Product recommendation method and device, electronic device, computer storage medium

Abstract

The present application provides a product recommendation method and device, electronic device, and computer storage medium. In the above method, the user behavior sequence generated by the user for the target product is matched with the target sequence pattern of the user obtained by mining, and the sample weight of the combined sample of the user and the target product is configured according to the matching result. Then, the sample weight of the combined sample is used to weight the combined feature vector constructed based on the combined features of the user features and the product features to obtain a weighted combined feature vector of the combined feature vector, and the weighted combined feature vector is used to calculate the cluster to which the combined sample belongs. The cluster is obtained by clustering the weighted combined feature vectors of multiple cluster samples. Finally, the target product is recommended to the user corresponding to the cluster sample in the cluster to which the combined sample belongs, thereby combining the sequence pattern mining and clustering algorithm to achieve product recommendation to the user.

Description

Technical Field

[0001] The present application relates to the technical field of product recommendation, and in particular to a product recommendation method and device, electronic equipment, and computer storage medium. Background Art

[0002] Personalized recommendation has important significance and wide application scenarios in e-commerce platforms, video software, advertising, etc. At present, the mainstream business personalized recommendation methods mainly include demographic-based recommendation, content-based recommendation and coordinated filtering algorithm recommendation.

[0003] The demography-based recommendation method mainly recommends products that are similar to the target user in demographic characteristics to the target user. Content-based recommendation recommends products that are similar to the target user's favorite products to the target user. The collaborative filtering algorithm recommendation mainly uses association algorithms, clustering algorithms, classification algorithms, regression algorithms, matrix decomposition, graph models, etc. to complete collaborative filtering, which can be specifically divided into user-based collaboration and item-based collaboration. Among them, the user-based collaborative method is to calculate and analyze users with similar preferences to the target user from historical preference data, and recommend the target user to the target user; the item collaboration is to compare the historical preference data of the target user and multiple users, and based on the target user and multiple users with common favorite products, predict products that meet the target user's preferences and recommend them to the user.

[0004] However, these methods mainly target target users and recommend products that may suit their preferences, but fail to accurately recommend target products to users who are most likely to like or need the products. Summary of the invention

[0005] Based on the above-mentioned deficiencies of the prior art, the present application provides a product recommendation method and device, electronic device, and computer storage medium to solve the problem of accurately recommending a target product to users who are most likely to like or need the product.

[0006] In order to achieve the above objectives, this application provides the following technical solutions:

[0007] The first aspect of the present application provides a product recommendation method, comprising:

[0008] Acquire product features of a target product and user features and a user behavior sequence of a user; wherein the user generates a user behavior indicated by the user behavior sequence for the target product;

[0009] Based on the matching results of the user behavior sequence and each determined target sequence pattern, a sample weight of a combined sample of the user and the target product is configured; wherein the matching result indicates that when the user behavior sequence matches the target sequence pattern, the sample weight of the combined sample is higher; the target sequence pattern is a frequent sequence pattern obtained by performing sequence pattern mining on the user behavior sequence generated when the user converts multiple products;

[0010] Using the sample weight of the combined sample, weighting the combined feature vector constructed based on the combined features of the user feature and the product feature to obtain a weighted combined feature vector of the combined feature vector;

[0011] The weighted combined feature vector of the combined feature vector is used to calculate the cluster to which the combined sample belongs; wherein the cluster is obtained by clustering the weighted combined feature vectors of multiple cluster samples, and the weighted combined feature vector of each cluster sample is obtained by weighting the combined feature vector of the cluster sample and the sample weight of the cluster sample; one cluster sample corresponds to one user and one product converted by the user;

[0012] The target product is recommended to users corresponding to the cluster samples in the cluster to which the combined sample belongs.

[0013] Optionally, in the above product recommendation method, the step of calculating the cluster to which the combined sample belongs by using the weighted combined feature vector of the combined feature vector includes:

[0014] Performing clustering using weighted combined feature vectors of a plurality of cluster samples to obtain a plurality of clusters;

[0015] Using the weighted combined feature vector of the combined feature vector, the distance between the combined sample and the cluster center of each cluster is calculated;

[0016] Determine the cluster corresponding to the cluster center having the shortest distance to the combined sample as the cluster to which the combined sample belongs;

[0017] Alternatively, clustering is performed using the weighted combined feature vector of the combined feature vector and the weighted combined feature vectors of a plurality of cluster samples to obtain a plurality of clusters, and the cluster to which the combined sample belongs is screened out from the plurality of clusters.

[0018] Optionally, in the above product recommendation method, after calculating the cluster to which the combined sample belongs by using the weighted combined feature vector of the combined feature vector, the method further comprises:

[0019] Calculate the conversion rate ratio of users corresponding to each of the cluster samples in the cluster to which the combined sample belongs;

[0020] The step of recommending the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs includes:

[0021] The target product is recommended to the user with the highest conversion rate.

[0022] Optionally, in the above-mentioned product recommendation method, the method for constructing the combined feature vector includes:

[0023] Performing multiple preprocessing on the user characteristics of the user and the product characteristics of the target product; wherein the preprocessing includes filtering processing, filling processing and derivation processing;

[0024] The preprocessed user features of the user are combined with the product features of the target product to obtain a combined feature of the user features and the product features;

[0025] Feature processing is performed on the combined features to obtain a combined feature vector of the combined features.

[0026] Optionally, in the above product recommendation method, configuring the sample weight of the combined sample of the user and the target product based on the matching result of the user behavior sequence and each target sequence pattern includes:

[0027] If the matching result indicates that the user behavior sequence contains the target sequence pattern, configuring the support rate of the longest target sequence pattern contained in the user behavior sequence as the combined sample weight;

[0028] If the matching result indicates that the user behavior sequence does not contain any of the target sequence patterns, the combined sample weight is configured as the minimum support rate set when mining sequence patterns, or as the support rate of the user behavior sequence in each of the conversion behavior sequences corresponding to the user.

[0029] Optionally, in the above-mentioned product recommendation method, the method for determining each target sequence pattern includes:

[0030] Finding each frequent sequence pattern of the user from the frequent sequence patterns of multiple users mined in advance, and determining each frequent sequence pattern of the user as the target sequence pattern;

[0031] Alternatively, sequence pattern mining is performed on the behavior sequence generated when the user converts to multiple products to obtain the user's frequent sequence pattern, and the user's frequent sequence pattern is determined as the target sequence pattern.

