A product recommendation method, system, device and medium

By performing feature encoding and incremental clustering of user historical data, user feature vectors are generated, and recommendation preference sets are grouped, the problem of mismatch between recommended products and user preferences in the prior art is solved, and the accuracy and conversion rate of product recommendations are improved. .

CN112070577BActive Publication Date: 2024-11-19SHENZHEN KANIU TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202010898544.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2020-08-31
Publication Date
2024-11-19
Estimated Expiration
2040-08-31

Smart Images

  • Figure CN112070577B_ABST
    Figure CN112070577B_ABST
Patent Text Reader

Abstract

The embodiment of the present invention discloses a method, system, device and medium for recommending products. The method includes: obtaining historical data of users, obtaining user feature data through preprocessing; encoding and feature concatenating the user feature data to obtain a first user feature vector; converting the first user feature vector into a second user feature vector of a preset length after passing through two fully connected layers; performing incremental clustering on the second user feature vector to divide the users corresponding to the second user feature vector into different clusters; obtaining a preference set of all users in each cluster; and recommending the preference set to all users in the corresponding cluster. By obtaining historical data of users to analyze user features, processing user features into feature vectors of a preset length for incremental clustering, obtaining preference sets of similar users and recommending them to relevant users, the problem of low conversion rate of recommended products is solved, and the accuracy and conversion rate of recommended products are improved.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The embodiments of the present invention relate to computer application technology, and more particularly to a product recommendation method, system, device and medium. Background Art

[0002] With the development of Internet technology, the Internet has become an indispensable part of people's lives. People's consumption, entertainment, learning, travel and financial management are all inseparable from the Internet.

[0003] At the same time, the variety of e-commerce products is becoming more and more abundant, promotional activities are emerging in an endless stream, and the amount of data and information generated is huge. When people need to obtain certain types of data, they can get the desired products or services through active searches or recommendations from related applications.

[0004] Users hope to get product recommendations that meet their preferences from a large amount of information. If the recommendations of the application cannot be well combined with the potential needs of users, people will have to spend a lot of time looking for information that meets their requirements in the face of excessive information, which will reduce the conversion rate of recommended products and reduce user stickiness. Inaccurate recommended products are often due to inaccurate user feature analysis and product recommendations are only based on products that users have purchased, and potential preferred products that users have not purchased cannot be obtained. At the same time, since new users have not purchased any products, accurate product recommendations cannot be made for new users. Summary of the invention

[0005] The present invention provides a commodity recommendation method, system, device and medium, so as to realize commodity recommendation according to user preference and improve the accuracy and conversion rate of recommended commodities.

[0006] In a first aspect, an embodiment of the present invention provides a product recommendation method, comprising:

[0007] Obtain the user's historical data and obtain the user's characteristic data through preprocessing;

[0008] Encoding the user characteristic data;

[0009] Performing feature concatenation processing on the encoded data to obtain a first user feature vector;

[0010] Converting the first user feature vector into a second user feature vector of a preset length after passing through two fully connected layers;

[0011] performing incremental clustering on the second user feature vector to classify users corresponding to the second user feature vector into different clusters;

[0012] Get the preference set of all users in each cluster;

[0013] The preference set is recommended to all users in the corresponding cluster.

[0014] Optionally, performing incremental clustering on the second user feature vector to classify users corresponding to the second user feature vector into different clusters includes:

[0015] Pre-clustering the second user feature vector using a clustering algorithm;

[0016] Save the pre-clustered cluster centers and class labels;

[0017] Obtain the feature vector newly added by the user, and calculate the distance to each of the cluster centers;

[0018] Assigning users corresponding to the newly added feature vectors to the nearest cluster centers;

[0019] Update the cluster center of each class.

[0020] Optionally, obtaining a preference set of users of each cluster includes:

[0021] Obtain the product information generated by each user in a cluster to form a preference pre-set;

[0022] Determining whether the number of commodities in the preference pre-set is greater than a preset value;

[0023] If the quantity of the commodities is greater than the preset quantity, the commodities in the preference pre-set are sorted, and a preset quantity of commodities are selected in sequence to form a preference set;

[0024] If the quantity of the commodities is not greater than the preset quantity, the preference pre-set is determined to be the preference set.

