk. Visit count limit: when the number of visits to a product by a user exceeds a certain threshold, it is considered that the user has "enough" explored the product and can stop walking. The embodiment of the present invention does not specifically limit the setting method of the stop rule.

[0108] It should be noted that the random walk algorithm is executed to extract the user access path, specifically: the random walk algorithm is executed according to the initial step size, number of iterations and stopping rules; for each iteration, a vector is randomly generated; the user's access location is determined according to the vector, and the user's new access location is calculated according to the initial step size; a transfer matrix is ​​constructed according to all access locations to obtain the user access path.

[0109] Specifically, the user access path is shown in formula (7).

[0110] (7)

[0111] in, represents the access path of user u; p i Represents the i-th financial product in the access path.

[0112] In practical applications, based on the user access path, an interest transfer matrix can be constructed to represent the probability that a user will transfer to other products after visiting a certain product.

[0113] Assume that the user accesses p in the path i and p j For two products, the transition probability can be calculated according to the following formula (8):

[0114] (8)

[0115] It is understandable that the interest transfer matrix can help identify which products are potential targets for user interest transfer. For example, if a user often transfers to p2 after visiting product p1, it can be considered that there is a strong correlation between p1 and p2, which may be a transfer of user interest.

[0116] It should be noted that potential recommendation paths can be identified based on the interest transfer matrix. For example, if a user shows a strong interest in product p1, the transfer matrix can be used to predict that the user may be interested in product p2 or p3, thereby generating a potential recommendation path.

[0117] Step S106: Calculate the probability of the target user's interest in the target financial product based on the behavior data and the user's access path, and generate a recommendation list containing multiple target financial products based on the interest probability.

[0118] It should be noted that the target user is any user; the target financial product is a financial product that has no interaction with the target user.

[0119] It should be noted that the specific implementation process of calculating the probability of a target user's interest in a target financial product based on behavioral data and user access paths is as follows: Figure 3 As shown, Figure 3 include:

[0120] Step S301: obtaining a plurality of financial products and the quantity of the financial products corresponding to the user access path.

[0121] It should be noted that the user access path is a sequence of financial products that the user has interacted with in the past. These financial interactions include but are not limited to browsing, clicking, purchasing and other behaviors.

[0122] Each financial product k represents a node on the path, and all financial products on the path p k All of these may have an impact on the prediction of users' interest in new financial products.

[0123] Step S302: Calculate the cosine similarity between each financial product and the target financial product in combination with the quantity.

[0124] Specifically, cosine similarity cosine(p,p k The calculation process of ) is shown in formula (9).

[0125] (9)

[0126] Among them, p and p k They are target financial product p and financial product p k The eigenvector of It represents their similarity in the feature space, and the denominator is the norm of the two financial product vectors.

[0127] Step S303: Obtain target behavior data of the target user for multiple financial products from the behavior data.

[0128] Step S304: obtaining multiple financial products corresponding to the target behavior data, and performing weighted average calculation based on the cosine similarity between each financial product and the target financial product to obtain the probability of the target user's interest in the target financial product.

[0129] It can be understood that the calculation process of the interest probability is as shown in formula (10).

[0130] (10)

[0131] in, is each financial product p in the behavior path of user u k The sum of cosine similarities with the target financial product p.

[0132] is the number of financial products corresponding to the access path of user u, which serves as a normalization factor so that the score will not be biased due to different path lengths.

[0133] It can be understood that based on the interest probability Score(u,p) of each target financial product p, a personalized recommendation can be made for user u, generating a recommendation list containing financial products with the highest interest scores. After sorting, the user may show a high interest in these products.

[0134] In some specific embodiments, through dynamic adjustment and optimization, feedback data of target users on target financial products in the recommendation list are collected, and the parameters of the random walk, such as transition probability and stopping conditions, are flexibly adjusted to ensure that the recommendation model can adapt to changing needs.

[0135] At the same time, optimize the recommendation strategy and improve its adaptability and accuracy to provide more personalized recommendation results.

[0136] In actual applications, through feedback learning loops, we continuously collect user feedback on recommended products, including clicks, purchases, ratings, and other behaviors, and automatically update recommendation models and strategies accordingly, thereby achieving continuous optimization of the recommendation system and ensuring long-term improvement in recommendation effectiveness.