[0032] Optionally, in the above product recommendation method, performing sequence pattern mining on the behavior sequence generated when the user converts to the plurality of sample products to obtain the frequent sequence pattern of the user includes:

[0033] Obtaining the behavior sequences generated when the user converts to each product respectively, and composing each of the behavior sequences into a behavior sequence set; wherein each of the behavior sequences consists of a plurality of behavior codes; and one behavior code refers to one user behavior;

[0034] Determine each behavior code in the behavior sequence set whose support is greater than or equal to the support threshold as a level 1 sequence pattern, and set the sequence level N to 2;

[0035] Obtaining a projected behavior sequence set corresponding to each N-1 level sequence pattern; wherein the projected behavior sequence set corresponding to the N-1 level sequence pattern includes the N-1 level suffix obtained from each of the behavior sequences in the behavior sequence set;

[0036] Combine each behavior code whose support in the corresponding projected behavior sequence set is greater than or equal to the support threshold and the N-1 level sequence pattern corresponding to the projected behavior sequence set into an N-level sequence pattern, and increment N by 1 and then return to execute to obtain the projected behavior sequence set corresponding to each N-1 level sequence pattern until there is no behavior code whose support in the corresponding projected behavior sequence set is greater than or equal to the support threshold;

[0037] The obtained sequence pattern of each level is determined as the frequent sequence pattern of the user.

[0038] A second aspect of the present application provides a product recommendation device, comprising:

[0039] A first acquisition unit is used to acquire product features of a target product and user features and a user behavior sequence of a user; wherein the user generates a user behavior indicated by the user behavior sequence for the target product;

[0040] a configuration unit, configured to configure a sample weight of a combined sample of the user and the target product based on a matching result between the user behavior sequence and each determined target sequence pattern; wherein the matching result indicates that when the user behavior sequence matches the target sequence pattern, the sample weight of the combined sample is higher; and the target sequence pattern is a frequent sequence pattern obtained by performing sequence pattern mining on the user behavior sequence generated when the user converts multiple products;

[0041] A weighting unit, configured to weight a combined feature vector constructed based on the combined features of the user features and the product features using the sample weight of the combined sample, to obtain a weighted combined feature vector of the combined feature vector;

[0042] A first calculation unit is used to calculate the cluster to which the combined sample belongs by using the weighted combined feature vector of the combined feature vector; wherein the cluster is obtained by clustering the weighted combined feature vectors of multiple cluster samples, and the weighted combined feature vector of each cluster sample is obtained by weighting the combined feature vector of the cluster sample and the sample weight of the cluster sample; one cluster sample corresponds to one user and one product converted by the user;

[0043] A recommendation unit is used to recommend the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs.

[0044] Optionally, in the above-mentioned product recommendation device, the first calculation unit includes:

[0045] A first clustering unit, configured to perform clustering using a weighted combination feature vector of a plurality of the cluster samples to obtain a plurality of clusters;

[0046] A distance calculation unit, used to calculate the distance between the combined sample and the cluster center of each cluster by using the weighted combined feature vector of the combined feature vector;

[0047] The determining unit determines the cluster corresponding to the cluster center having the shortest distance to the combined sample as the cluster to which the combined sample belongs;

[0048] Alternatively, the first computing unit includes: a first computing subunit, used to perform clustering using the weighted combined feature vector of the combined feature vector and the weighted combined feature vectors of multiple cluster samples to obtain multiple clusters, and filter out the cluster to which the combined sample belongs from the multiple clusters.

[0049] Optionally, the recommended device of the above product further includes:

[0050] A second calculation unit is used to calculate the conversion rate ratio of users corresponding to each of the cluster samples in the cluster to which the combined sample belongs;

[0051] Wherein, when the recommendation unit performs the step of recommending the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs, it is used to:

[0052] The target product is recommended to the user with the highest conversion rate.

[0053] Optionally, in the above-mentioned product recommendation device, a feature vector construction unit is further included, and the feature vector construction unit includes:

[0054] A preprocessing unit, used to perform multiple preprocessing on the user characteristics of the user and the product characteristics of the target product; wherein the preprocessing includes filtering processing, filling processing and derivation processing;

[0055] A combining unit, used for combining the pre-processed user features of the user with the product features of the target product to obtain a combined feature of the user features and the product features;

[0056] The feature processing unit is used to perform feature processing on the combined feature to obtain a combined feature vector of the combined feature.

[0057] Optionally, in the above-mentioned product recommendation device, the configuration unit, when the matching result indicates that the user behavior sequence includes the target sequence pattern, is used to configure the support rate of the longest target sequence pattern contained in the user behavior sequence as the combined sample weight; when the matching result indicates that the user behavior sequence does not include any of the target sequence patterns, the configuration unit is used to configure the combined sample weight as the minimum support rate set when mining sequence patterns, or as the support rate of the user behavior sequence in each of the conversion behavior sequences corresponding to the user.

[0058] Optionally, the product recommendation device further comprises: a target sequence pattern determination unit, wherein the target sequence pattern determination unit comprises:

[0059] A searching unit, searching for each frequent sequence pattern of the user from the frequent sequence patterns of multiple users mined in advance, and determining each frequent sequence pattern of the user as the target sequence pattern;

[0060] Alternatively, the target sequence pattern determination unit includes: a mining unit for performing sequence pattern mining on the behavior sequence generated when the user converts to multiple products, obtaining the user's frequent sequence pattern, and determining the user's frequent sequence pattern as the target sequence pattern.

[0061] Optionally, in the recommended device of the above product, the excavation unit includes:

[0062] A second acquisition unit is used to respectively acquire the behavior sequences generated when the user converts to each product, and to form a behavior sequence set from each of the behavior sequences; wherein each of the behavior sequences is composed of a plurality of behavior codes; and one of the behavior codes refers to one user behavior;

[0063] A first-level sequence pattern mining unit, used for determining each behavior coding whose support is greater than or equal to a support threshold in the behavior sequence set as a first-level sequence pattern, and setting the sequence level N to 2;

[0064] A third acquisition unit is used to obtain a projection behavior sequence set corresponding to each N-1 level sequence pattern; wherein the projection behavior sequence set corresponding to the N-1 level sequence pattern includes the N-1 level suffix obtained from each of the behavior sequences in the behavior sequence set;

[0065] A sequence pattern mining unit, used for combining each behavior code whose support in the corresponding projected behavior sequence set is greater than or equal to the support threshold and the N-1 level sequence pattern corresponding to the projected behavior sequence set into an N-level sequence pattern, and returning to execute after incrementing N by 1 to obtain the projected behavior sequence set corresponding to each N-1 level sequence pattern until there is no behavior code whose support in the corresponding projected behavior sequence set is greater than or equal to the support threshold;

[0066] The sequence pattern determination unit is used to determine each level of the obtained sequence pattern as the frequent sequence pattern of the user.

[0067] A third aspect of the present application provides a computer storage medium for storing a computer program, wherein when the computer program is executed, it is used to implement the product recommendation method as described in any one of the above.

[0068] A fourth aspect of the present application discloses an electronic device, including a memory and a processor;

[0069] Wherein, the memory is used to store programs;

[0070] The processor is used to execute the program, and when the program is executed, it is specifically used to implement the product recommendation method as described in any one of the first aspects above.

