Personalized preferential commodity recommendation method and system

Abstract

The invention provides a preferential commodity personalized recommendation method and system, and relates to the technical field of data processing, and the method comprises the steps: obtaining historical commodity browsing data of a target user; establishing a weighted undirected graph reflecting the interaction relationship between the target user and each browsed commodity; calculating a correlation score between the target user and each browsed commodity in the weighted undirected graph through a restart random walk algorithm with a dynamic restart probability; calculating the similarity between the commodity characteristics of each browsed commodity through the Euclidean distance, and determining a bridging score in an inverse proportion relationship with the similarity; determining a recommendation score of each browsed commodity according to the correlation score and the bridging score of the target user and each browsed commodity; calculating commodity feature similarity between each preferential commodity and the browsed commodity with the highest recommendation score through a mahalanobis distance; and pushing the preferential commodities to the target user according to the mahalanobis distance from small to large. And the diversity of recommendation results is improved while the recommended preferential commodities are ensured to accord with user preferences.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and system for personalized recommendation of discounted products. Background Art

[0002] Product recommendation is a technology that uses algorithms and user data to recommend products that users may be interested in. It helps users discover products of potential interest and improves their shopping experience by analyzing users' browsing, clicking, and purchasing behaviors, combined with product attributes and other users' behaviors.

[0003] Product recommendation plays a key role in connecting users and products in a massive amount of product information, solving the problem of users having difficulty choosing in an information overload environment. It not only improves user experience and helps users quickly find the products they need, but also increases platform sales, attracts users to buy through personalized recommendations, and optimizes the platform's revenue structure.

[0004] However, existing product recommendation methods are overly limited to users' historical purchase records, resulting in over-recommendations of historically purchased products or products close to historically purchased products. This causes the recommended content to be too concentrated and lacks diversity, which may cause users to lose interest in repeated recommendations, thereby reducing their stickiness to the platform. Summary of the invention

[0005] In order to solve the technical problem that the commodity recommendation method in the prior art is overly limited to the user's historical purchase records, resulting in excessive recommendation of historically purchased commodities and commodities close to historically purchased commodities, resulting in over-concentration of recommended content and lack of diversity, which may cause users to lose interest in repeated recommendations and thus reduce stickiness to the platform, the present invention provides a personalized recommendation method and system for discounted products.

[0006] The technical solution provided by the embodiment of the present invention is as follows:

[0007] First aspect

[0008] An embodiment of the present invention provides a personalized recommendation method for discounted products, comprising:

[0009] S1: Obtain historical product browsing data of a target user, wherein the historical product browsing data includes multiple browsed products, product features of each browsed product, browsing time of each browsed product, and number of clicks on each browsed product;

[0010] S2: Based on the historical product browsing data, a weighted undirected graph is established to reflect the interactive relationship between the target user and each browsed product, wherein the nodes of the weighted undirected graph include the target user node and the browsed product node, the edge between each node in the weighted undirected graph is the connection between the target user and each browsed product, and the weight of each edge is the preference degree obtained based on the number of clicks and browsing time of the target user on each browsed product;

[0011] S3: Based on the preference degree, the relevance score between the target user and each browsed product in the weighted undirected graph is calculated by a restart random walk algorithm with dynamic restart probability;

[0012] S4: Calculate the correlation between the product features of each browsed product through the Euclidean distance, and determine a bridge score that is in inverse proportion to the correlation, wherein the bridge score reflects the span of the target user's preference for different browsed products;

[0013] S5: Determine the recommendation score of each browsed product according to the relevance score and bridge score between the target user and each browsed product;

[0014] S6: Calculate the similarity of product features between each discounted product and the browsed product with the highest recommendation score through Mahalanobis distance;

[0015] S7: Push the discounted products to the target users in descending order of Mahalanobis distance.

[0016] Second aspect

[0017] An embodiment of the present invention provides a personalized recommendation system for discounted products, including:

[0018] processor;

[0019] A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method for personalized recommendation of discounted products as described in the first aspect is implemented.