[0137] In the embodiment of the present invention, by constructing a user-product bipartite graph and combining it with a random walk algorithm, the accuracy of recommendations is significantly improved, and the complex relationship between users and products can be accurately captured, user interests can be predicted, and highly relevant recommendation results can be provided. In addition, by analyzing the user behavior path, a personalized recommendation path is discovered, and the recommendation strategy can be dynamically adjusted to ensure consistency with the user's real-time needs and interests. At the same time, the recommendation strategy can be optimized in real time according to market changes and user feedback to improve the timeliness and flexibility of the recommendation. Through the feedback learning cycle, continuous self-optimization is carried out, and the strategy is automatically adjusted based on the user's behavioral feedback to improve the recommendation effect and user satisfaction. In general, this solution performs well in terms of recommendation accuracy, personalization, adaptability and self-optimization, which can improve user experience, enhance the market competitiveness of financial products, and bring greater commercial value to financial institutions.

[0138] For a better Figure 1The method shown is explained and further illustrated in conjunction with the following examples.

[0139] Assume that the user set ;

[0140] Financial product collection ;

[0141] User-product interaction data (simplified as purchase behavior): U1 purchased p1 and p2; U2 purchased p2 and p3; U3 purchased p3 and p4; U4 purchased p1 and p4.

[0142] First, construct a bipartite graph , where E contains the edges (u1,p1),(u1,p2),(u2,p2),(u2,p3),(u3,p3),(u3,p4),(u4,p1),(u4,p4).

[0143] Next, assuming that a rating of 1 indicates purchase, construct the rating matrix R: .

[0144] Similarity calculation: Calculate the cosine similarity between users and products. For example, calculate the similarity between u1 and u2:

[0145]

[0146] Again, walking strategy design: set the initial step size step=10.0 and the number of iterations N=100.

[0147] Randomly generate a vector u, for example, u=[0.6,-0.8].

[0148] Update the location of user u1:

[0149]

[0150] Calculate the new position and adjust it to a reasonable product range based on business rules.

[0151] Furthermore, assuming that the path of the random walk is .

[0152] Therefore, the probability of interest is calculated as:

[0153] Calculate u1’s interest score for p4:

[0154]

[0155] Assumptions ,

[0156] but: .

[0157] Corresponding to the recommendation method based on random walk algorithm and graph theory provided in the above embodiment of the present invention, see Figure 4 , shows a structural block diagram of a recommendation device based on a random walk algorithm and graph theory provided by an embodiment of the present invention, the device includes: a collection unit 401, a first construction unit 402, a second construction unit 403, a processing unit 404, an extraction unit 405 and a generation unit 406.

[0158] The collection unit 401 is used to collect the credit data of multiple users and the characteristic data of multiple financial products, as well as the behavior data of multiple users for multiple financial products.

[0159] The first construction unit 402 is used to construct a bipartite graph according to the credit data, feature data and behavior data.

[0160] The second construction unit 403 is used to obtain the rating data of multiple users on multiple financial products according to the bipartite graph, construct a rating matrix, and construct a user similarity matrix and a product similarity matrix according to the rating matrix.

[0161] The processing unit 404 is used to perform Laplace regularization processing on the user similarity matrix and the product similarity matrix to obtain a target user similarity matrix and a target product similarity matrix.

[0162] The extraction unit 405 is used to set the initial step size, the number of iterations and the stopping rule corresponding to the random walk algorithm based on the target user similarity matrix and the target product similarity matrix, and execute the random walk algorithm to extract the user access path.

[0163] The extraction unit 405 is specifically used to: execute the random walk algorithm according to the initial step size, the number of iterations and the stopping rule; randomly generate a vector for each iteration; determine the user's access location according to the vector, and calculate the user's new access location according to the initial step size; construct a transfer matrix according to all access locations to obtain the user's access path.

[0164] The generating unit 406 is used to calculate the interest probability of the target user in the target financial product according to the behavior data and the user access path, and generate a recommendation list containing multiple target financial products according to the interest probability; the target user is any user; the target financial product is a financial product with which there is no interaction between the target user and the target user.