[0025] Optionally, the user feature data includes user dense features and user sparse features, and when encoding the user feature data, the encoding process includes:

[0026] The user dense features are processed by One-Hot encoding; the user sparse features are processed by Embedding encoding.

[0027] Optionally, the user's historical data includes, but is not limited to, device information, geographic location information, and demographic information, as well as product browsing, clicking, collection, and purchase information.

[0028] In a second aspect, an embodiment of the present invention further provides a product recommendation system, including:

[0029] User data acquisition module, used to obtain user historical data and obtain user feature data after preprocessing;

[0030] An encoding processing module, used for performing encoding processing and feature splicing processing on the user feature data to obtain a first user feature vector;

[0031] A fully connected layer processing module, used for converting the first user feature vector into a second user feature vector of a preset length after passing through two fully connected layers;

[0032] A clustering processing module, used for performing incremental clustering on the second user feature vector, and dividing the users corresponding to the second user feature vector into different clusters;

[0033] A preference acquisition module is used to obtain the preference set of users in each cluster;

[0034] The product recommendation module is used to recommend the preference set to all users in the corresponding cluster.

[0035] Optionally, the clustering processing module includes:

[0036] A pre-clustering submodule, used for pre-clustering the second user feature vector using a clustering algorithm;

[0037] The saving submodule is used to save the pre-clustering cluster centers and class labels;

[0038] A calculation submodule, used to obtain the feature vector newly added by the user and calculate the distance to each of the cluster centers;

[0039] A clustering submodule, used to group users corresponding to the newly added feature vectors into the nearest cluster centers;

[0040] The update submodule is used to update the cluster center of each class.

[0041] In a third aspect, an embodiment of the present invention further provides a product recommendation device, the product recommendation device comprising:

[0042] one or more processors;

[0043] a storage device for storing one or more programs,

[0044] When the one or more programs are executed by the one or more processors, the one or more processors implement the product recommendation method as described in any one of the first aspects.

[0045] In a fourth aspect, an embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the product recommendation method as described in any one of the first aspects.

[0046] The embodiment of the present invention obtains the user's historical data to analyze the user's characteristics, processes the user characteristics into a feature vector of a preset length for incremental clustering, obtains a preference set of similar users, and recommends them to related users. This solves the problem that the recommended products do not match the user's expected consumption preferences or the recommended products are relatively single, resulting in a low conversion rate for the recommended products, and the problem of being unable to make accurate product recommendations to new users. This improves the accuracy and conversion rate of recommended products, and achieves the effect of accurately recommending products to new users who have not purchased any products. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 A flowchart of a product recommendation method provided in Embodiment 1 of the present invention;

[0048] Figure 2 A schematic diagram of a sub-process of a product recommendation method provided in Embodiment 2 of the present invention;

[0049] Figure 3 A schematic diagram of a sub-process of another product recommendation method provided in Embodiment 2 of the present invention;

[0050] Figure 4A A schematic diagram of the structure of a product recommendation system provided in Embodiment 3 of the present invention;

[0051] Figure 4B A schematic diagram of the structure of a clustering processing module of a product recommendation system provided in Embodiment 3 of the present invention;

[0052] Figure 5 This is a structural schematic diagram of a product recommendation device provided in Example 4 of the present invention. DETAILED DESCRIPTION

[0053] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0054] Embodiment 1

[0055] Figure 1 This is a flow chart of a product recommendation method provided in Embodiment 1 of the present invention. This embodiment is applicable to the case where the product preferences of a user's consumption are obtained through user historical data and product recommendations are made. The method can be executed by a product recommendation system and specifically includes the following steps:

[0056] Step 110: Obtain the user's historical data and obtain user feature data through preprocessing.