[0020] The third aspect

[0021] An embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the program is executed by a processor, the method for personalized recommendation of discounted products as described in the first aspect is implemented.

[0022] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0023] In the present invention, the browsing commodity information of the target user that is not limited to the purchased commodities is obtained, and the potential interests and real needs of the user can be fully captured, thereby improving the coverage and personalization of the recommendation and meeting the diverse preferences of the user. A weighted undirected graph is established based on the data in combination with the number of clicks and the browsing time to capture the user's interest preference, and the weight is used to reflect the preference degree of the number of clicks and the browsing time, accurately capturing the user's interest and improving the personalization and accuracy of the recommendation. Then, based on the preference degree, the relevance score between the target user and each browsing commodity in the weighted undirected graph is calculated by a restart random walk algorithm with a dynamic restart probability, and then the bridge score reflecting the span of the preference degree of the target user for different browsing commodities is calculated by Euclidean distance. Under the assumption that each feature is independent of each other, the correlation between the commodity features of each browsing commodity is calculated, so that the difference of each feature is amplified separately, which helps to amplify the span of the preference degree and helps to emphasize the diversity of the user's browsing commodities in the recommendation process. The preference span is amplified in combination with the Euclidean distance, which can accurately capture the difference in user's interest in different commodities, emphasize the diversity of the recommended content, avoid single recommendation, and meet the needs of user diversity and potential interests. Then, the recommendation score of each browsed product is determined based on the relevance score and the bridge score. Finally, the Mahalanobis distance that takes into account the feature relevance is used to calculate the similarity between the current discounted product to be recommended and the browsed product with the highest recommendation score, and the similarity is recommended to the target user. This can effectively screen products that meet user preferences and ensure the diversity and accuracy of recommendation results by emphasizing feature relevance. It effectively captures user preferences, ensures that recommendations meet user preferences, and improves the diversity of recommendation results, customer satisfaction and customer stickiness. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 A flowchart of a personalized recommendation method for discounted products provided by an embodiment of the present invention;

[0026] Figure 2 A schematic diagram of the structure of a personalized recommendation system for discounted products provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0027] The technical solution of the present invention is described below in conjunction with the accompanying drawings.

[0028] In the embodiments of the present invention, words such as "exemplarily" and "for example" are used to indicate examples, illustrations or explanations. Any embodiment or design described as "example" in the present invention should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of the word "example" is intended to present the concept in a specific way. In addition, in the embodiments of the present invention, the meaning expressed by "and / or" can be both, or it can be either of the two.

[0029] In order to make the technical problems, technical solutions and advantages to be solved by the present invention more clear, a detailed description will be given below with reference to the accompanying drawings and specific embodiments.

[0030] Reference Manual Attached Figure 1 , showing a flow chart of a method for personalized recommendation of discounted products provided in an embodiment of the present invention.

[0031] The embodiment of the present invention provides a method for personalized recommendation of discounted products, which can be implemented by a personalized recommendation device for discounted products, which can be a terminal or a server. The processing flow of the method for personalized recommendation of discounted products may include the following steps:

[0032] S1: Obtain the historical product browsing data of the target user.

[0033] The historical product browsing data includes multiple browsed products, product features of each browsed product, browsing time of each browsed product, and number of clicks on each browsed product.

[0034] It should be noted that by obtaining the user's historical browsing data, including product features, browsing time and click counts, the user's interaction with the product is fully recorded. By covering browsing behavior rather than just purchase records, it is possible to capture the user's potential interests and needs, expand the coverage of recommendations, and improve the accuracy of personalized recommendations and user experience. Specifically, this target user is a single user, that is, an individual who needs to be recommended a discounted product.

[0035] S2: Based on historical product browsing data, a weighted undirected graph is established to reflect the interactive relationship between the target user and each browsed product.

[0036] Among them, the nodes of the weighted undirected graph include target user nodes and browsed product nodes. The edges between the nodes in the weighted undirected graph are the connections between the target user and the browsed products. The weight of each edge is the preference degree based on the number of clicks and browsing time of the target user on each browsed product.

