Object recommendation method and device, computing equipment and computer readable storage medium

Abstract

Embodiments of the invention provide an object recommendation method and apparatus, a computing device and a computer readable storage medium. The object recommendation method comprises the steps of obtaining object features of a target object and object features of a reference object; determining similar objects from the reference objects based on the feature similarity between the object features of the target object and the object features of the reference objects; determining the preference information between the candidate user and the target object based on the preference information between the candidate user and the similar object and the feature similarity between the object feature of the target object and the object feature of the similar object; and determining a target user from the candidate users based on the preference information between the candidate users and the target object, and recommending the target object to the target user. The target users having potential preferences for the target object are mined, and the potential user recommendation of the target object completes the user broken circle of the target object, so that the object recommendation of the potential users can be accurately, flexibly and fully mined.

Description

Technical Field

[0001] The embodiments of this specification relate to the technical field of big data, and more particularly to an object recommendation, apparatus, computing device, and computer-readable storage medium. Background Art

[0002] With the development of big data technology, object recommendation has shifted from traditional wide-ranging recommendation to accurate recommendation based on feature analysis.

[0003] At present, the traditional funnel recommendation method matches objects with corresponding features based on the user's user features. This method has insufficient generalization ability, and the objects recommended to users are likely to be related objects that they have browsed before, resulting in the inability of object recommendations to break through the circle.

[0004] In view of the above problems, how to effectively achieve user cross-circle recommendation is a problem. One way is to use the method of similarity expansion between users (co-extension) to achieve user cross-circle, but this method will only find new user groups with highly similar characteristics to existing users, resulting in poor cross-circle effect and failure to fully tap potential users for recommendation. Another way is to calculate the matching degree between all users and all objects for recall, which lacks the accuracy of recommendations for specific objects, making the entire process too black-box. Due to the isolation between the operation strategy and the matching recall algorithm, the formulation of the actual operation strategy is inconvenient and cannot be adjusted quickly and flexibly. Therefore, there is an urgent need for an accurate and flexible object recommendation method that can fully tap potential users. Summary of the invention

[0005] In view of this, an embodiment of the present specification provides an object recommendation method. One or more embodiments of the present specification also relate to an object recommendation device, a computing device, a computer-readable storage medium, and a computer program product to solve the technical defects existing in the prior art.

[0006] According to a first aspect of an embodiment of this specification, there is provided an object recommendation method, including: Obtaining object features of the target object and object features of the reference object; determining similar objects from the reference objects based on feature similarities between object features of the target object and object features of the reference object; determining preference information between the candidate user and the target object based on preference information between the candidate user and the similar objects and feature similarity between an object feature of the target object and an object feature of the similar objects; Based on the preference information between the candidate users and the target objects, the target user is determined from the candidate users, and the target object is recommended to the target user.

[0007] According to a second aspect of an embodiment of this specification, there is provided an object recommendation device, including: An acquisition module, configured to acquire object features of a target object and object features of a reference object; a similarity determination module configured to determine similar objects from the reference objects based on feature similarities between object features of the target object and object features of the reference object; A user matching module configured to determine preference information between the candidate user and the target object based on preference information between the candidate user and the similar objects and feature similarity between object features of the target object and object features of the similar objects; The recommendation module is configured to determine a target user from the candidate users based on preference information between the candidate users and the target object, and recommend the target object to the target user.

[0008] According to a third aspect of an embodiment of this specification, a computing device is provided, including: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the above method are implemented.

[0009] According to a fourth aspect of the embodiments of this specification, a computer-readable storage medium is provided, which stores a computer program / instruction, and the steps of the above method are implemented when the computer program / instruction is executed by a processor.

[0010] According to a fifth aspect of the embodiments of this specification, a computer program product is provided, comprising a computer program / instruction, which implements the steps of the above method when executed by a processor.