[0057] The user's historical data is obtained through data embedding, where the user's historical data includes but is not limited to device information, geographic location information and demographic information, as well as information about browsing, clicking, collecting and purchasing of goods. Device information includes information about mobile terminals such as mobile phones and tablets that users have used to make purchases, such as models, etc. The geographical location is obtained through the mobile terminal used by the user, including the user's work and living place; demographic information includes the user's gender, age, occupation, education, etc.; product information includes product name, product category, product price, etc.; the above information can be pre-processed and analyzed to obtain user feature data.

[0058] Step 120: Encode and concatenate the user feature data to obtain a first user feature vector.

[0059] After obtaining the user feature data, it must be encoded to obtain the encoded characters that the system can process. Among them, user feature data includes user dense features and user sparse features; user dense features include traditional features such as user identity information, which are not particularly high in dimension; user sparse features include product information generated by users, such as the list of products that users have clicked and browsed, which are high in dimension and large in number.

[0060] The encoding methods of user feature data include but are not limited to the following three methods:

[0061] First, after word embedding processing, each user sparse feature is mapped into a vector. Different user sparse features are cross-operated to obtain cross features. The cross features and user dense features are concatenated to obtain the first user feature vector as the input of the fully connected layer.

[0062] Secondly, firstly, the user dense features are discretized and converted into discrete features, and then the discretized user dense features and user sparse features are word embedded respectively, and then the user dense features and user sparse features after word embedding are cross-linked and concatenated to obtain the first user feature vector as the input of the fully connected layer.

[0063] Third, each user dense feature is multiplied by the word embedding matrix for word embedding processing, and then multiplied by the weight vector to obtain a feature vector, which is equivalent to mapping a dense feature into a feature vector. The user sparse feature is then concatenated with the feature vector after word embedding processing to obtain the first user feature vector as the input of the fully connected layer.

[0064] Optionally, when encoding the user feature data, it includes: using One-Hot encoding for user dense features; using Embedding encoding for user sparse features; and then performing feature concatenation operations.

[0065] One-Hot encoding, also known as one-bit effective encoding, mainly uses N-bit state registers to encode N states, each state consists of independent register bits, and only one bit is valid at any time. One-Hot encoding is the representation of categorical variables as binary vectors. This first requires mapping categorical values ​​to integer values. Then, each integer value is represented as a binary vector, which is zero value except for the index of the integer, which is marked as 1. Using One-Hot encoding for discrete features will make the distance calculation between features more reasonable.

[0066] Embedding encoding is essentially a mapping from semantic space to vector space, while trying to maintain the relationship between the original samples in the semantic space in the vector space. For example, two words with similar semantics are also close in position in the vector space. Embedding encoding is used to reduce the dimensionality of a large number of user sparse features to a low-dimensional space.

[0067] Step 130: Convert the first user feature vector into a second user feature vector of a preset length after passing through two fully connected layers.

[0068] The concatenated first user feature vector is then provided to a deep network model for learning, and is converted into a second user feature vector of a preset length after passing through two fully connected layers. The number of network layers here can be set according to the effect without limitation. Experiments show that generally two fully connected layers can achieve good results.

[0069] Step 140: perform incremental clustering on the second user feature vector to classify users corresponding to the second user feature vector into different clusters.

[0070] The process of dividing a collection of physical or abstract objects into multiple classes consisting of similar objects is called clustering. The clusters generated by clustering are a collection of data objects that are similar to objects in the same cluster and different from objects in other clusters. Incremental clustering mainly regards incremental data as time series data or data in a specific order.

[0071] Incremental clustering is performed on the second user feature vector, specifically clustering the users corresponding to the second user feature vector into different clusters. When a new feature vector is added, the cluster center is kept unchanged, and the cluster to which the user belongs can be changed, that is, the relationship between the user and the cluster center can be changed.

[0072] Among them, new users have not purchased any products and cannot be classified according to the purchase preferences obtained from the browsing, clicking, collection and purchase information of products. However, based on device information, geographic location information and demographic information, clusters of other users with similar conditions such as occupation, income, living or working place to the new users can be obtained, and the new users can be divided into corresponding clusters.

[0073] Step 150: Obtain the preference set of all users in each cluster.