[0071] The present application provides a product recommendation method, for a target product, obtaining the product features of the target product, the user features of the user, and the user behavior sequence generated by the user for the target product. Then, based on the matching results of the user behavior sequence and the determined target sequence patterns, the sample weight of the combined sample of the user and the target product is configured, and the combined feature vector constructed based on the combined features of the user features and the product features is weighted by using the sample weight of the combined sample to obtain the weighted combined feature vector of the combined feature vector, and finally the weighted combined feature vector is used to calculate the cluster to which the combined sample belongs, and the target product is recommended to the user corresponding to the cluster sample in the cluster to which the combined sample belongs. Since the target sequence pattern is a frequent sequence obtained in advance by mining the behavioral sequence of the user converted to multiple products, the present application configures the sample weight for the sample by reflecting the historical behavioral commonality of the user's conversion product, and the matching result shows that when the user behavior sequence matches the target sequence pattern, the sample weight of the combined sample is higher, so when the weighted combined feature vector obtained by using the sample weight to cluster the combined feature vector is clustered, the user's behavioral commonality is emphasized. Moreover, since the combined feature vector is constructed by using product features and user features, the correlation between products and users is also taken into account. Therefore, the target product is recommended to the user corresponding to the cluster sample in the cluster to which the combined sample belongs, and the target product can be accurately recommended to the user who is most likely to like or need the product. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0073] Figure 1 A schematic diagram of a flow chart of a method for constructing a weighted combined feature vector provided in an embodiment of the present application;

[0074] Figure 2 A schematic diagram of a flow chart of a method for constructing a combined feature vector provided in an embodiment of the present application;

[0075] Figure 3 A flowchart of a method for mining frequent sequence patterns provided in an embodiment of the present application;

[0076] Figure 4 A flowchart of a recommended method for a product provided by another embodiment of the present application;

[0077] Figure 5A schematic flow chart of a method for constructing a combined feature vector of a combined sample provided in another embodiment of the present application;

[0078] Figure 6 A schematic diagram of a flow chart of a method for calculating the cluster to which a combined sample belongs according to another embodiment of the present application;

[0079] Figure 7 A schematic diagram of a recommended device structure of a product provided by another embodiment of the present application;

[0080] Figure 8 A schematic diagram of the structure of a feature vector construction unit provided in another embodiment of the present application;

[0081] Fig. 9 A schematic structural diagram of an excavation unit provided in another embodiment of the present application;

[0082] Fig.10 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0083] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0084] In this application, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprise", "include" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the statement "comprises a ..." does not exclude the presence of other identical elements in the process, method, article or device including the element.

[0085] Products can be understood as anything that is produced and provided to people for use and consumption, and can meet people's needs. Products can specifically include tangible things, such as daily necessities, food, etc., as well as intangible things, such as services, advertisements, videos, online novels, etc., and of course, a combination of the two. In order to achieve the expected benefits, personalized recommendation systems are used in many fields to ensure that the products provided to users meet the preferences or needs of users. Since the scope of products is relatively wide, in order to more clearly illustrate the product recommendation method provided by this application, this application is mainly described as an example of product recommendation on an e-commerce platform, but the method provided in any embodiment of this application is applicable to products on e-commerce platforms as well as other products, such as videos, music, advertisements, etc.

[0086] In order to recommend a product to the user who is most likely to like or need the product, the present application provides a product recommendation method. Since the weighted combination feature vector of multiple cluster samples is required when implementing the product recommendation method provided by the present application, the present application embodiment provides a method for constructing a weighted combination feature vector, see Figure 1 , the method may include the following steps:

[0087] S101. Obtain user characteristics of multiple users and product characteristics of multiple products.

[0088] Among them, user characteristics may include user personal information and user behavior information, etc., which are used to reflect the characteristics of the user. For example, in an e-commerce platform, the obtained user characteristics may include the user's basic attribute characteristics, such as age, gender, education, city, and other personal basic attribute characteristics, as well as the user's consumption characteristics, such as the total number of payments, total amount, the number of payments in a specified time period, the payment amount distribution, the average amount per payment, etc., and may also include user behavior characteristics, such as browsing time, number of web page clicks, etc. It should be noted that in different fields, due to the difference in products, the way of consuming products is different, so the obtained user characteristics can be selected according to the actual application scenario.

[0089] Product features may include basic product attribute information, such as the type of movie, the length of music, the brand of the product, etc., and may also include product consumption characteristics, such as movie reviews, the number of times music is played, the number of times the product is purchased, etc. For example, in an e-commerce platform, the basic attributes of the obtained product features may include product category, product price, product brand, product purchase reviews, etc., and the product consumption characteristics of the obtained product features may include the number of times the product has been purchased, the number of times it has been clicked, the number of times it has been added to the shopping cart, and the number of times similar items have been purchased.

[0090] S102: construct a combined feature vector of multiple cluster samples by using the user features of the user and the product features of the product.

[0091] Among them, the combined feature vector of a cluster sample is constructed by the user features of a user and the product features of a product converted by the user. In other words, the combination of a user and a product converted by the user corresponds to a cluster sample. In addition, a user can be combined with multiple products converted by the user, that is, a user corresponds to multiple cluster samples. Similarly, a product can be combined with multiple users who converted the product, that is, a product can also correspond to multiple cluster samples.

[0092] It should be noted that user conversion to a product refers to the user's behavior towards a product that the product provider expects. For example, in an e-commerce platform, the behavior that the product provider usually expects is for the user to purchase the provided product. Therefore, when the user purchases the provided product, it is considered that the user has achieved conversion to the product. For products such as movies or music, usually when the user orders the provided movie or music, it is determined that the user has converted the ordered product.

[0093] Specifically, for each user, the user characteristics of the user are evaluated and combined with the product characteristics of each product converted by the user, that is, the two separate characteristics are combined to obtain multiple combined features, and the combined features are feature processed to obtain a multi-dimensional combined feature vector.

[0094] See also Figure 2 A specific implementation of step S102 includes the following steps:

[0095] S201, performing multiple preprocessing on the acquired user features and product features, the preprocessing including filtering processing, filling processing and derivation processing.

[0096] Among them, the filtering process may include discarding features with too many missing values ​​and outlier processing. Specifically, for the processing of discarding features with too many missing values, a missing value filtering threshold may be set. When the number of missing data of a feature exceeds the set missing value filtering threshold, all data of the feature will be deleted, and single-value features may be further deleted. Among them, the set missing value filtering threshold may be equal to the product of the total sample size and the set filtering ratio. For outlier processing, features with too large feature values ​​and outliers ranked in the top n% may be discarded based on feature distribution, where n is set according to feature value distribution.

[0097] Filling can be understood as filling the missing values ​​of features with relatively few missing values. Specifically, the missing values ​​of continuous features can be filled with the mean, and the missing values ​​of discrete features can be filled with constants as separate categories.

[0098] Feature derivation processing can combine and derive features through feature transformation, squaring, addition and subtraction.

[0099] S202 , for each user, respectively, concatenate and combine the pre-processed user features of the user and the product features of each product converted by the user to obtain the combined features of the cluster samples corresponding to the user and each product converted by the user.

[0100] S203 , performing feature processing on the combined features of each cluster sample respectively to obtain a combined feature vector of the combined features of each cluster sample.

[0101] Optionally, continuous features can be processed by binning and discretization, while discrete features can be one-hot encoded.

[0102] S103, for each user, respectively, obtaining the conversion behavior sequence generated when the user converts to each product, and performing sequence pattern mining on the conversion behavior sequence generated when the user converts to each product to obtain the frequent sequence pattern of the user.