[0011] In one embodiment of the present specification, the object features of the target object and the object features of the reference object are obtained, and based on the feature similarity between the object features of the target object and the object features of the reference object, similar objects are determined from the reference objects, similar objects are recalled, and feature similarity data support is provided for the subsequent determination of preference information. Based on the preference information between the candidate user and the similar objects, and the feature similarity between the object features of the target object and the object features of the similar objects, the preference information between the candidate user and the target object is determined, and similar objects are used as a bridge to determine the preference information between the candidate user and the target object by reference to the candidate user's preference information for similar objects. Based on the preference information between the candidate user and the target object, the target user is determined from the candidate users, and the target object is recommended to the target user, and the target user who has a potential preference for the target object is excavated. The potential user recommendation of the target object completes the user breakthrough of the target object, and realizes accurate, flexible, and fully excavated object recommendation of potential users. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a flow chart of an object recommendation method provided by an embodiment of this specification; Figure 2 It is a schematic diagram of the architecture of an object recommendation method provided by an embodiment of this specification; Figure 3 is a flowchart of an object recommendation method provided by an embodiment of this specification; Figure 4 is a process flow chart of an object recommendation method for mining a marketing population of a target SPU provided by an embodiment of the present specification; Figure 5 is a structural diagram of an object recommendation device provided by an embodiment of this specification; Figure 6 It is a structural block diagram of a computing device provided by an embodiment of this specification. DETAILED DESCRIPTION

[0013] Many specific details are described in the following description to facilitate a full understanding of this specification. However, this specification can be implemented in many other ways than those described herein, and those skilled in the art can make similar generalizations without violating the connotation of this specification, so this specification is not limited to the specific implementation disclosed below.

[0014] The terms used in one or more embodiments of the present invention are only for the purpose of describing specific embodiments, and are not intended to limit one or more embodiments of the present invention. The singular forms of "a", "said" and "the" used in one or more embodiments of the present invention are also intended to include plural forms, unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used in one or more embodiments of the present invention refers to and includes any or all possible combinations of one or more associated listed items.

[0015] It should be understood that, although the terms first, second, etc. may be used to describe various information in one or more embodiments of the present invention, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of one or more embodiments of the present invention, the first may also be referred to as the second, and similarly, the second may also be referred to as the first. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0016] In addition, it should be noted that the data involved in one or more embodiments of the present invention are information and data authorized by the user or fully authorized by all parties, and the statistics, use and processing of the relevant data need to comply with the relevant laws, regulations and standards of the relevant countries and regions, and provide corresponding operation entrances for users to choose to authorize or refuse.

[0017] First, the terms involved in one or more embodiments of this specification are explained.

[0018] Standard Product Unit (SPU) is the basic unit of product information aggregation. It is a set of reusable and easily searchable standardized information that describes the characteristics of a product.

[0019] Funnel user group: refers to the target user group selected through traditional marketing strategies. These strategies are usually based on the user's existing behavioral data (such as browsing history, purchase history, etc.), and use a wide-net approach to gradually narrow the scope of potential customers, similar to the marketing funnel model. The characteristic of this user group is that the objects recommended to them are likely to be limited to those they have previously been interested in, making it difficult to break through the existing interest circle, limiting the possibility of wider exploration and discovery.

[0020] Reverse funnel user group: Compared with the traditional concept of funnel user group, the reverse funnel user group refers to the potential target user group identified through innovative marketing strategies. This method is not limited to the historical behavior patterns of users, but attempts to tap into potential users who have not been reached by traditional marketing methods by analyzing the behavioral similarities and preference differences between users and combining specific object characteristics. The reverse funnel user group emphasizes breaking the information cocoon effect and encouraging users to reach a wider range of objects, thereby expanding and diversifying the user group and improving the effectiveness and coverage of marketing activities.

[0021] SparkSQL: Designed for processing structured data and supporting SQL queries, allowing users to perform efficient data processing and analysis through SQL or DataFrame API.

[0022] In this specification, an object recommendation method is provided. This specification also relates to an object recommendation device, a computing device, a computer-readable storage medium and a computer program product, which are described in detail one by one in the following embodiments.

[0023] See also Figure 1 , Figure 1 A flowchart of an object recommendation method provided by an embodiment of the present specification is shown, including the following specific steps: Step 102: Obtain object features of the target object and object features of the reference object.

[0024] The embodiments of this specification are applied to a system platform of a big data analysis and personalized recommendation system platform, such as an e-commerce platform, a social media platform, or a community content application.

[0025] The target object is a specific object that needs to be recommended. The target object can be a specific object entity, including but not limited to: goods, services and content, for example, a certain product or service that needs to be marketed, an advertisement that needs to be marketed, or an article, audio, or video content that needs to be pushed. The target object can also be a basic unit of object information aggregation, for example, the SPU of goods or services, or a collection of a type of articles, audio, or video.