[0037] Among them, a weighted undirected graph is a graph structure in which nodes represent objects (such as users and products), edges represent the relationship between them, and the weight of the edge is used to quantify the strength or importance of the relationship. A weighted undirected graph is constructed based on the user's historical browsing data. The target user node and the browsed product node are connected by edges. The weight of the edge is calculated based on the user's click count and browsing time, reflecting the user's preference for each product. This process can transform the complex interaction relationship between users and products into a mathematical graph model, providing structured basic data for subsequent recommendation algorithms.

[0038] In a possible implementation, the weighted undirected graph is specifically:

[0039] G=(V,E,A)

[0040] A=(a ij )

[0041]

[0042] Where V represents the node set including the target user node and the browsed product node, E represents the edge set including the browsing time of each purchased product by the target user, A represents the target user's preference matrix for each browsed product, and a ij Indicates the preference of target user i for browsing product j, l ij represents the browsing time of target user i for browsing product j, L represents the maximum browsing time of target user i for all browsing products, c ij represents the number of clicks of target user i on browsed product j, C represents the maximum number of clicks of target user i on all browsed products, α1 and α2 represent the browsing time weight and click number weight respectively, H(L) and H(C) represent the entropy of browsing time and click number respectively, φ 浏览商品 and φ 已购商品 They represent the browsed product set and the purchased product set respectively, and G represents a weighted undirected graph about V, E and W.

[0043] Specifically, the target user i may be any user who visits the sales platform, and j refers to the product that the user has browsed on the sales platform. Browsing a product means that the user clicks on the product link corresponding to the product.

[0044] It should be noted that, through the node and weight connection between the target user and the browsed product, the weight distribution of browsing time and number of clicks is comprehensively considered, and the normalization method is used to accurately characterize the user's preference intensity for different products. At the same time, combined with the edge weight and feature distribution information of the user's historical interaction behavior, the multi-dimensional characteristics of user interests can be effectively captured. In addition, by dynamically balancing the influence weights of clicks and browses, the flexibility and accuracy of preference modeling are ensured, providing a more structured and quantitative preference expression for subsequent recommendations. The advantage of the preference degree is that it combines the two behavioral characteristics of browsing time and number of clicks, and ensures that the contribution of different feature dimensions is reasonably measured by normalization. At the same time, when determining the weight, the weight is dynamically allocated through information entropy, which can not only balance the importance of features, but also adjust adaptively according to data characteristics, thereby improving the accuracy and flexibility of preference expression.

[0045] Specifically, Among them, c i represents the number of clicks on the i-th browsed product, log represents the logarithmic function, Indicates c i The probability distribution of Represents the total number of clicks for all viewed products. Among them, l i represents the browsing time of the i-th browsed product, It represents the total browsing time of all browsed products. Optionally, α1 and α2 can be directly set to 0.5 respectively.

[0046] S3: Based on the preference degree, the relevance score between the target user and each browsed item in the weighted undirected graph is calculated by a restart random walk algorithm with dynamic restart probability.

[0047] Among them, the dynamic restart probability means that during the random walk, the probability of returning to the starting point (target user node) at each step is dynamically changing, rather than a fixed value. It is adjusted according to the current walk state, allowing the system to flexibly balance local exploration and global exploration. The dynamic restart probability can improve the diversity and accuracy of recommendations. The restart random walk algorithm is a method for calculating node relevance in a graph structure. It starts from the target node (user node), and each step has a certain probability of walking along the edge to the neighboring node, and at the same time has a certain probability of returning to the starting point. Through repeated walks, the algorithm can capture the multi-level relationship between the user and other nodes and output a relevance score. The relevance score represents the strength of the association between the target user node and each product node. It comprehensively considers the user's direct interest in the product (such as clicks, browsing time) and indirect relationships (such as interest in other similar products) to measure the attractiveness of the product to the user.

[0048] Through the dynamic restart random walk algorithm, weighted undirected graphs and dynamic restart probability, the relevance score between the target user and each product is calculated on the basis of balancing local interest and potential interest exploration. This process can not only highlight the products of the user's current interest, but also explore diverse potential preferences, providing basic support for accurate and diverse recommendations.