[0103] Among them, the conversion behavior sequence generated by the user's conversion to the product mainly refers to the behavior sequence of the user's final conversion to the product and the multiple behaviors before the conversion to the product in a more time-ordered combination. For example, the process of the user's conversion to item 1 is: click button a on page A to enter page B, then click button b to enter page C after browsing for a while, and click button c on page C to purchase item 1. The conversion behavior sequence generated when the user converts to item 1 can be recorded as AaBbCc.

[0104] Sequential pattern mining refers to mining patterns with high frequency of occurrence relative to time or other patterns. In this application, for each user, it is to mine the behavioral sequences with high frequency of occurrence in the process of product conversion of the user, so as to obtain the frequent sequence patterns of the user.

[0105] Optionally, the sequence pattern mining can be performed on the conversion behavior sequence generated when each user converts to each sample product based on the prefixspan algorithm. Since the sequence pattern mining is performed for each user separately, and the method of performing the sequence pattern mining for each user is the same, the mining process for only one user is described.

[0106] For details on the process of mining a user's sequence pattern based on the prefixspan algorithm, see Figure 3 , including the following steps:

[0107] S301. Acquire conversion behavior sequences generated when a user converts to each product respectively, and group the conversion behavior sequences into a behavior sequence set, where each conversion behavior sequence consists of multiple behavior codes, and one behavior code refers to one user behavior.

[0108] It should be noted that, since it is impossible to directly process the behavior sequence expressed in words for calculation, in the embodiment of the present application, a behavior code is used to refer to a user behavior, and a plurality of corresponding behavior codes are used to refer to the conversion behavior sequence for the convenience of calculation. For example, the process of a user switching to item 1 is: entering the e-commerce platform, then registering and logging in, and clicking to enter the item details page after browsing the page for a while, clicking the favorite button to collect the item after browsing for a while, and then clicking to add to the shopping cart to purchase the item. Please refer to Table 1, the correspondence between user behavior and behavior code is as follows:

[0109] Table 1

[0110] User Conduct Behavior Coding Purchase behavior h Add to cart behavior g Collection behavior f Comment Behavior e Search Behavior d Login behavior c Registration behavior b

[0111] Therefore, according to the relationship between user behavior and behavior expression in Table 1, the conversion behavior sequence of the user to item 1 is expressed as: bcafgh.

[0112] S302: Determine each behavior code in the behavior sequence set whose support is greater than or equal to the support threshold as a level 1 sequence pattern, and set the sequence level N to 2.

[0113] The support degree can be understood as the number of behavior sequences in which the behavior code appears. Specifically, the support degree threshold can be set as the product of the total number of conversion behavior sequences and the set minimum support rate.

[0114] In this embodiment, the level of a sequence pattern is equal to the length of the sequence pattern. In other words, if a sequence pattern is an N-level sequence pattern, the sequence pattern contains N behavior codes. Therefore, each level 1 sequence pattern contains only one behavior code, and the support is greater than or equal to the support threshold.

[0115] For example, referring to the example in the description of step S301, the transformation behavior sequence of the user transforming to item 1 is represented as: bcafgh, and the transformation behavior sequence of the user transforming to item 2 is: bcdaghf. Assume that the support threshold is set to 2. Then, in the two transformation behavior sequences, 6 behavior codes are included: h, g, f, d, c, b. Since f only appears in one transformation behavior sequence, the behavior codes that meet the support threshold are only: h, g, d, c, b, so the first-level sequence pattern is: h, g, d, c, b.

[0116] S303: Obtain a projection behavior sequence set corresponding to each N-1 level sequence pattern, wherein the projection behavior sequence set corresponding to the N-1 level sequence pattern includes an N-1 level suffix obtained from each transformation behavior sequence in the behavior sequence set.

[0117] It should be noted that the obtained N-level sequence pattern is the N-item prefix of the conversion behavior sequence, so the projection behavior sequence set corresponding to the N-1-level sequence pattern is the set of suffixes corresponding to each N-1-item prefix.

[0118] For example, based on the example in step S302, the mined first-level sequence pattern is: h, g, d, c, b, then the 1-item prefix is: h, g, d, c, b. Referring to Table 2, the suffixes of each 1-item prefix are as follows:

[0119] Table 2

[0120]

[0121] Therefore, the set of projection behavior sequences corresponding to the level 1 sequence pattern is: cafgh, cdaghf, afgh, daghf, fgh, ghf, h, hf, f.

[0122] It can be understood that in this embodiment, when step S203 is executed for the first time, N is equal to 2, and what is obtained by executing step S203 at this time is the projected corpus of each level 1 sequence pattern in the filtered training corpus. Afterwards, since N is incremented by 1 by executing step S206, when step S203 is executed for the second time, N is equal to 3, and what is obtained by step S203 is the projected corpus of each level 2 sequence pattern in the filtered training corpus. When it is executed for the third time, it becomes the projected corpus of the level 3 sequence pattern, and so on.

[0123] S304. Combine each behavior code whose support is greater than or equal to the support threshold in the corresponding projected behavior sequence set and the N-1 level sequence pattern corresponding to the projected behavior sequence set into an N level sequence pattern, and increment N by 1 and return to execute step S303 to obtain the projected behavior sequence set corresponding to each N-1 level sequence pattern until there is no behavior code whose support is greater than or equal to the support threshold in the corresponding projected behavior sequence set.

[0124] Specifically, only when mining the 1st level sequence pattern, the behavior sequence set consisting of the original conversion behavior sequences is used as the mining target. When mining the N-level sequence pattern with N greater than 1, the projected behavior sequence set of the N-1-level sequence pattern is used as the mining target. Therefore, the method of mining each level of sequence pattern is the same, only the mining target is different.

[0125] It should be noted that, since the support of the N-1 level sequence pattern is greater than or equal to the support threshold, the support of the combination of each behavior code whose support is greater than or equal to the support threshold in the corresponding projected behavior sequence set and the N-1 level sequence pattern corresponding to the projected behavior sequence set is also greater than or equal to the support threshold. Therefore, it can also be understood as combining each behavior code in the corresponding projected behavior sequence set with the corresponding N-1 level sequence pattern, and determining the combination whose support is greater than or equal to the support threshold as the N level sequence pattern. Then, N is incremented by 1 and returned to execute to obtain the projected behavior sequence set corresponding to each N-1 level sequence pattern until there is no behavior code whose support is greater than or equal to the support threshold in the corresponding projected behavior sequence set.

[0126] For example, when the example in the description of step S303 mines the first-level sequence pattern as: h, g, d, c, b, the second-level sequence pattern is mined based on the projected behavior sequence set of the first-level sequence pattern, and the projected behavior sequence set of the second-level sequence pattern is obtained, as shown in Table 3.

[0127] Table 3

[0128]

[0129] Similarly, the third-level sequence pattern is mined from the projected behavior sequence set of the second-level sequence pattern, and the projected behavior sequence set of the third-level sequence pattern is obtained, as shown in Table 4.