[0026] The object features of the target object are a set of information that defines the object attributes of the target object. The object features are used to compare with the object features of other objects to determine similarity and preference matching. The object features of the target object include, but are not limited to, the visual appearance, type, label, keyword, functional characteristics, etc. of the object. For example, the brand name, model, price, function, applicable population, etc. of a certain product; another example is the target audience, distribution channel, creative elements, expected effect, etc. of an advertisement; another example is the release time, theme, author, keyword, label, etc. of an article, audio, or video content that needs to be pushed. Object features can be represented in the form of a coding vector for calculating subsequent feature similarity.

[0027] The reference object is a set of objects that can be used as a reference to complete object recall. Similar to the target object, any reference object can be a specific object entity or a basic unit of object information aggregation.

[0028] The object features of the reference object are a set of information that defines the object attributes of the reference object. The object features of the reference object are used to compare with the object features of the target object to determine similarity and preference matching. Similar to the target object, the object features of the reference object may include but are not limited to the visual appearance, type, label, keyword, functional characteristics, etc. of the object. The object features can be represented in the form of a coding vector for calculating subsequent feature similarity.

[0029] To obtain the object features of the target object, one optional method is to directly obtain the object features of the target object from the system through database query or data interface call. For example, in an e-commerce platform, the brand name, model, price and other detailed information of a certain product can be directly obtained through the product management system. Another optional method is to crawl the object features of the target object through crawler technology, for example, crawl user comments, ratings and other data from social media platforms or other e-commerce websites as the object features of the target object. Another optional method is to use a feature extraction algorithm to extract the object features of the target object. For example, for an article or video, its main content and discussion topics can be extracted through a natural language algorithm.

[0030] To obtain the object characteristics of the reference object, one optional method is to directly obtain the object characteristics of the reference object from the system through database query or data interface call, for example, the e-commerce platform batch imports detailed information of all products on sale, including brand, model, price, etc. Another optional method is to capture the object characteristics of the reference object through crawler technology, for example, through crawler technology, batch capture user comments, ratings and other data from social media platforms or other e-commerce websites as the object characteristics of the reference object. Another optional method is to use a feature extraction algorithm to extract the object characteristics of the reference object. For example, for all articles or videos in the database, the main content and discussion topics can be extracted through a natural language algorithm, which is not limited here.

[0031] For example, in a community content application, it is necessary to promote m target SPUs of multiple merchant brands according to business rules. The m target SPUs build a target SPU list, and obtain the embedded coding vectors of the brand name, model, price, function, applicable population and other features of the m target SPUs through the product management system: , and obtain the embedded coding vectors of the brand name, model, price, function, applicable population and other features of N candidate SPUs through the product management system: ,in, Represents a reference object, Represents a reference object The object type.

[0032] Obtaining the object features of the target object and the object features of the reference object provides data support for the object set and the basis for recalling similar objects in the subsequent determination of similar objects from the reference object.

[0033] Step 104: Determine similar objects from the reference objects based on the feature similarity between the object features of the target object and the object features of the reference objects.

[0034] The feature similarity between the object features of the target object and the object features of the reference object is a quantitative indicator of the degree of object association between the target object and the reference object. The calculation method of the feature similarity includes but is not limited to: cosine similarity, Jaccard similarity coefficient, Euclidean distance, etc. The feature similarity between the object features of the target object and the object features of the reference object can be a single dimension, multiple dimensions, or multimodal feature similarity combined with visual space.

[0035] The similar objects are specific objects recalled from the reference objects through similar objects, and are objects selected from the reference objects and having a high feature similarity with the target object.

[0036] Based on the feature similarity between the object features of the target object and the object features of the reference object, similar objects are determined from the reference objects. An optional method is: based on the feature similarity between the object features of the target object and the object features of the reference object and a preset threshold, similar objects are determined from the reference objects, wherein the preset threshold can be a feature similarity threshold or an object number threshold, which is not limited here.

[0037] Exemplarily, based on the embedded coding vector of the target SPU: And the embedded coding vector of the candidate SPU: Calculate the feature similarity: Based on feature similarity and a preset threshold, a set of n similar SPUs is recalled from N candidate SPUs: ,in, Indicates the number of similar SPUs recalled.

[0038] Based on the feature similarity between the object features of the target object and the object features of the reference object, similar objects are determined from the reference objects, similar object recall is completed, and feature similarity data support is provided for the subsequent determination of preference information.

[0039] Step 106: Determine the preference information between the candidate user and the target object based on the preference information between the candidate user and the similar objects and the feature similarity between the object feature of the target object and the object feature of the similar objects.