[0074] Obtain the preference set of all users in each cluster, and collect all the information of all products clicked or purchased by all users to form a preference set; for example, there are three users A, B, and C in a cluster, and the products purchased by A are hamburger A1 and French fries A2; the products purchased by B are ice cream B1 and fried chicken B2, and the product clicked and browsed is coupon B3; the products purchased by C are hamburger A1, fried chicken B2, and family bucket C1. Then the product preference set in this class is hamburger A1, French fries A2, ice cream B1, fried chicken B2, coupon B3, and family bucket C1.

[0075] Step 160: Recommend the preference set to all users in the corresponding cluster.

[0076] Users in a cluster have the same or similar preferences and are motivated to purchase products that other users in the cluster have purchased. Therefore, recommending the preference set "hamburger A1, French fries A2, ice cream B1, fried chicken B2, coupon B3, and family bucket C1" obtained by the above method to all users A, B, and C in the cluster corresponding to this preference set can improve the conversion rate of the recommended products.

[0077] The embodiment of the present invention obtains the user's historical data to analyze the user's characteristics, processes the user characteristics into a feature vector of a preset length for incremental clustering, obtains a preference set of similar users, and recommends them to related users. This solves the problem that the recommended products do not match the user's expected consumption preferences or the recommended products are relatively single, resulting in a low conversion rate for the recommended products, and the problem of being unable to make accurate product recommendations to new users. This improves the accuracy and conversion rate of recommended products, and achieves the effect of accurately recommending products to new users who have not purchased any products.

[0078] Embodiment 2

[0079] Figure 2 A schematic diagram of a sub-process of a product recommendation method provided in Embodiment 2 of the present invention. Figure 3 A schematic diagram of a sub-process of another product recommendation method provided in Embodiment 2 of the present invention.

[0080] On the basis of the above embodiment, incremental clustering is performed on the second user feature vector to classify users corresponding to the second user feature vector into different clusters, which specifically includes the following steps:

[0081] Step 141: Use a clustering algorithm to pre-cluster the second user feature vector.

[0082] The second user feature vector is processed by a clustering algorithm, such as the K-Means clustering algorithm. Specifically, the K-Means clustering algorithm is an iterative clustering analysis algorithm. The data is pre-divided into K groups, and K user features are randomly selected as the initial cluster centers. Then the distance between each user feature and each seed cluster center is calculated, and the user corresponding to each user feature is assigned to the cluster center closest to it. The cluster centers and the objects assigned to them represent a cluster. Each time a sample is assigned, the cluster center of the cluster is recalculated based on the existing user features in the cluster. This process will be repeated until a certain termination condition is met. The termination condition can be that no (or a minimum number) of user features are reassigned to different clusters, no (or a minimum number) of cluster centers change again, and the sum of squared errors is locally minimized.

[0083] Step 142: Save the pre-clustering cluster centers and class labels.

[0084] The cluster centers and class labels obtained by pre-clustering are saved, and the cluster centers are kept unchanged. The class labels are the same labels of the same cluster, and the meaning of the class labels remains unchanged for a long time, so that the system can better learn this feature.

[0085] Step 143: Obtain the feature vector newly added by the user, and calculate the distance to each cluster center.

[0086] At present, the research on incremental clustering mainly regards incremental data as time series data or data in a specific order. The technical solution of this embodiment uses the results of the previous clustering, and each time a new feature vector of the user is obtained, the distance between the new feature vector of the user and each cluster center is calculated.

[0087] Step 144: Assign users corresponding to the newly added feature vectors to the cluster centers with the closest distances.

[0088] According to the calculated user's newly added feature vector and the clustering of each cluster center, each user's newly added feature vector is divided into the existing clusters, that is, the user's newly added feature vector is assigned to the cluster with the center closest to it and the center is moved to the newly added feature vector. In other words, the user's newly added feature vector will not affect the original division. The advantage is that there is no need to re-cluster all the data every time.

[0089] Step 145: Update the cluster center of each class.