[0130] Table 4

[0131]

[0132] Similarly, the same method is used to continuously mine until a 5-level sequence pattern is mined. The projected behavior sequence set corresponding to the 5-level sequence pattern obtained has only one projected behavior sequence left, which can no longer meet the requirement of being equal to or greater than the support degree. Therefore, the sequence pattern mining ends.

[0133] S305: Determine each level of obtained sequence patterns as the user's frequent sequence patterns.

[0134] S104 . For each cluster sample, respectively, based on the matching result between the conversion behavior sequence corresponding to the cluster sample and the frequent sequence pattern of the user corresponding to the cluster sample, configure the sample weight of the cluster sample.

[0135] It should be noted that the conversion behavior sequence corresponding to a cluster sample refers to the conversion behavior sequence generated when the user corresponding to the cluster sample converts the product corresponding to the cluster sample.

[0136] The sample weights of cluster samples are configured so that the algorithm can pay more attention to cluster samples that conform to the user's behavior habits during subsequent cluster learning, that is, to cluster samples whose conversion behavior sequences of the corresponding users contain corresponding frequent sequence patterns. Therefore, when the conversion behavior sequence corresponding to the cluster sample matches the frequent sequence pattern of the user corresponding to the cluster sample, the sample weights of the cluster samples configured are greater than the sample weights configured when they do not match. It should be noted that when a conversion behavior sequence contains a frequent sequence pattern, it means that the conversion behavior sequence matches the frequent sequence pattern.

[0137] Optionally, in an embodiment of the present application, if a frequent sequence pattern is included in the conversion behavior sequence, the sample weight is configured as the support rate of the longest frequent sequence pattern included in the conversion behavior sequence. Among them, the support rate of a frequent sequence pattern is equal to the ratio of the number of conversion behavior sequences in which the frequent sequence pattern appears in the frequent sequence pattern of the corresponding user to the total number of conversion behavior sequences. Thus, the same sample weight is configured for the cluster samples corresponding to the conversion behavior sequences containing the same longest frequent sequence pattern. In addition, since the longer the frequent sequence pattern, the smaller its support rate is usually, the closer the length of the longest frequent sequence pattern included, the closer the sample weights configured for the cluster samples corresponding to the conversion behavior sequences. The closer the length of the longest frequent sequence pattern included, the more similar the conversion behavior sequences are, so cluster samples with higher behavioral commonality can be clustered into one category.

[0138] Optionally, if the conversion behavior sequence does not contain any of the frequent sequence patterns, the sample weight is configured as the minimum support rate set when mining the sequence pattern, or as the support rate of the conversion behavior sequence in each conversion behavior sequence of the user corresponding to the cluster sample.

[0139] It should be noted that if a conversion behavior sequence corresponding to a user and a behavior sequence shared by other conversion behavior sequences corresponding to the user do not meet the set support threshold, the conversion behavior sequence does not contain any of the frequent sequence patterns.

[0140] Since the mined frequent sequence patterns are all greater than or equal to the set support threshold, they are all greater than the set minimum support rate. Therefore, when any of the frequent sequence patterns are not included, the configured sample weight is not less than the sample weight configured when the frequent sequence pattern is not included. The support rate of the conversion behavior sequence in each conversion behavior sequence of the user corresponding to the cluster sample is the ratio of the number of conversion behavior sequences that contain the conversion behavior sequence in each conversion behavior sequence of the user to the total number of each conversion behavior sequence of the user. Obviously, this ratio is usually smaller than the minimum support rate, so it can meet the requirement of not being greater than the sample weight configured during matching.

[0141] It should also be noted that the above is only one of the optional ways to configure the sample weights of cluster samples, and other configuration methods can be used, such as configuring the same first preset sample weights uniformly when matching, or configuring sample weights according to the number of matched frequent sequence patterns. Similarly, when there is no match, the same first preset weight can also be uniformly configured, and the first preset weight is greater than the second preset sample weight.

[0142] S105 . For each cluster sample, weight the combined feature vector of the cluster samples using the sample weight of the cluster sample to obtain a weighted combined feature vector of the cluster samples.

[0143] It should be noted that step S101 to step S102 are steps for constructing the combined feature vector of cluster samples, while step S103 to step S104 are steps for obtaining the sample weights of cluster samples. It can be seen that the two are independent of each other and can be performed separately. Therefore, step S103 to step S104 are not limited to being performed after step S105.

[0144] Based on the weighted combination feature vector of the cluster samples provided in the above embodiment, another embodiment of the present application provides a product recommendation method, see Figure 3 , the method comprises the following steps:

[0145] S401, obtaining product features of a target product and user features and a user behavior sequence of a user, wherein the user generates user behaviors indicated by the user behavior sequence for the target product.

[0146] It should be noted that, since new products need to be recommended to users in a timely manner, in the embodiments of the present application, the target product mainly refers to the product that has not been converted by the user, that is, it can be understood that the target product can be a new product. Similarly, the acquired user behavior sequence can be a behavior sequence generated by the user for the product that does not include conversion behavior. Optionally, if the user behavior sequence generated by the user indicates that the user has come into contact with the target product, for example, it includes behaviors such as the user browsing the purchase page of the target product, then it can be determined that the behavior sequence is generated for the target product.

[0147] S402: Based on the matching results between the user behavior sequence and each determined target sequence pattern, a sample weight of a combination sample of the user and the target product is configured.

[0148] Among them, when the matching result indicates that the user behavior sequence matches the target sequence pattern, the sample weight of the combined sample is higher. It should be noted that, as in the above embodiment, when the user behavior sequence contains a certain target sequence pattern, it means that the user behavior sequence matches the target sequence pattern, and the sample weight configured at this time is greater than the sample weight configured when the user behavior sequence does not match any target sequence pattern.

[0149] It should be noted that the sample weight configuration method of the combined sample needs to be the same as the sample weight configuration method of the clustered sample. Therefore, also optionally, if the matching result indicates that the user behavior sequence contains a target sequence pattern, the support rate of the longest target sequence pattern contained in the user behavior sequence is configured as the combined sample weight. For example, each target sequence pattern is: a, ab, abe, abeh, and the user behavior sequence is: abde, then the target sequence pattern matched with the user behavior sequence includes: a, ab, abe, and the longest target sequence pattern contained in the user behavior sequence is abe. If the support rate of abe is 0.6, the sample weight of the combined sample is configured to 0.6. If the matching result indicates that the user behavior sequence does not contain any target sequence pattern, then the combined sample weight is configured to the minimum support rate set when mining the sequence pattern, or configured to the support rate of the user behavior sequence in each of the conversion behavior sequences corresponding to the user. According to the configuration method, reference can be made to step S104 in the above embodiment, which will not be repeated here.