[0040] Candidate users are user groups for whom object recommendations can be made. Target users need to be determined from among the candidate users to complete object recommendations. Candidate users can be user groups of all users or partial user groups selected based on business needs, which is not limited here.

[0041] The preference information between the candidate user and the similar objects is a quantitative indicator that characterizes the candidate user's preference for the similar objects, including but not limited to: preference score, click rate, conversion rate, etc.

[0042] The feature similarity between the object features of the target object and the object features of the similar objects is a quantitative indicator of the degree of object association between the target object and the similar objects. The calculation methods of the feature similarity include but are not limited to: cosine similarity, Jaccard similarity coefficient, Euclidean distance, etc. The feature similarity between the object features of the target object and the object features of the similar objects can be a single dimension, multiple dimensions, or a multimodal feature similarity combined with a visual space. The feature similarity between the object features of the target object and the object features of the similar objects is included in the feature similarity between the object features of the target object and the object features of the reference object in the above step 104.

[0043] The preference information between the candidate user and the target object is a quantitative indicator that characterizes the candidate user's preference for the target object, including but not limited to: preference score, weighted click rate, weighted conversion rate, etc.

[0044] Since both feature similarity and preference information are quantitative indicators, similar objects are used as a bridge in step 106 to determine the quantitative indicator of preference information between candidate users and target objects. Compared with the method of expanding similarity between users (cooperative expansion) to achieve user breakthrough, it avoids the formation of a gradually narrowing and solidified funnel user group. It can not only determine users who have a direct interest in the target object, but also identify those who may become potential users due to their high preference for similar objects although they have not directly shown interest. This greatly expands the scope of target users determined subsequently, promotes the diversification of the user group of target users, forms an anti-funnel user group, and achieves more extensive and effective object recommendation.

[0045] Based on the preference information between the candidate user and the similar objects, and the feature similarity between the object feature of the target object and the object feature of the similar objects, the preference information between the candidate user and the target object is determined. An optional formula is shown in Formula 1: Formula 1 in, Characterize candidate users, Characterize the target object, Representing similar objects, Represents the preference information between candidate users and target objects, Characterizes the feature similarity between the object features of the target object and the object features of similar objects, Characterize the preference information between the candidate user and the similar object. Simple linear calculations such as weighting, adding and subtracting constants, or nonlinear calculations such as summation and integration can be performed based on Formula 1, which are all protected in the embodiments of this specification and are not described here.

[0046] For example, based on the preference distribution between i candidate users and n similar SPUs, , and the feature similarity between m target SPUs and n similar SPUs , using the above formula 1, calculate the preference score between i candidate users and m target SPUs , select the one with the highest preference score is the preference score of i candidate users for the target SPU: .

[0047] Using similar objects as a bridge, by referring to and evaluating the preference information of candidate users for similar objects, the preference information between candidate users and target objects is determined, providing data support for the subsequent determination of target users from candidate users.

[0048] Step 108: Based on the preference information between the candidate users and the target object, determine the target user from the candidate users, and recommend the target object to the target user.

[0049] The target user is the user who makes the object recommendation. The target user is obtained by screening from the candidate users. The target user has a high probability of showing a preference interest in the target object and can be identified as a user with potential interest.

[0050] Based on the preference information between the candidate users and the target object, the target user is determined from the candidate users. An optional method is: based on the preference information between the candidate users and the target object and a preset threshold, the target user is determined from the candidate users, wherein the preset threshold can be a preference information threshold or a user number threshold, which is not limited here.

[0051] One optional way to recommend target objects to target users is to display recommended content directly in the user interface through a personalized recommendation system. For example, in an e-commerce platform, recommended products can be displayed to users through the user's personal homepage, product recommendation column, or email notification. For another example, on a social media platform, recommended content can be displayed through a recommendation position in a dynamic stream or a private message push. For another example, in a community content application, a contextual recommendation pop-up window triggered by user behavior can pop up relevant recommendations in real time when a user browses a specific type of content. For example, when a user reads an article about "outdoor camping", the sidebar dynamically loads a link to purchase camping equipment or video content on the same topic. Another optional method is to recommend target objects to target users across devices and platforms. For example, after a user browses a food tutorial video on the mobile terminal of a community content application, related kitchenware product advertisements are recommended to the smart TV terminal in the user's home through cross-device ID mapping. Another optional method is to recommend target objects to target users through their user relationships. For example, in a community content application, when a certain content is collected by a user in the user's interest group, the content is automatically pushed to the private message list of all members of the group. For example, by analyzing the common preference characteristics of the user's social circle, a special recommendation column called "Friends are also paying attention" is generated on the user's personal homepage, dynamically aggregating high-interaction content from similar user groups. This is not limited here.