[0090] During the off-peak period of business, the current user feature vector data is re-clustered and the cluster center of each class is fully updated to eliminate the local deviation that may be introduced by incremental clustering, so as to improve the accuracy of the system.

[0091] Based on the above embodiment, obtaining the preference set of users in each cluster specifically includes the following steps:

[0092] Step 151: Obtain commodity information generated by each user of a cluster to form a preference pre-set.

[0093] Get the product information generated by each user in a cluster, including browsing, clicking, favorites and purchase information of the product. Not only the purchase information, but also the browsing, clicking, favorites and other information are obtained to analyze the purchase needs and expected consumption of users in the cluster and form a pre-set of product preferences for users in this cluster.

[0094] Step 152: Determine whether the quantity of commodities in the preference pre-set is greater than a preset quantity.

[0095] When there are many users in the cluster or the users have a lot of product information, too many product preference pre-sets are obtained, and recommending all products to all users will reduce the product recommendation conversion rate. It is necessary to determine whether the number of products in the preference pre-set is greater than the preset value. If the number of products is greater than the preset number, execute step 153; if the number of products is not greater than the preset number, execute step 154.

[0096] Step 153: sort the commodities in the preference pre-set, and select a preset number of commodities in sequence to form a preference set.

[0097] When the number of products in the product preference pre-set is greater than the preset number, the products in the preference pre-set are sorted in descending order according to the number of clicks and purchases, and a preset number of products are selected in turn to form a preference set. The preference set obtained in this way has a higher conversion rate for product recommendations.

[0098] Step 154: Determine the preference pre-set as the preference set.

[0099] When the number of products in the product preference pre-set is not greater than the preset number, the obtained preference pre-set is determined as the preference set. At this time, the number of products in the preference set is small and the user preferences in the cluster are more concentrated. Therefore, the conversion rate of product recommendations based on this preference set is higher.

[0100] Embodiment 3

[0101] Figure 4A This is a schematic diagram of the structure of a product recommendation system provided by Embodiment 3 of the present invention. Figure 4B This is a schematic diagram of the structure of a clustering processing module of a product recommendation system provided in Embodiment 3 of the present invention.

[0102] A product recommendation system provided by an embodiment of the present invention can execute a product recommendation method provided by any embodiment of the present invention, and has functional modules and beneficial effects corresponding to the execution method.

[0103] like Figure 4A As shown, a product recommendation system includes:

[0104] The user data acquisition module 410 is used to acquire the user's historical data and obtain the user's characteristic data through preprocessing;

[0105] Obtain the user's historical data, where the user's historical data includes but is not limited to device information, geographic location information and demographic information, as well as information about browsing, clicking, collecting and purchasing of goods. Device information includes information about mobile terminals such as mobile phones and tablets that users have used to make purchases, such as models, etc.; geographic location is obtained through the mobile terminal used by the user, including the user's work and living place; demographic information includes the user's gender, age, occupation, education, etc.; product information includes product name, product category, product price, etc.; pre-processing of the above information can be analyzed to obtain user feature data.

[0106] The encoding processing module 420 is used to perform encoding processing and feature concatenation processing on the user feature data to obtain a first user feature vector.

[0107] After obtaining the user feature data, it needs to be encoded to obtain the encoded characters that the system can process. Among them, the user feature data includes user dense features and user sparse features.

[0108] The encoding processing module 420 includes:

[0109] The encoding submodule (not shown) is used to encode the user dense features and the user sparse features respectively.

[0110] When encoding the user feature data, it includes: one-hot encoding for user dense features; embedding encoding for user sparse features; and then feature concatenation.

[0111] The splicing submodule (not shown) is used to perform feature splicing processing on the encoded data to obtain a first user feature vector.

[0112] The fully connected layer processing module 430 is used to convert the first user feature vector into a second user feature vector of a preset length after passing through two fully connected layers.

[0113] The concatenated first user feature vector is then provided to a deep network model for learning, and is converted into a second user feature vector of a preset length after passing through two fully connected layers. The number of network layers here can be set according to the effect without limitation. Experiments show that generally two fully connected layers can achieve good results.