[0165] In the embodiment of the present invention, by constructing a user-product bipartite graph and combining it with a random walk algorithm, the accuracy of recommendations is significantly improved, and the complex relationship between users and products can be accurately captured, user interests can be predicted, and highly relevant recommendation results can be provided. In addition, by analyzing the user behavior path, a personalized recommendation path is discovered, and the recommendation strategy can be dynamically adjusted to ensure consistency with the user's real-time needs and interests. At the same time, the recommendation strategy can be optimized in real time according to market changes and user feedback to improve the timeliness and flexibility of the recommendation. Through the feedback learning cycle, continuous self-optimization is carried out, and the strategy is automatically adjusted based on the user's behavioral feedback to improve the recommendation effect and user satisfaction. In general, this solution performs well in terms of recommendation accuracy, personalization, adaptability and self-optimization, which can improve user experience, enhance the market competitiveness of financial products, and bring greater commercial value to financial institutions.

[0166] Combination Figure 4 The device also includes: an optimization unit, which is used to collect feedback data from target users on target financial products in the recommendation list and optimize the parameters of the random walk algorithm.

[0167] Combination Figure 4 As shown in the content, the second construction unit 403 includes: a first scoring acquisition module, a sum calculation module, a first cosine similarity calculation module, a second scoring acquisition module and a second cosine similarity calculation module.

[0168] The first scoring module is used to obtain scores of any two users on multiple financial products according to the bipartite graph.

[0169] The calculation and sum module is used to calculate the sum of the products of the ratings of two users on the financial products that they have rated together.

[0170] The first cosine similarity calculation module is used to calculate the cosine similarity between two users based on the score and the sum of the score products to construct a user similarity matrix.

[0171] The second scoring module is used to obtain the scores between any two financial products given by any user according to the bipartite graph.

[0172] The second cosine similarity calculation module is used to calculate the cosine similarity between the two financial products based on the scores between the two financial products to construct a product similarity matrix.

[0173] Combination Figure 4 The content shown, the generating unit 406 includes: a first acquisition module, a first calculation module, a second acquisition module and a second calculation module.

[0174] The first acquisition module is used to acquire multiple financial products and the quantity of financial products corresponding to the user access path.

[0175] The first calculation module is used to calculate the cosine similarity between each financial product and the target financial product in combination with the quantity.

[0176] The second acquisition module is used to acquire target behavior data of the target user for multiple financial products from the behavior data.

[0177] The second calculation module is used to obtain multiple financial products corresponding to the target behavior data, and perform weighted average calculation based on the cosine similarity between each financial product and the target financial product to obtain the probability of the target user's interest in the target financial product.

[0178] Each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can refer to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system or system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without creative work.

[0179] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0180] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present invention. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the scope of the present invention. Therefore, the present invention will not be limited to the embodiments shown herein, but rather to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A recommendation method based on random walk algorithm and graph theory, characterized in that: The method comprises: Collecting credit data of multiple users and characteristic data of multiple financial products, as well as behavioral data of multiple users with respect to multiple financial products; Constructing a bipartite graph based on the credit data, the feature data and the behavior data; According to the bipartite graph, the rating data of the plurality of users on the plurality of financial products are obtained to construct a rating matrix, and a user similarity matrix and a product similarity matrix are constructed according to the rating matrix; Performing Laplace regularization processing on the user similarity matrix and the product similarity matrix to obtain a target user similarity matrix and a target product similarity matrix; Setting the initial step size, number of iterations and stopping rule corresponding to the random walk algorithm based on the target user similarity matrix and the target product similarity matrix, and executing the random walk algorithm to extract the user access path; The probability of a target user's interest in a target financial product is calculated based on the behavior data and the user access path, and a recommendation list containing multiple target financial products is generated based on the interest probability; the target user is any of the users; and the target financial product is a financial product with which there is no interaction between the target user and the target user.

2. The method according to claim 1, characterized in that: Also includes: Collect feedback data of the target user on the target financial product in the recommendation list, and optimize the parameters of the random walk algorithm.