[0052] For example, based on the preference scores of i candidate users for the target SPU: and a preset user quantity threshold, and determine the anti-funneling user group of the target users of each target SPU from the i candidate users.

[0053] In the embodiments of the present specification, similar objects are used as a bridge, and the preference information between the candidate users and the target object is determined by reference and evaluation. Based on the preference information between the candidate users and the target object, the target user is determined from the candidate users, and the target object is recommended to the target user, thereby discovering the target users who have potential preferences for the target object. The potential user recommendation of the target object completes the user circle of the target object, and realizes accurate and flexible object recommendation that can fully tap potential users.

[0054] In an optional embodiment of the present specification, step 104 includes the following specific steps: Based on the feature similarity between the object feature of the target object and the object feature of the reference object, and a preset feature similarity threshold, similar objects are determined from the reference objects.

[0055] The preset feature similarity threshold is a dynamic value determined based on business needs or historical data, and is used to filter reference objects with significant similarity. The threshold can be determined in the following ways: Empirical threshold: Set a benchmark value based on domain knowledge, such as cosine similarity ≥ 0.7.

[0056] Dynamic adjustment: Verify the conversion rate under different thresholds through A / B testing and select the optimal value.

[0057] Distribution analysis: Count the quantiles of the feature similarity of the reference object and take the top 20% as the threshold.

[0058] By presetting the feature similarity threshold, the white-box recommendation process control is completed. This method not only improves the transparency of the recommendation system, but also enables operators to flexibly adjust the recommendation strategy according to actual business needs, ensuring that the recommendation results are in line with user interests and meet business goals.

[0059] For example, based on feature similarity and the preset feature similarity threshold =0.75, recall a set of n similar SPUs from N candidate SPUs: .

[0060] In the embodiments of the present specification, based on the feature similarity between the object features of the target object and the object features of the reference object, and a preset feature similarity threshold, similar objects are accurately and effectively recalled from the reference objects, providing accurate and effective object support for subsequent determination of preference information.

[0061] In an optional embodiment of the present specification, step 104 includes at least one of the following: Determining similar objects from the reference objects based on object features of the target object and object features of the reference object and feature similarities between object dimensions; Determine similar objects from the reference objects based on the object characteristics of the target object and the object characteristics of the reference object and the characteristic similarity between the brand dimensions; Based on the object features of the target object and the object features of the reference object, and the feature similarity between the category dimensions, similar objects are determined from the reference objects.

[0062] The object dimension is a feature set dimension that defines the basic attributes of the object entity, focusing on the physical characteristics and functional parameters of the object itself. The object dimension of an object includes: visual appearance, model, functional parameters, technical specifications, and usage scenario characteristics. The feature similarity in the object dimension is a quantitative indicator of the degree of object association between two objects in the object dimension. A certain dress is a simple black dress in visual appearance, with a straight cut and sleeveless design, and a low-key metal zipper as a decorative element. Another object with a high similarity in visual appearance may be a black vest dress with the same simple style, with a similar straight cut, but with an additional belt to highlight the waistline. The two are very similar in color, style, and overall design style. It can be determined that the two have a high feature similarity in the object dimension of visual appearance. For example, consider two smartphones. Although these two phones belong to different brands and have differences in operating systems, user interfaces, etc., they are both positioned in the high-end market and have almost the same screen size, resolution, refresh rate and other features. At the same time, they also show high consistency in memory capacity, storage options, etc. Based on this information, in the model dimension, these two phones show a high degree of feature similarity.

[0063] The brand dimension is a feature set dimension that defines brand association attributes, focusing on analyzing brand-level cognitive associations and market positioning. For example, although product A and product B are smartphones, they are the same type of products, but they belong to different brands. The feature similarity in the brand dimension is a quantitative indicator of the degree of object association between two objects in the brand dimension.

[0064] The category dimension is a feature set dimension that defines the object classification system, focusing on the hierarchical relationship and semantic association of objects in the classification tree. For example, product A is a smartphone and product C is a smartwatch. Although they are different types of products, they both belong to the category of "electronic equipment". The feature similarity in the category dimension is a quantitative indicator of the degree of object association between two objects in the category dimension.