[0049] In a possible implementation, the correlation score is calculated as follows:

[0050]

[0051] Among them, h q represents the relevance score between the target user and the qth browsed item, p re represents the restart probability, e q represents the unit vector describing the position of the qth browsed product in the weighted undirected graph, A represents the normalized preference matrix, T represents the current number of walking steps, T0 represents the expected number of walking steps, β represents the adjustment coefficient, e represents the base of the natural logarithm, and D represents the degree matrix, where the diagonal elements of the degree matrix are the degrees of the nodes.

[0052] The degree of a node specifically refers to the sum of the weights of all edges associated with the node.

[0053] Among them, the unit vector e q It is a standard indicator vector, which is used to "select" the position of the qth browsed product node in the weighted undirected graph, that is, to locate the position of the qth node (browsed product) in the weighted undirected graph.

[0054] Among them, the expected number of walking steps indicates that the system expects to complete local exploration within the expected number of walking steps. If T<T0 and the walking has not reached the expected number of steps, the system maintains a high restart probability. If T>T0 and the walking reaches the expected number of steps, the restart probability begins to decrease. Control the switching timing between local and global exploration in random walks to optimize the personalization and diversity performance of the recommendation system. Optionally, the expected number of walking steps can be set to 1000, and the adjustment coefficient can be set to a value between 0.1 and 1, such as 0.5.

[0055] It should be noted that the dynamic restart probability and random walk calculation relevance score are combined, and the restart probability is adjusted according to the number of walk steps to balance local and global exploration, so as to ensure that the system prioritizes the user's current interests in the initial stage, and then gradually expands the scope to explore potential interests, thereby improving the diversity and accuracy of recommendations. The use of normalized preference matrices and degree matrices can effectively eliminate the impact caused by differences in the number of connections or weight scales between different nodes, so that the interaction between users and products can be quantified more fairly. This normalization process ensures that the degree of preference can reflect the true intensity of interest during the calculation process, while avoiding excessive bias of highly connected nodes, thereby optimizing the preference modeling of the recommendation system and improving the personalization effect of recommendations and the fairness of results.

[0056] S4: Calculate the similarity between the product features of each browsed product through the Euclidean distance, and determine a bridging score that is inversely proportional to the similarity.

[0057] Among them, the bridge score reflects the target user's preference span for different browsed products.

[0058] Among them, Euclidean distance is a mathematical tool for measuring the straight-line distance between two points (here, product feature vectors). The bridge score is a score calculated inversely proportional to the similarity of product features, which is used to measure the target user's preference span for products of different categories. The larger the span, the more interested the user is in cross-category products, and the system will give priority to recommending diversified products. The smaller the span, the more concentrated the user's interest is, and the bridge score helps avoid too single recommendation results. The Euclidean distance is used to quantify the differences in the feature dimensions of browsed products. The similarity between the product features of each browsed product is calculated under the assumption that each feature is independent of each other, so that the difference of each feature is amplified separately, which helps to amplify the preference span and helps to emphasize the diversity of user browsed products in the recommendation process. The similarity between products is calculated, and the bridge score is obtained based on the inverse proportional relationship, reflecting the span of user preference between product categories. The bridge score improves diversity in recommendations, avoids too concentrated recommendation results, and finds a balance between preference span and accuracy, providing users with recommendation results that are both in line with their interests and novel.

[0059] In a possible implementation, the commodity characteristics include commodity price, commodity production date, and commodity type.

[0060] In a possible implementation, S4 specifically includes:

[0061] S401: Calculate the similarity between the product features of each browsed product:

[0062]

[0063] Among them, Ri [j,k] represents the similarity between the jth and kth browsed items browsed by the i-th target user, f j and f k denote the product feature vectors of the jth and kth browsed products respectively, σ denotes the Gaussian kernel bandwidth parameter, It means calculating the square of Euclidean distance, and exp means the natural exponential function.