[0114] The clustering processing module 440 is used to perform incremental clustering on the second user feature vector, and classify the users corresponding to the second user feature vector into different clusters.

[0115] The second user feature vector is incrementally clustered, specifically, the users corresponding to the second user feature vector are clustered into different clusters. When a new feature vector is added, the cluster center is kept unchanged, and the cluster to which the user belongs can be changed, that is, the relationship between the user and the cluster center can be changed.

[0116] Among them, new users have not purchased any products and cannot be classified according to the purchase preferences obtained from the browsing, clicking, collection and purchase information of products. However, based on device information, geographic location information and demographic information, clusters of other users with similar conditions such as occupation, income, living or working place to the new users can be obtained, and the new users can be divided into corresponding clusters.

[0117] A preference acquisition module 450 is used to acquire a preference set of users of each cluster;

[0118] Obtain the preference set of all users in each cluster, and collect all the information of all products clicked or purchased by all users to form a preference set; for example, there are three users A, B, and C in a cluster, and the products purchased by A are hamburger A1 and French fries A2; the products purchased by B are ice cream B1 and fried chicken B2, and the product clicked and browsed is coupon B3; the products purchased by C are hamburger A1, fried chicken B2, and family bucket C1. Then the product preference set in this class is hamburger A1, French fries A2, ice cream B1, fried chicken B2, coupon B3, and family bucket C1.

[0119] The product recommendation module 460 is used to recommend the preference set to all users in the corresponding cluster.

[0120] Users in a cluster have the same or similar preferences and are motivated to purchase products that other users in the cluster have purchased. Therefore, recommending the preference set obtained by the above method to all users in the cluster corresponding to this preference set can improve the conversion rate of recommended products.

[0121] The embodiment of the present invention obtains the user's historical data to analyze the user's characteristics, processes the user characteristics into a feature vector of a preset length for incremental clustering, obtains a preference set of similar users, and recommends them to related users. This solves the problem that the recommended products do not match the user's expected consumption preferences or the recommended products are relatively single, resulting in a low conversion rate for the recommended products, and the problem of being unable to make accurate product recommendations to new users. This improves the accuracy and conversion rate of recommended products, and achieves the effect of accurately recommending products to new users who have not purchased any products.

[0122] Based on the above embodiments, Figure 4BAs shown, optionally, the clustering processing module includes:

[0123] The pre-clustering submodule 441 is used to pre-cluster the second user feature vector using a clustering algorithm.

[0124] The second user feature vector is processed by a clustering algorithm, such as the K-Means clustering algorithm. Specifically, the K-Means clustering algorithm is an iterative clustering analysis algorithm. The data is pre-divided into K groups, and K user features are randomly selected as the initial cluster centers. Then the distance between each user feature and each seed cluster center is calculated, and the user corresponding to each user feature is assigned to the cluster center closest to it. The cluster centers and the objects assigned to them represent a cluster. Each time a sample is assigned, the cluster center of the cluster is recalculated based on the existing user features in the cluster. This process will be repeated until a certain termination condition is met. The termination condition can be that no (or a minimum number) of user features are reassigned to different clusters, no (or a minimum number) of cluster centers change again, and the sum of squared errors is locally minimized.

[0125] The saving submodule 442 is used to save the pre-clustering cluster centers and class labels.

[0126] The cluster centers and class labels obtained by pre-clustering are saved, and the cluster centers are kept unchanged. The class labels are the same labels of the same cluster, and the meaning of the class labels remains unchanged for a long time, so that the system can better learn this feature.

[0127] The calculation submodule 443 is used to obtain the feature vector newly added by the user and calculate the distance to each cluster center.

[0128] At present, the research on incremental clustering mainly regards incremental data as time series data or data in a specific order. The technical solution of this embodiment uses the results of the previous clustering, and each time a new feature vector of the user is obtained, the distance between the new feature vector of the user and each cluster center is calculated.

[0129] The clustering submodule 444 is used to group the users corresponding to the newly added feature vectors into the cluster centers with the closest distance.