[0065] It should be noted that by completing the feature similarity calculation for the object features of the target object and the object features of the reference object from three dimensions, namely, object dimension, brand dimension and category dimension, and then recalling similar objects in a mixed manner, we can comprehensively recall as many specific objects as possible that can be used for similar object recall from multiple dimensions, thus providing a broader reference for the subsequent determination of target users.

[0066] Figure 2 A schematic diagram of the architecture of an object recommendation method provided by an embodiment of this specification is shown. Figure 2 As shown: The reverse funnel user group mining for the target object includes two main parts: the recall of similar objects of the target object and the calculation of preference information between the user and the target object.

[0067] Among them, the similarity object recall of the target object includes: First, the feature similarity between objects in three dimensions, namely object dimension, brand dimension and category dimension, is calculated. Secondly, multi-dimensional mixed recall of similar objects is performed to recall similar objects of the target object in three dimensions respectively to form a set of similar objects.

[0068] The calculation of preference information between the user and the target object includes: First, the preference information of the reference object and similar objects is calculated: the candidate users and similar objects are fully connected to calculate the preference information, and the preference information between the 6 candidate users and similar objects is calculated; secondly, the preference information of the reference object and the target object is calculated, and the preference information between the 6 candidate users and the target object is calculated.

[0069] Finally, based on the preference information between the six candidate users and the target objects, the anti-funnel user group consisting of target users is determined from the candidate users.

[0070] Exemplarily, based on the embedded coding vector of the target SPU: And the embedding coding vector of the candidate SPU: Calculate the feature similarity in three dimensions: SPU dimension, brand dimension and category dimension: ,in, Indicates SPU in spu, brand, and category dimensions; . Preset thresholds based on feature similarity and mixed recall =0.75, then sort all candidate SPUs in descending order according to feature similarity and perform mixed recall to obtain a set of n similar SPUs: ,in, Indicates the number of similar SPUs recalled.

[0071] In the embodiments of this specification, through multi-dimensional feature similarity calculation and hybrid recall strategy, not only the recall accuracy and diversity of similar objects are improved, but also potential target user groups are effectively mined. Combining the quantitative indicators of the three dimensions of object, brand, and category, more accurate matching and recommendation of target objects are achieved, which promotes users to break through the circle.

[0072] In an optional embodiment of the present specification, before step 106, the following specific steps are also included: Based on the user behavior data of the candidate user with respect to the similar object, preference information between the candidate user and the similar object is determined.

[0073] The user behavior data of candidate users for similar objects is the record data of the candidate users' interactive behaviors for similar objects, including but not limited to search, interaction (click, favorite, purchase, evaluation, comment, etc.), and browsing. User behavior data reflects the candidate users' interest and preference for similar objects. For example, user 1 browsed product B twice and rated product B 4 stars.

[0074] Based on the user behavior data of the candidate user for similar objects, the preference information between the candidate user and the similar object is determined. One optional method is: using a statistical algorithm, based on the user behavior data of the candidate user for similar objects, the preference information between the candidate user and the similar object is determined, for example, calculating the number of clicks, browsing time, purchase frequency and other indicators of a candidate user on a similar object, and establishing a preference scoring system between each candidate user and the similar object based on these statistical data. The higher the preference score, the greater the candidate user's interest in the similar object. Another optional method is: using a machine learning algorithm, based on the user behavior data of the candidate user for similar objects, the preference information between the candidate user and the similar object is determined, for example, logistic regression, decision tree, random forest, gradient boosting decision tree, neural network, etc., and model training is performed based on historical data sets. The historical data set contains user behavior data with known preferences as input features and corresponding preference scores or labels as output targets, which are not limited here.

[0075] For example, a user behavior set B of i candidate users for n similar SPUs is selected according to business requirements. A statistical algorithm is used based on the user behavior data of i candidate users for n similar SPUs. , determine the preference score between candidate users and similar objects: ,in Represents the user, , .

[0076] In the embodiments of this specification, based on the preference information of candidate users for similar objects, the preference information between candidate users and similar objects is accurately determined, thereby improving the accuracy of object recommendation. Not only is a behavior-based preference scoring system constructed, but also the user's potential interests are deeply mined, and the target object can be effectively identified and recommended to potentially interested users.