[0064] S402: Calculate the similarity mean:

[0065]

[0066] Among them, s i represents the total number of browsed items browsed by the i-th target user, represents the total relevance score between the browsed items browsed by the i-th target user, Represents the mean similarity between the browsed items browsed by the i-th target user.

[0067] S403: Determine a bridging score based on the similarity mean:

[0068]

[0069] Among them, b i The bridging scores between the browsed items browsed by the ith target user reflect the span of the ith target user's preference for different browsed items.

[0070] The bridge score reflects the feature span between the target user's browsed products. The larger the bridge score, the larger the span between the target user's browsed products, that is, the weaker the correlation. The higher the bridge score, the more likely it is to recommend products of different relevance to the user, thus recommending more diverse results. For example, identify users who are connected to different product categories and recommend multi-category products that they are interested in.

[0071] It should be noted that by calculating the feature similarity between browsed products and combining it with the Gaussian kernel function, the feature span of user-preferred products can be quantified. The larger the bridging score, the greater the diversity of the products browsed by users. The system can make recommendations for products in different categories to improve the diversity of recommendation results. At the same time, the calculation of the mean similarity smoothes the impact of individual abnormal product features, ensuring the stability of the bridging score, and providing an effective quantitative indicator for the recommendation system to find a balance between accuracy and diversity.

[0072] S5: Determine the recommendation score of each browsed product according to the relevance score and the bridge score between the target user and each browsed product.

[0073] It should be noted that the recommendation score of each browsed product is calculated by combining the relevance score and bridge score between the target user and the browsed product. The relevance score reflects the user's current interest, and the bridge score measures the preference span. The combination of the two ensures that the recommendation is both accurate and diverse, meeting the balance between user personalized needs and exploring potential interests.

[0074] In a possible implementation, the recommendation score is calculated as follows:

[0075] t q =h q ×b i

[0076] Among them, t q represents the recommendation score of the qth browsed product, h q represents the relevance score between the target user and the qth browsed product, b i The bridging scores between the browsed items browsed by the ith target user reflect the span of the ith target user's preference for different browsed items.

[0077] It should be noted that the combination of the relevance score and the bridge score between users and products not only reflects the user's direct preference for products, but also takes into account the user's preference span for browsing products, ensuring that the recommendations are both accurate and diverse. In this way, the system can balance the mining of users' current interests and potential interests, improve the personalization and diversity of recommended content, and meet users' multi-dimensional needs.

[0078] S6: Calculate the similarity of product features between each discounted product and the browsed product with the highest recommendation score through Mahalanobis distance.

[0079] Among them, Mahalanobis distance is a distance measurement method that measures the similarity between different features and can take into account the correlation between features. Mahalanobis distance normalizes the features through the inverse matrix of the covariance matrix, so that the scale and correlation of different features are effectively processed. Using Mahalanobis distance, the similarity of candidate products is accurately calculated by combining the correlation and global distribution of product features. There is usually a correlation between product features (for example, the price range of certain types of products is relatively concentrated, or the production date and product category have a dependency relationship). Mahalanobis distance captures the potential relationship between product features through the covariance matrix, making the similarity calculation more accurate. The goal is to calculate the exact similarity between candidate discounted products and users' highly rated products to provide high-quality matches for recommendation results.

[0080] In a possible implementation, the method for calculating the similarity of product features is specifically as follows:

[0081]

[0082] Among them, fv represents the feature vector of the vth discounted product, f q1 represents the feature vector of the browsed product with the highest recommendation score. The subscript T represents transposition. Σ -1 represents the inverse of the covariance matrix between the discounted product features and the features of the browsed product features, d M (f v ,f q1 ) indicates f v With f q1 The Mahalanobis distance between them is the similarity of product features between the vth discounted product and the browsed product with the highest recommendation score.

[0083] Among them, the characteristic dimensions of the discounted product feature vector are consistent with the characteristic dimensions of the commodity feature vector, that is, they both include three characteristic dimensions: commodity price, commodity production date and commodity type.