3. The method according to claim 1, characterized in that The constructing of a user similarity matrix and a product similarity matrix based on the bipartite graph includes: Obtaining ratings of any two of the users on the plurality of financial products according to the bipartite graph; Calculating the sum of the products of the scores of the two users for the financial product scored together; Based on the score and the sum of the score products, calculating the cosine similarity between the two users to construct a user similarity matrix; Obtaining, according to the bipartite graph, a rating given by any user to any two of the financial products; The cosine similarity between the two financial products is calculated based on the scores between the two financial products to construct a product similarity matrix.

4. The method according to claim 1, characterized in that The executing the random walk algorithm to extract the user access path includes: Execute the random walk algorithm according to the initial step size, the number of iterations and the stopping rule; For each iteration, a vector is randomly generated; the user's access position is determined according to the vector, and the user's new access position is calculated according to the initial step length; A transfer matrix is ​​constructed based on all access locations to obtain the user access path.

5. The method according to claim 1, characterized in that The calculating the probability of the target user's interest in the target financial product according to the behavior data and the user access path includes: Acquire a plurality of the financial products corresponding to the user access path and the quantity of the financial products; Calculating the cosine similarity between each of the financial products and the target financial product based on the quantity; Acquire target behavior data of a target user for a plurality of the financial products from the behavior data; A plurality of financial products corresponding to the target behavior data are obtained, and a weighted average calculation is performed based on the cosine similarity between each of the financial products and the target financial product to obtain the probability of the target user's interest in the target financial product.

6. A recommendation device based on random walk algorithm and graph theory, characterized in that: The device comprises: A collection unit, used to collect credit data of multiple users and feature data of multiple financial products, as well as behavior data of multiple users with respect to multiple financial products; A first construction unit, configured to construct a bipartite graph according to the credit data, the feature data and the behavior data; A second construction unit is used to obtain rating data of the plurality of users on the plurality of financial products according to the bipartite graph, construct a rating matrix, and construct a user similarity matrix and a product similarity matrix according to the rating matrix; A processing unit, used for performing Laplace regularization processing on the user similarity matrix and the product similarity matrix to obtain a target user similarity matrix and a target product similarity matrix; An extraction unit, used to set an initial step size, a number of iterations and a stop rule corresponding to a random walk algorithm based on the target user similarity matrix and the target product similarity matrix, and execute the random walk algorithm to extract user access paths; A generating unit is used to calculate the probability of a target user's interest in a target financial product based on the behavior data and the user access path, and to generate a recommendation list containing multiple target financial products based on the interest probability; the target user is any of the users; and the target financial product is a financial product with which there is no interaction between the target user and the target user.

7. The device according to claim 6, characterized in that Also includes: An optimization unit is used to collect feedback data from the target user on the target financial product in the recommendation list, and optimize the parameters of the random walk algorithm.

8. The device according to claim 6, characterized in that The second building unit comprises: A first scoring acquisition module is used to obtain scores of any two of the users on the plurality of financial products according to the bipartite graph; A calculation module, used for calculating the sum of the products of the scores of the two users for the financial product scored together; A first cosine similarity calculation module, used for calculating the cosine similarity between two users based on the score and the sum of the score products, so as to construct a user similarity matrix; A second scoring acquisition module is used to obtain a score between any two of the financial products given by any one of the users according to the bipartite graph; The second cosine similarity calculation module is used to calculate the cosine similarity between the two financial products based on the scores between the two financial products to construct a product similarity matrix.

9. The device according to claim 6, characterized in that The extraction unit is specifically used for: The random walk algorithm is executed according to the initial step size, the number of iterations and the stopping rule; a vector is randomly generated for each iteration; the user's access position is determined according to the vector, and the user's new access position is calculated according to the initial step size; a transfer matrix is ​​constructed according to all access positions to obtain the user's access path.

10. The device according to claim 6, characterized in that The generating unit comprises: A first acquisition module, configured to acquire a plurality of the financial products corresponding to the user access path and the quantity of the financial products; A first calculation module, used to calculate the cosine similarity between each of the financial products and the target financial product based on the quantity; A second acquisition module is used to acquire target behavior data of a target user for a plurality of the financial products from the behavior data; The second calculation module is used to obtain multiple financial products corresponding to the target behavior data, and perform weighted average calculation based on the cosine similarity between each of the financial products and the target financial product to obtain the probability of the target user's interest in the target financial product.

Images