[0150] The target sequence pattern is a frequent sequence pattern obtained by performing sequence pattern mining on the user behavior sequence generated when the user converts to multiple products. Specifically, each frequent sequence pattern of the user can be found from the frequent sequence patterns of multiple users obtained by preset mining, that is, from the frequent sequence patterns of each user mined in the above embodiment, each frequent sequence pattern of the user is found and each frequent sequence pattern of the user is determined as the target sequence pattern, so that there is no need to perform sequence pattern mining for the user additionally. When the frequent sequence of the user is not included in the frequent sequence patterns of each user mined in advance, sequence pattern mining is performed on the behavior sequence generated when the user converts to multiple products to obtain the frequent sequence pattern of the user, and the frequent sequence pattern of the user is determined as the target sequence pattern.

[0151] Among them, the behavior sequence generated when the user converts to multiple products is subjected to sequence pattern mining, and the method of obtaining the user's frequent sequence pattern is the same as the method of sequence pattern mining performed in the process of constructing the weighted combination feature vector of the clustering samples. Therefore, the behavior sequence generated when the user converts to multiple products is subjected to sequence pattern mining to obtain the user's frequent sequence pattern as follows: the behavior sequence generated when the user converts to each product is obtained respectively, and each behavior sequence is combined into a behavior sequence set; each behavior code in the behavior sequence set with a support greater than or equal to the support threshold is determined as a level 1 sequence pattern, and the sequence level N is set to 2; the projected behavior sequence set corresponding to each N-1 level sequence pattern is obtained; each behavior code with a support greater than or equal to the support threshold in the corresponding projected behavior sequence set and the N-1 level sequence pattern corresponding to the projected behavior sequence set are combined into an N level sequence pattern, and N is incremented by 1 and then returned to execute to obtain the projected behavior sequence set corresponding to each N-1 level sequence pattern until there is no behavior code with a support greater than or equal to the support threshold in the corresponding projected behavior sequence set; each level of sequence pattern obtained is determined as the frequent sequence pattern of the user. For a more specific sequence mining process, reference may be made to steps S301 to S305 in the above embodiment, which will not be described in detail here.

[0152] It should also be noted that the sequence pattern of each user's new conversion behavior sequence can be mined again at regular intervals, thereby continuously updating the user's frequent sequence pattern.

[0153] S403: using the sample weight of the combined sample, weighting the combined feature vector constructed based on the combined features of the user feature and the product feature to obtain a weighted combined feature vector of the combined feature vector.

[0154] Specifically, a combined feature vector is constructed by using the acquired user features of the user and the product features of the target product, and then the combined feature vector is multiplied by the sample weight of the combined sample to obtain a weighted combined feature vector of the combined feature vector.

[0155] See also Figure 5 , a method for constructing a combined feature vector of user features and product features specifically includes the following steps:

[0156] S501: Perform multiple preprocessing on the user characteristics of the user and the product characteristics of the target product.

[0157] Among them, the preprocessing includes filtering processing, filling processing and derivation processing.

[0158] It should be noted that the features filtered during filtering need to be the same as those filtered during filtering of the features of the clustered samples, and the processing methods and objects of the filling and derivative processing also need to be the same as those during preprocessing of the features of the clustered samples. The specific implementation of step S501 can be referred to step S201 in the above method embodiment, and will not be repeated here.

[0159] S502: Combine the pre-processed user features of the user with the product features of the target product to obtain combined features of the user features and the product features.

[0160] Specifically, the product features may be directly concatenated after the user features, thereby obtaining a combination feature of the user features and the product features.

[0161] S503: Perform feature processing on the combined features to obtain a combined feature vector of the combined features.

[0162] It should be noted that the method of performing feature processing on the combined features of the combined samples is the same as the method of performing feature processing on the combined features of the clustered samples, so reference may be made to step S203 in the above embodiment accordingly and will not be repeated here.

[0163] S404: Calculate the cluster to which the combined sample belongs by using the weighted combined feature vector of the combined feature vector.

[0164] Among them, multiple clusters are obtained by clustering the weighted combined feature vectors of the cluster samples constructed in the above embodiments. The weighted combined feature vector of each cluster sample is obtained by weighting the combined feature vector of the cluster sample and the sample weight of the cluster sample. One cluster sample corresponds to one user and one product converted by the user.

[0165] Optionally, the weighted combined feature vector of the combined feature vector and the weighted combined feature vectors of multiple clustering samples may be directly used for clustering, thereby obtaining multiple clusters, and then filtering out the cluster to which the combined sample belongs from the multiple clusters.

[0166] Alternatively, see Figure 6 , the cluster to which the combined sample belongs is calculated by using the weighted combined feature vector of the combined feature vector in the following manner, the method specifically comprising:

[0167] S601, clustering is performed using weighted combined feature vectors of multiple clustering samples to obtain multiple clusters.

[0168] Optionally, a K-means clustering algorithm may be used for clustering. Of course, other clustering algorithms, such as the EM algorithm, may also be used.

[0169] For , K-means clustering algorithm is used for clustering, and clustering is terminated when the criterion function converges. The formula of the criterion function is Among them, J is the cohesion, which is used to measure the clustering effect, k is the total number of clusters, is the jth sample in cluster i; is the center vector of cluster i, and its calculation formula is: m i is the total number of samples in cluster i. It can be seen that the prior art treats each sample equally during the clustering process. For sample and the center vector of cluster i The similarity can be calculated using the cosine of the vector angle.

[0170] It should be noted that in the embodiment of the present application, clustering is performed taking into account the sample weights, and the calculation formula of the clustering criterion function is also:

[0171] However, the formula is the center vector of cluster i after weighting the cluster samples, so its calculation formula is: Among them, w j is the weight of cluster sample j, no longer multiplied by Since the meanings of other parameters are the same as those of the parameters in the formula of the first criterion function mentioned above, they will not be described in detail.

[0172] S602: Using the weighted combined feature vector of the combined feature vector, calculate the distance between the combined sample and the cluster center of each cluster.

[0173] Optionally, the cosine of the vector angle may be used to calculate the distance between the weighted combined feature vector of the combined feature vector and the vector of the cluster center of each cluster.

[0174] S603: Determine the cluster corresponding to the cluster center with the shortest distance to the combined sample as the cluster to which the combined sample belongs.

[0175] S405: Recommend the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs.

[0176] Due to the influence of sample weights, the clustering process places more emphasis on the commonality of user behavior. In addition, since the combined feature vector is constructed using product features and user features, the clustering process also fully considers the correlation between products and users. Therefore, by recommending the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs, the target product can be accurately recommended to the user who is most likely to convert the product, thereby effectively improving the conversion rate of the target product.

[0177] Optionally, after calculating the cluster to which the combined sample belongs, before executing step S405, the following may be further included:

[0178] Calculate the conversion rate ratio of users corresponding to each cluster sample in the cluster to which the combined sample belongs.

[0179] It should be noted that the higher the user's conversion rate is, the higher the rate at which the user ultimately purchases the product after generating a behavioral sequence for the product. Therefore, if the target product is recommended to the user, the user is most likely to purchase the product. Therefore, the specific implementation method of step S405 is: recommend the target product to the user with the highest conversion rate.

[0180] That is to say, the target product is not recommended to the user corresponding to each cluster sample in the cluster to which the combined sample belongs. In order to ensure that the target product is recommended to other users, multiple combined samples can be constructed for the target product, that is, the method provided by this application is executed for each user who generates a user behavior sequence for the target product, so as to recommend the target product to the user with the highest conversion rate determined each time.