[0084] It should be noted that the use of Mahalanobis distance, combined with the inverse matrix of the covariance matrix, effectively considers the correlation and distribution characteristics between different features, avoiding the problem of simple distance calculation ignoring the dependency between features. By normalizing the features, the contribution of features of different scales to the similarity calculation is balanced, which improves the accuracy of feature similarity calculation, thereby ensuring a high match between the recommended products and the target products in the multi-dimensional feature space.

[0085] S7: Push the discounted products to the target users in the order of Mahalanobis distance from small to large.

[0086] In a possible implementation manner, after S7, the method further includes:

[0087] Update historical product browsing data at preset intervals.

[0088] It is understandable that updating historical product browsing data at preset intervals can dynamically capture changes in user interests, ensure that the recommendation system always makes personalized recommendations based on the latest user behavior, and improve the real-time and accuracy of recommendation results.

[0089] It should be noted that those skilled in the art can set the preset duration according to actual needs, and the present invention is not limited thereto.

[0090] In the actual application process, first, the historical browsing data of the target user is obtained, including product features, clicks and browsing time, to fully capture the user's interactive behavior. Then, a weighted undirected graph is constructed based on this data, and the preference relationship between users and products is quantified using the weights of nodes and edges. Subsequently, the relevance score between the target user and each product is calculated by dynamically restarting the random walk algorithm to balance the mining of local interests and potential preferences. Then, the similarity of product features is calculated using the Euclidean distance to generate a bridging score, which amplifies the user's preference span and improves the recommendation diversity. Next, the recommendation score of each browsed product is determined by combining the relevance score and the bridging score. After that, the Mahalanobis distance is used to calculate the similarity between the discounted product and the product with the highest recommendation score to capture the potential association between product features. Finally, the discounted products are sorted from high to low by similarity and pushed to users to achieve accurate and diverse recommendation results.

[0091] The beneficial effects brought about by the technical solution provided by the embodiment of the present invention include at least:

[0092] In the present invention, the browsing commodity information of the target user that is not limited to the purchased commodity is obtained, and the potential interests and real needs of the user can be fully captured, thereby improving the coverage and personalization of the recommendation and meeting the diverse preferences of the user. A weighted undirected graph is established based on the data in combination with the number of clicks and the browsing time to capture the user's interest preference, and the weight is used to reflect the preference degree of the number of clicks and the browsing time, accurately capturing the user's interest and improving the personalization and accuracy of the recommendation. After that, based on the preference degree, the relevance score between the target user and each browsing commodity in the weighted undirected graph is calculated by a restart random walk algorithm with a dynamic restart probability, and then the bridge score reflecting the span of the target user's preference degree for different browsing commodities is calculated by Euclidean distance. Under the assumption that each feature is independent of each other, the similarity between the commodity features of each browsing commodity is calculated, so that the difference of each feature is amplified separately, which helps to amplify the span of the preference degree and helps to emphasize the diversity of the user's browsing commodities in the recommendation process. The preference span is amplified in combination with the Euclidean distance, which can accurately capture the difference in user's interest in different commodities, emphasize the diversity of the recommended content, avoid single recommendation, and meet the needs of user diversity and potential interests. Then, the recommendation score of each browsed product is determined based on the relevance score and the bridge score. Finally, the Mahalanobis distance that takes into account the feature relevance is used to calculate the similarity between the current discounted product to be recommended and the browsed product with the highest recommendation score, and the similarity is recommended to the target user. This can effectively screen products that meet user preferences and ensure the diversity and accuracy of recommendation results by emphasizing feature relevance. It effectively captures user preferences, ensures that recommendations meet user preferences, and improves the diversity of recommendation results, customer satisfaction and customer stickiness.

[0093] Reference Manual Attached Figure 2, showing a structural schematic diagram of a personalized recommendation system for discounted products provided by the present invention.

[0094] The present invention further provides a personalized recommendation system 20 for discounted products, which is applied to the above-mentioned personalized recommendation method for discounted products, and comprises:

[0095] Processor 201.

[0096] The memory 202 stores computer-readable instructions. When the computer-readable instructions are executed by the processor 201, the method for personalized recommendation of discounted products in the method embodiment is implemented.