[0077] In an optional embodiment of the present specification, the user behavior data includes at least one type of user behavior data of search, interaction and in-depth browsing; Based on the user behavior data of the candidate user for the similar object, determining the preference information between the candidate user and the similar object includes the following specific steps: Based on the user behavior data of the candidate user for the similar object, calculating at least one type of behavior preference score of the candidate user for the similar object; At least one type of behavior preference score is weighted to obtain preference information between candidate users and similar objects.

[0078] Search is the behavior of users searching for objects by entering keywords or other query conditions. For example, a user searches for "wireless headphones" on an e-commerce platform, which indicates that the user may be interested in purchasing wireless headphones.

[0079] Interaction refers to the behavior of users interacting with objects, including but not limited to clicking, collecting, sharing, commenting, etc. For example, a user likes and shares an article about healthy eating on social media, which shows that the user is interested in the topic and is willing to spread it to others.

[0080] In-depth browsing refers to the behavior of users browsing an object for a duration exceeding a threshold. For example, a user watches a video in a community content application for more than 3 minutes.

[0081] At least one type of behavior preference is a quantitative indicator of the candidate user's preference for similar objects under at least one type, which is used to evaluate the user's interest in similar objects and potential behavioral intention under each behavior type. For example, the preference score for deep browsing is higher than the preference score for searching.

[0082] For example, according to similar SPU Types Setting weights for preference scores . Based on the user behavior data set of i candidate users for n similar SPUs , calculate the three types of behavior preference scores of i candidate users for n similar SPUs: ; , The three types of behavior preference scores are weighted to obtain the preference scores between i candidate users and n similar SPUs: .

[0083] In the embodiments of this specification, by comprehensively analyzing the data of candidate users under various behavior types such as search, interaction and in-depth browsing, the behavior preference scores for similar objects are calculated, and these preference scores are weighted to accurately quantify user interests and potential behavior intentions, which not only improves the depth of understanding of user preferences, but also enables object recommendations to more accurately identify high-potential target users. By adjusting the weights of different behavior types, the accuracy and personalization level of recommendations can be effectively improved. At the same time, operators can flexibly adjust the weights of behavior types according to different business needs to achieve more flexible and accurate user group positioning.

[0084] In an optional embodiment of the present specification, before step 106, the following specific steps are also included: Get the user collection; Based on the user recommendation conditions of the target object, candidate users are determined from the user set.

[0085] The user set is a collection of user groups on the platform, including registered users on the platform or active users with behavior records. The user set can be obtained by exporting the full user list from the user database, or dynamically obtaining user groups that meet the activity threshold through the real-time data interface, for example, users who have logged in ≥ 3 times in the past 30 days.

[0086] The user recommendation conditions for the target object are user screening logic set according to business rules, which are used to define the range of recommended users for the target object. For example, if a product is not suitable for minors, the age range needs to be set to over 18 years old. For another example, if a product cannot be provided to foreign users, the geographical location needs to be set in China. User recommendation conditions include but are not limited to: basic attributes: geographical location, age range, gender, device type; behavioral characteristics: historical click-through rate threshold, purchase frequency range, content interaction depth; interest tags: interest classification tags based on user portraits (such as "technology enthusiasts", "mother and baby users"); time window: specify the time range of user behavior (such as search behavior in the past 7 days); exclusion rules: filter users who have purchased the target object or blacklisted users.

[0087] For example, a community content application needs to promote outdoor sports SPU, and the user recommendation conditions set include: 1. Basic attributes: Age 18-45, and located in a city popular for mountain sports; 2. Behavioral characteristics: browsing outdoor-related content ≥ 5 times in the past 30 days, and the single stay time ≥ 60 seconds; 3. Interest tags: "outdoor sports" or "self-driving tour" tags; 4. Exclusion rules: Filter users who have purchased similar products in the past 7 days.

[0088] In the embodiments of this specification, based on the user recommendation conditions of the target object, candidate users are initially screened from the user set, which reduces unnecessary computing resource consumption and processing time, and improves the efficiency and accuracy of the recommendation system. Through refined user screening logic, the system can focus on potential users who are really likely to be interested in the target object, thereby avoiding the waste of resources caused by pushing content to irrelevant users. At the same time, operators can flexibly adjust the screening rules according to different business needs to achieve more flexible and accurate user group positioning.

[0089] Since the preference information between candidate users and target objects needs to be fully connected, the order of magnitude required for calculation is too large, often reaching more than one trillion. Processing such a large amount of data at one time in programming modes such as SparkSQL will result in performance bottlenecks and will not be convenient for subsequent business expansion. Therefore, group parallel computing can be used to convert the computing task of step 106 into a parallel computing task to improve computing efficiency.