[0181] Another embodiment of the present application provides a product recommendation device, see Figure 7 , the device comprises:

[0182] The first acquisition unit 701 is used to acquire product features of a target product and user features and user behavior sequences of a user.

[0183] Among them, the user behavior referred to by the user behavior sequence generated by the user for the target product.

[0184] The configuration unit 702 is used to configure the sample weight of the combination sample of the user and the target product based on the matching result between the user behavior sequence and each determined target sequence pattern.

[0185] Among them, the matching result shows that when the user behavior sequence matches the target sequence pattern, the sample weight of the combined sample is higher; the target sequence pattern is a frequent sequence pattern obtained by performing sequence pattern mining on the user behavior sequence generated when the user converts multiple products.

[0186] The weighting unit 703 is used to weight the combined feature vector constructed based on the combined features of the user features and the product features by using the sample weight of the combined sample to obtain a weighted combined feature vector of the combined feature vector.

[0187] The first calculation unit 704 is used to calculate the cluster to which the combined sample belongs by using the weighted combined feature vector of the combined feature vector.

[0188] The cluster is obtained by clustering the weighted combination feature vectors of multiple cluster samples, and the weighted combination feature vector of each cluster sample is obtained by weighting the combination feature vector of the cluster sample and the sample weight of the cluster sample. One cluster sample corresponds to one user and one product converted by the user.

[0189] The recommendation unit 705 is used to recommend the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs.

[0190] Optionally, in a product recommendation device provided in another embodiment of the present application, the first calculation unit 704 includes: a first clustering unit, a distance calculation unit and a determination unit.

[0191] The first clustering unit is used to perform clustering using weighted combined feature vectors of multiple clustering samples to obtain multiple clusters.

[0192] The distance calculation unit is used to calculate the distance between the combined sample and the cluster center of each cluster by using the weighted combined feature vector of the combined feature vector.

[0193] The determination unit determines the cluster corresponding to the cluster center with the shortest distance to the combined sample as the cluster to which the combined sample belongs;

[0194] Optionally, in a product recommendation device provided by another embodiment of the present application, the first computing unit 704 includes: a first computing subunit, used to perform clustering using a weighted combined feature vector of a combined feature vector and a weighted combined feature vector of a plurality of clustering samples to obtain a plurality of clusters, and to filter out the cluster to which the combined sample belongs from the plurality of clusters.

[0195] Optionally, in the recommended device of the product provided in another embodiment of the present application, it also includes:

[0196] The second calculation unit is used to calculate the conversion rate ratio of users corresponding to each cluster sample in the cluster to which the combined sample belongs.

[0197] Among them, in the embodiment of the present application, when the recommendation unit executes to recommend the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs, it is used to: recommend the target product to the user with the highest conversion rate.

[0198] Optionally, in another embodiment of the present application, the product recommendation device further includes a feature vector construction unit. Figure 8 , feature vector building unit, including the following units:

[0199] The preprocessing unit 801 is used to perform multiple preprocessing on the user characteristics of the user and the product characteristics of the target product, wherein the preprocessing includes filtering processing, filling processing and derivation processing.

[0200] The combining unit 802 is used to combine the pre-processed user features of the user with the product features of the target product to obtain a combined feature of the user features and the product features.

[0201] The feature processing unit 803 is used to perform feature processing on the combined feature to obtain a combined feature vector of the combined feature.

[0202] Optionally, in a recommendation device for a product provided by another embodiment of the present application, the configuration unit 702 is used to configure the support rate of the longest target sequence pattern contained in the user behavior sequence as the combined sample weight when the matching result indicates that the user behavior sequence includes the target sequence pattern; when the matching result indicates that the user behavior sequence does not include any target sequence pattern, the configuration unit 702 is used to configure the combined sample weight to the minimum support rate set when mining the sequence pattern, or to the support rate of the user behavior sequence in each conversion behavior sequence corresponding to the user.

[0203] Optionally, in the product recommendation device provided in another embodiment of the present application, it may further include: a target sequence pattern determination unit.

[0204] The target sequence pattern determination unit may include: a search unit that searches for each frequent sequence pattern of a user from the frequent sequence patterns of multiple users mined in advance, and determines each frequent sequence pattern of the user as the target sequence pattern.

[0205] Alternatively, the target sequence pattern determination unit may also include: a mining unit for performing sequence pattern mining on the behavior sequence generated when the user converts to multiple products, obtaining the user's frequent sequence pattern, and determining the user's frequent sequence pattern as the target sequence pattern.

[0206] Optionally, in the recommended device of the product provided by another embodiment of the present application, see Fig. 9 , the mining unit includes:

[0207] The second acquisition unit 901 is used to respectively acquire the behavior sequences generated when the user converts to each product, and compose each behavior sequence into a behavior sequence set, wherein each behavior sequence consists of multiple behavior codes, and one behavior code refers to one user behavior.

[0208] The first-level sequence pattern mining unit 902 is used to determine each behavior coding in the behavior sequence set whose support is greater than or equal to the support threshold as a first-level sequence pattern, and set the sequence level N to 2.

[0209] The third acquisition unit 903 is used to obtain a projection behavior sequence set corresponding to each N-1 level sequence pattern.

[0210] The projected behavior sequence set corresponding to the N-1 level sequence pattern includes the N-1 level suffixes obtained from each behavior sequence in the behavior sequence set.

[0211] The sequence pattern mining unit 904 is used to combine each behavior coding whose support is greater than or equal to the support threshold in the corresponding projected behavior sequence set and the N-1 level sequence pattern corresponding to the projected behavior sequence set into an N-level sequence pattern, and then increment N by 1 and return to execute to obtain the projected behavior sequence set corresponding to each N-1 level sequence pattern until there is no behavior coding whose support is greater than or equal to the support threshold in the corresponding projected behavior sequence set.

[0212] The sequence pattern determining unit 905 is used to determine each level of obtained sequence pattern as a frequent sequence pattern of the user.

[0213] It should be noted that the specific working process of each unit in the product recommendation device provided in each of the above embodiments can refer to the specific implementation process of the steps in the above method embodiments, which will not be repeated here.

[0214] Another embodiment of the present application provides a computer storage medium for storing a computer program. When the computer program is executed, it is used to implement a product recommendation method provided in any one of the above embodiments.

[0215] The present application also provides an electronic device, such as Fig.10 As shown, the electronic device includes a memory 1001 and a processor 1002 .

[0216] Wherein, the memory 1001 is used to store computer programs;

[0217] The processor 1002 is used to execute the above-mentioned computer program, specifically to implement the product recommendation method provided in any embodiment of the present application.

[0218] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0219] These computer program instructions may also be stored in a computer readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture including an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0220] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0221] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0222] The memory may include non-permanent memory in a computer-readable medium, random access memory (RAM) and / or non-volatile memory in the form of read-only memory (ROM) or flash RAM. The memory is an example of a computer-readable medium.