[0097] The personalized recommendation system 20 for discounted products provided by the present invention can execute the above-mentioned personalized recommendation method for discounted products and achieve the same or similar technical effects. To avoid repetition, the present invention will not elaborate on them.

[0098] It should be understood that the processor in the embodiment of the present invention may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0099] It should also be understood that the memory in the embodiments of the present invention may be a volatile memory or a non-volatile memory, or may include both volatile and non-volatile memories. Among them, the non-volatile memory may be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM), or a flash memory. The volatile memory may be a random access memory (RAM), which is used as an external cache. By way of example and not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic random access memory (DRAM), synchronous DRAM (SDRAM), double data rate synchronous dynamic random access memory (DDR SDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous link DRAM (SLDRAM), and direct rambus RAM (DR RAM).

[0100] The above embodiments can be implemented in whole or in part by software, hardware (such as circuits), firmware or any other combination. When implemented by software, the above embodiments can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present invention is generated in whole or in part. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions can be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired (such as infrared, wireless, microwave, etc.). The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server or data center that contains one or more available media sets. The available medium can be a magnetic medium (for example, a floppy disk, a hard disk, a tape), an optical medium (for example, a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state hard disk.

[0101] It should be understood that the term "and / or" in this article is only a description of the association relationship of associated objects, indicating that there can be three relationships. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. A and B can be singular or plural. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship, but it may also indicate an "and / or" relationship. Please refer to the context for specific understanding.

[0102] In the present invention, "at least one" means one or more, and "more than one" means two or more. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b, or c can be represented by: a, b, c, ab, ac, bc, or abc, where a, b, c can be single or multiple.

[0103] It should be understood that in various embodiments of the present invention, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0104] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can 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.

[0105] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described equipment, devices and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0106] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another device, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0107] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0108] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0109] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention can be essentially or partly embodied in the form of a software product that contributes to the prior art. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media that can store program codes, such as a USB flash drive, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk.

[0110] An embodiment of the present invention provides a computer-readable storage medium having a computer program stored thereon. When the program is executed by a processor, the method for personalized recommendation of discounted products as described in the method embodiment is implemented.

[0111] A computer-readable storage medium provided by the present invention can implement the steps and effects of the personalized recommendation method for discounted products of the above method embodiment. To avoid repetition, the present invention will not go into details.

[0112] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions within the technical scope disclosed by the present invention, which should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.

[0113] There are a few points to note:

[0114] (1) The drawings of the embodiments of the present invention only relate to the structures related to the embodiments of the present invention, and other structures may refer to the general design.

[0115] (2) For the sake of clarity, in the drawings used to describe the embodiments of the present invention, the thickness of the layers or regions is exaggerated or reduced, that is, these drawings are not drawn according to the actual scale. It is understood that when an element such as a layer, film, region or substrate is referred to as being "on" or "under" another element, the element may be "directly" "on" or "under" the other element or there may be intermediate elements.

[0116] (3) In the absence of conflict, the embodiments of the present invention and the features therein may be combined with each other to obtain new embodiments.

[0117] The above are only specific embodiments of the present invention, but the protection scope of the present invention is not limited thereto. The protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A personalized recommendation method for discounted products, characterized in that: include: S1: Acquire historical product browsing data of a target user, wherein the historical product browsing data includes multiple browsed products, product features of each browsed product, browsing time of each browsed product, and number of clicks on each browsed product; S2: establishing a weighted undirected graph reflecting the interactive relationship between the target user and each browsed product based on the historical product browsing data, wherein the nodes of the weighted undirected graph include target user nodes and browsed product nodes, the edges between the nodes in the weighted undirected graph are connections between the target user and each browsed product, and the weight of each edge is the preference degree obtained based on the number of clicks and browsing time of the target user on each browsed product; S3: Based on the preference degree, calculating the relevance score between the target user and each browsed product in the weighted undirected graph by using a restart random walk algorithm with a dynamic restart probability; S4: calculating the similarity between the product features of each browsed product through the Euclidean distance, and determining a bridging score that is in inverse proportion to the similarity, wherein the bridging score reflects the span of the target user's preference for different browsed products; S5: determining a recommendation score for each browsed product according to the relevance score between the target user and each browsed product and the bridge score; S6: Calculate the similarity of product features between each discounted product and the browsed product with the highest recommendation score through Mahalanobis distance; S7: Pushing the discounted products to the target users in the order of the Mahalanobis distance from small to large.