[0130] According to the calculated user's newly added feature vector and the clustering of each cluster center, each user's newly added feature vector is divided into the existing clusters, that is, the user's newly added feature vector is assigned to the cluster with the center closest to it and the center is moved to the newly added feature vector. In other words, the user's newly added feature vector will not affect the original division. The advantage is that there is no need to re-cluster all the data every time.

[0131] The cluster center remains unchanged over a period of time, and the relationship between the user and the cluster center can be changed according to the cluster to which the user feature vector belongs. For example, a user may belong to one cluster today and another cluster tomorrow. The advantage of this is that the meaning of the same cluster label remains unchanged for a long time, making it easier for the sorting layer to learn this feature better.

[0132] The updating submodule 445 is used to update the cluster center of each class.

[0133] During the off-peak period of business, the current user feature vector data is re-clustered and the cluster center of each class is fully updated to eliminate the local deviation that may be introduced by incremental clustering, so as to improve the accuracy of the system.

[0134] Embodiment 4

[0135] Figure 5 A schematic diagram of the structure of a commodity recommendation device provided in the fourth embodiment of the present invention, such as Figure 5 As shown, the device includes a processor 50, a memory 51, an input device 52 and an output device 53; the number of processors 50 in the device can be one or more. Figure 5 A processor 50 is taken as an example; the processor 50, memory 51, input device 52 and output device 53 in the device can be connected by a bus or other means. Figure 5 The example of connecting through bus is taken in the following.

[0136] The memory 51 is a computer-readable storage medium that can be used to store software programs, computer executable programs and modules, such as program instructions / modules corresponding to the product recommendation method in the embodiment of the present invention (for example, the user data acquisition module 410, the encoding processing module 420, the fully connected layer processing module 430, the clustering processing module 440, the preference acquisition module 450 and the product recommendation module 460 in the theme product recommendation system). The processor 50 executes various functional applications and data processing of the device by running the software programs, instructions and modules stored in the memory 51, that is, realizing the above-mentioned product recommendation method.

[0137] The memory 51 may mainly include a program storage area and a data storage area, wherein the program storage area may store an operating system and at least one application required for a function; the data storage area may store data created according to the use of the terminal, etc. In addition, the memory 51 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some instances, the memory 51 may further include a memory remotely arranged relative to the processor 50, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0138] The input device 52 may be used to receive input digital or character information and generate key signal input related to user settings and function control of the device. The output device 53 may include a display device such as a display screen.

[0139] Embodiment 5

[0140] Embodiment 5 of the present invention further provides a storage medium containing computer executable instructions, wherein the computer executable instructions are used to perform a search method when executed by a computer processor, the method comprising:

[0141] Obtain the user's historical data and obtain the user's characteristic data through preprocessing;

[0142] The user feature data is encoded and concatenated to obtain a first user feature vector;

[0143] Converting the first user feature vector into a second user feature vector of a preset length after passing through two fully connected layers;

[0144] performing incremental clustering on the second user feature vector to classify users corresponding to the second user feature vector into different clusters;

[0145] Get the preference set of all users in each cluster;

[0146] Recommend the preference set to all users in the corresponding cluster.

[0147] Of course, the storage medium containing computer executable instructions provided by an embodiment of the present invention is not limited to the method operations described above, and can also execute related operations in the product recommendation method provided by any embodiment of the present invention.

[0148] Through the above description of the implementation methods, the technicians in the relevant field can clearly understand that the present invention can be implemented by means of software and necessary general hardware, and of course it can also be implemented by hardware, but in many cases the former is a better implementation method. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as a computer floppy disk, read-only memory (ROM), random access memory (RAM), flash memory (FLASH), hard disk or optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, server, or network device, etc.) to execute the methods described in each embodiment of the present invention.

[0149] It is worth noting that in the embodiment of the above-mentioned search device, the various units and modules included are only divided according to functional logic, but are not limited to the above-mentioned division, as long as the corresponding functions can be achieved; in addition, the specific names of the functional units are only for the convenience of distinguishing each other, and are not used to limit the scope of protection of the present invention.