[0223] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0224] Those skilled in the art will appreciate that the embodiments of the present application may be provided as methods, systems or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0225] The above description is only a preferred embodiment of the present application and an explanation of the technical principles used, and is not intended to limit the present application. For those skilled in the art, the present application may have various modifications and variations. The scope of the invention involved in the present application is not limited to the technical solution formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, a technical solution formed by replacing the above-mentioned features with the technical features with similar functions disclosed in this application (but not limited to).

Claims

1. A product recommendation method, characterized in that: include: Acquire product features of a target product and user features and a user behavior sequence of a user; wherein the user generates a user behavior indicated by the user behavior sequence for the target product; Based on the matching results of the user behavior sequence and each determined target sequence pattern, a sample weight of a combined sample of the user and the target product is configured; wherein the matching result indicates that when the user behavior sequence matches the target sequence pattern, the sample weight of the combined sample is higher; the target sequence pattern is a frequent sequence pattern obtained by performing sequence pattern mining on the user behavior sequence generated when the user converts multiple products; Using the sample weight of the combined sample, weighting the combined feature vector constructed based on the combined features of the user feature and the product feature to obtain a weighted combined feature vector of the combined feature vector; The weighted combined feature vector of the combined feature vector is used to calculate the cluster to which the combined sample belongs; wherein the cluster is obtained by clustering the weighted combined feature vectors of multiple cluster samples, and the weighted combined feature vector of each cluster sample is obtained by weighting the combined feature vector of the cluster sample and the sample weight of the cluster sample; one cluster sample corresponds to one user and one product converted by the user; The target product is recommended to users corresponding to the cluster samples in the cluster to which the combined sample belongs.

2. The method according to claim 1, characterized in that The step of calculating the cluster to which the combined sample belongs by using the weighted combined feature vector of the combined feature vector comprises: Performing clustering using weighted combined feature vectors of a plurality of cluster samples to obtain a plurality of clusters; Using the weighted combined feature vector of the combined feature vector, the distance between the combined sample and the cluster center of each cluster is calculated; Determine the cluster corresponding to the cluster center having the shortest distance to the combined sample as the cluster to which the combined sample belongs; Alternatively, clustering is performed using the weighted combined feature vector of the combined feature vector and the weighted combined feature vectors of a plurality of cluster samples to obtain a plurality of clusters, and the cluster to which the combined sample belongs is screened out from the plurality of clusters.

3. The method according to claim 1, characterized in that After calculating the cluster to which the combined sample belongs by using the weighted combined feature vector of the combined feature vector, the method further includes: Calculate the conversion rate ratio of users corresponding to each of the cluster samples in the cluster to which the combined sample belongs; The step of recommending the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs includes: The target product is recommended to the user with the highest conversion rate.

4. The method according to claim 1, characterized in that: The method for constructing the combined feature vector comprises: Performing multiple preprocessing on the user characteristics of the user and the product characteristics of the target product; wherein the preprocessing includes filtering processing, filling processing and derivation processing; The preprocessed user features of the user are combined with the product features of the target product to obtain a combined feature of the user features and the product features; Feature processing is performed on the combined features to obtain a combined feature vector of the combined features.

5. The method according to claim 2, characterized in that: The configuring, based on the matching results of the user behavior sequence and each target sequence pattern, to obtain the sample weight of the combined sample of the user and the target product comprises: If the matching result indicates that the user behavior sequence contains the target sequence pattern, configuring the support rate of the longest target sequence pattern contained in the user behavior sequence as the combined sample weight; If the matching result indicates that the user behavior sequence does not contain any of the target sequence patterns, the combined sample weight is configured as the minimum support rate set when mining sequence patterns, or as the support rate of the user behavior sequence in each of the conversion behavior sequences corresponding to the user.

6. The method according to claim 1, characterized in that The method for determining each target sequence pattern comprises: Finding each frequent sequence pattern of the user from the frequent sequence patterns of multiple users mined in advance, and determining each frequent sequence pattern of the user as the target sequence pattern; Alternatively, sequence pattern mining is performed on the behavior sequence generated when the user converts to multiple products to obtain the user's frequent sequence pattern, and the user's frequent sequence pattern is determined as the target sequence pattern.

7. The method according to claim 6, characterized in that The performing sequence pattern mining on the behavior sequence generated when the user converts to the plurality of sample products to obtain the frequent sequence pattern of the user includes: Obtaining the behavior sequences generated when the user converts to each product respectively, and composing each of the behavior sequences into a behavior sequence set; wherein each of the behavior sequences consists of a plurality of behavior codes; and one of the behavior codes refers to one user behavior; Determine each behavior code in the behavior sequence set whose support is greater than or equal to the support threshold as a level 1 sequence pattern, and set the sequence level N to 2; Obtaining a projected behavior sequence set corresponding to each N-1 level sequence pattern; wherein the projected behavior sequence set corresponding to the N-1 level sequence pattern includes the N-1 level suffix obtained from each of the behavior sequences in the behavior sequence set; Combine each behavior code whose support in the corresponding projected behavior sequence set is greater than or equal to the support threshold and the N-1 level sequence pattern corresponding to the projected behavior sequence set into an N-level sequence pattern, and increment N by 1 and then return to execute to obtain the projected behavior sequence set corresponding to each N-1 level sequence pattern until there is no behavior code whose support in the corresponding projected behavior sequence set is greater than or equal to the support threshold; The obtained sequence pattern of each level is determined as the frequent sequence pattern of the user.

8. A product recommendation device, characterized in that: include: A first acquisition unit is used to acquire product features of a target product and user features and a user behavior sequence of a user; wherein the user generates a user behavior indicated by the user behavior sequence for the target product; a configuration unit, configured to configure a sample weight of a combined sample of the user and the target product based on a matching result between the user behavior sequence and each determined target sequence pattern; wherein the matching result indicates that when the user behavior sequence matches the target sequence pattern, the sample weight of the combined sample is higher; and the target sequence pattern is a frequent sequence pattern obtained by performing sequence pattern mining on the user behavior sequence generated when the user converts multiple products; A weighting unit, configured to weight a combined feature vector constructed based on the combined features of the user features and the product features using the sample weight of the combined sample, to obtain a weighted combined feature vector of the combined feature vector; A first calculation unit is used to calculate the cluster to which the combined sample belongs by using the weighted combined feature vector of the combined feature vector; wherein the cluster is obtained by clustering the weighted combined feature vectors of multiple cluster samples, and the weighted combined feature vector of each cluster sample is obtained by weighting the combined feature vector of the cluster sample and the sample weight of the cluster sample; one cluster sample corresponds to one user and one product converted by the user; A recommendation unit is used to recommend the target product to the user corresponding to the cluster sample in the cluster to which the combined sample belongs.

9. A computer storage medium, characterized in that Used to store a computer program, which, when executed, is used to implement the product recommendation method according to any one of claims 1 to 7.

10. An electronic device, characterized in that: including memory and processor; Wherein, the memory is used to store programs; The processor is used to execute the program, and when the program is executed, it is specifically used to implement the product recommendation method according to any one of claims 1 to 7.

Images