2. The method for personalized recommendation of discounted products according to claim 1, characterized in that: The weighted undirected graph is specifically: G=(V,E,A) A=(a ij ) Where V represents the node set including the target user node and the browsed product node, E represents the edge set including the browsing time of each purchased product by the target user, A represents the target user's preference matrix for each browsed product, and a ij Indicates the preference of target user i for browsing product j, l ij represents the browsing time of target user i for browsing product j, L represents the maximum browsing time of target user i for all browsing products, c ij represents the number of clicks of target user i on browsed product j, C represents the maximum number of clicks of target user i on all browsed products, α1 and α2 represent the browsing time weight and click number weight respectively, H(L) and H(C) represent the entropy of browsing time and click number respectively, φ 浏览商品 and φ 已购商品 They represent the browsed product set and the purchased product set respectively, and G represents a weighted undirected graph about V, E and W.

3. The method for personalized recommendation of discounted products according to claim 2, characterized in that: The calculation method of the relevance score is specifically as follows: Among them, h q represents the relevance score between the target user and the qth browsed item, p re represents the restart probability, e q represents the unit vector describing the position of the qth browsed product in the weighted undirected graph, represents the normalized preference matrix, T represents the current number of steps, T0 represents the expected number of steps, β represents the adjustment coefficient, e represents the base of the natural logarithm, and D represents the degree matrix, where the diagonal elements of the degree matrix are the degrees of the nodes.

4. The method for personalized recommendation of discounted products according to claim 1, characterized in that: The product characteristics include product price, product production date and product type.

5. The method for personalized recommendation of discounted products according to claim 1, characterized in that: The S4 specifically includes: S401: Calculate the similarity between the product features of each browsed product: Among them, R i [j,k] represents the similarity between the jth and kth browsed items browsed by the i-th target user, f j and f k denote the product feature vectors of the jth and kth browsed products respectively, σ denotes the Gaussian kernel bandwidth parameter, Indicates the calculation of the square of the Euclidean distance, and exp represents the natural exponential function; S402: Calculate the similarity mean: Among them, s i represents the total number of browsed items browsed by the i-th target user, represents the total relevance score between the browsed items browsed by the i-th target user, represents the mean similarity between the browsed items browsed by the i-th target user; S403: Determine the bridging score based on the similarity mean: Among them, b i The bridging scores between the browsed items browsed by the ith target user reflect the span of the ith target user's preference for different browsed items.

6. The method for personalized recommendation of discounted products according to claim 1, characterized in that: The calculation method of the recommendation score is specifically as follows: t q =h q ×b i Among them, t q represents the recommendation score of the qth browsed product, h q represents the relevance score between the target user and the qth browsed product, b i The bridging scores between the browsed items browsed by the ith target user reflect the span of the ith target user's preference for different browsed items.

7. The method for personalized recommendation of discounted products according to claim 1, characterized in that: The method for calculating the similarity of the product features is specifically as follows: Among them, f v represents the feature vector of the vth discounted product, f q1 represents the feature vector of the browsed product with the highest recommendation score. The subscript T represents transposition. Σ -1 represents the inverse of the covariance matrix between the discounted product features and the features of the browsed product features, d M (f v ,f q1 ) indicates f v With f q1 The Mahalanobis distance between them is the similarity of product features between the vth discounted product and the browsed product with the highest recommendation score.

8. The method for personalized recommendation of discounted products according to claim 1, characterized in that: After S7, the method further includes: The historical product browsing data is updated at preset time intervals.

9. A personalized recommendation system for discounted products, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method for personalized recommendation of discounted products as claimed in any one of claims 1 to 8 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method for personalized recommendation of discounted products as claimed in any one of claims 1 to 8 is implemented.

Images