[0150] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A product recommendation method, characterized in that: include: Obtain the user's historical data and obtain the user's characteristic data through preprocessing; The user feature data is encoded and concatenated to obtain a first user feature vector; Converting the first user feature vector into a second user feature vector of a preset length after passing through two fully connected layers; performing incremental clustering on the second user feature vector to classify users corresponding to the second user feature vector into different clusters; Get the preference set of all users in each cluster; recommending the preference set to all users in the corresponding cluster; The step of obtaining a preference set of users of each cluster includes: Obtain the product information generated by each user in a cluster to form a preference pre-set; Determining whether the quantity of commodities in the preference pre-set is greater than a preset quantity; If the quantity of the commodities is greater than the preset quantity, the commodities in the preference pre-set are sorted, and a preset quantity of commodities are selected in sequence to form a preference set; If the quantity of the commodities is not greater than the preset quantity, determining the preference pre-set as the preference set; If the number of commodities is greater than a preset number, the commodities in the preference pre-set are sorted, and a preset number of commodities are sequentially selected to form a preference set, including: When the number of commodities in the preference pre-set is greater than a preset number, the commodities in the preference pre-set are sorted in descending order according to the number of clicks and purchases, and a preset number of commodities are selected in sequence to form a preference set; The performing incremental clustering on the second user feature vector to classify users corresponding to the second user feature vector into different clusters includes: Pre-clustering the second user feature vector using a clustering algorithm; Save the pre-clustered cluster centers and class labels; Obtain the feature vector newly added by the user, and calculate the distance to each of the cluster centers; Assigning users corresponding to the newly added feature vectors to the nearest cluster centers; Update the cluster center of each class.

2. The commodity recommendation method according to claim 1, characterized in that: The user feature data includes user dense features and user sparse features. When the user feature data is encoded, it includes: The user dense features are processed by One-Hot encoding; the user sparse features are processed by Embedding encoding.

3. The commodity recommendation method according to claim 1, characterized in that: The user's historical data includes but is not limited to device information, geographic location information and demographic information, as well as product browsing, clicking, collection and purchase information.

4. A product recommendation system, characterized in that: include: User data acquisition module, used to obtain user historical data and obtain user feature data after preprocessing; An encoding processing module, used for performing encoding processing and feature splicing processing on the user feature data to obtain a first user feature vector; A fully connected layer processing module, used for converting the first user feature vector into a second user feature vector of a preset length after passing through two fully connected layers; A clustering processing module, used for performing incremental clustering on the second user feature vector, and dividing the users corresponding to the second user feature vector into different clusters; A preference acquisition module is used to obtain the preference set of users in each cluster; A product recommendation module, used for recommending the preference set to all users in the corresponding cluster; A preference set determination module is used to obtain the commodity information generated by each user of a cluster to form a preference pre-set; Determining whether the quantity of commodities in the preference pre-set is greater than a preset quantity; If the quantity of the commodities is greater than the preset quantity, the commodities in the preference pre-set are sorted, and a preset quantity of commodities are selected in sequence to form a preference set; If the quantity of the commodities is not greater than the preset quantity, determining the preference pre-set as the preference set; A sorting module, for sorting the commodities in the preference pre-set in descending order according to the number of clicks and purchases when the number of commodities in the preference pre-set is greater than a preset number, and selecting a preset number of commodities in turn to form a preference set; The cluster processing module comprises: A pre-clustering submodule, used for pre-clustering the second user feature vector using a clustering algorithm; The saving submodule is used to save the pre-clustering cluster centers and class labels; A calculation submodule, used to obtain the feature vector newly added by the user and calculate the distance to each of the cluster centers; A clustering submodule, used to group users corresponding to the newly added feature vectors into the nearest cluster centers; The update submodule is used to update the cluster center of each class.

5. A commodity recommendation device, characterized in that: The commodity recommendation device comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the commodity recommendation method as described in any one of claims 1-3.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the product recommendation method as described in any one of claims 1 to 3 is implemented.