[0090] In an optional embodiment of the present specification, before step 106, the following specific steps are also included: According to the number of similar objects, the number of groups is determined, and multiple groups are obtained by dividing according to the number of groups; Step 106 includes the following specific steps: Based on the preference information between the candidate user and the similar objects of the plurality of groups and the feature similarity between the object feature of the target object and the object feature of the reference object, the preference information between the candidate user and the target object is calculated in parallel by group.

[0091] Determine the number of groups based on the number of similar objects, including but not limited to the following methods: Fixed grouping strategy: Each group is set to contain k similar objects, and the total number of groups = ⌈total number of similar objects / k⌉. For example, when k=100, 1000 similar objects are divided into 10 groups.

[0092] Dynamic resource adaptation: Dynamic allocation based on the number of nodes in the computing cluster, with each computing node processing one group. If the cluster has 20 available nodes, similar objects are evenly divided into 20 groups.

[0093] Feature clustering grouping: perform secondary clustering on similar objects and automatically divide them into groups based on feature similarity. For example, 1,000 similar objects can be clustered into 15 feature clusters using the K-means algorithm.

[0094] Parallel calculation of preference information between candidate users and target objects, including but not limited to the following methods: Distributed parallel computing: Using SparkSQL's RDD parallel mechanism, different groups of computing tasks are distributed to multiple distributed nodes for synchronous execution. Each node independently completes the calculation of the preference points of the specified group and then summarizes the results.

[0095] Multi-threaded concurrent computing: Create a thread pool in a single-machine environment, with each thread processing a group of data. For example, use Java's ForkJoinPool to dynamically balance the load of each thread.

[0096] For example, the number of units of the spu group is set to . According to the number of similar SPUs ,Sure Number of groups: . Set the parallelism of the task to , according to the number of groups , divided into multiple groups: Based on the preference information between i candidate users and q groups of similar SPUs, and the feature similarity between the SPU features of the target SPU and the SPU features of the candidate SPUs, the preference scores between the candidate users and the target SPU are calculated in parallel for each of the q groups. In the embodiments of this specification, similar objects are intelligently divided into groups according to their number, and concurrent computing is used to significantly improve the efficiency and scalability of preference information calculation, which not only reduces the pressure of single-node computing, but also greatly shortens the processing time, making complex calculations on large-scale data sets efficient and feasible, enhancing the response speed of object recommendations, and reducing resource requirements.

[0097] In an optional embodiment of the present specification, determining the target user from the candidate users based on the preference information between the candidate users and the target object in step 108 includes the following specific steps: Based on the preference information between the candidate users and the target object, and a preset preference information threshold, a target user is determined from the candidate users.

[0098] The preset preference information threshold can be set according to business needs or historical data to screen out user groups with high potential interest in the target object. The preset preference information threshold flexibly adapts to business needs. For example, in an e-commerce platform, you can set the threshold of preference information such as click-through rate and weighted conversion rate to screen out target users who are likely to buy a certain product.

[0099] For example, the user recall threshold is set according to the number of anti-funneling people actually needed. Based on the preference scores of i candidate users for the target SPU: and user recall threshold , determine the anti-funneling user group of target users for each target SPU from i candidate users.

[0100] In the embodiments of this specification, based on the preference information between the candidate users and the target object and the preset preference information threshold, the target user is determined from the candidate users, which realizes the accurate screening of potential high-interest users, and not only improves the accuracy of object recommendation. At the same time, the operator can flexibly adjust the preference information threshold according to different business needs to achieve more flexible and accurate user group positioning.

[0101] In an optional embodiment of the present specification, the target users include core users, high potential users and generalized users, and the preset preference information thresholds include a preset first preference information threshold, a second preference information threshold and a third preference information threshold, and the first preference information threshold, the second preference information threshold and the third preference information threshold decrease in sequence; Based on the preference information between the candidate users and the target object, and a preset preference information threshold, determining the target user from the candidate users includes the following specific steps: When the preference information between the candidate user and the target object exceeds a first preference information threshold, the candidate user is selected as a core user; When the preference information between the candidate user and the target object exceeds a second preference information threshold, screening the candidate user as a high-potential user; When the preference information between the candidate user and the target object exceeds the third preference information threshold, the candidate user is screened out as a generalized user.

[0102] Core users are user groups that have a clear preference for the target object and a high probability of behavioral conversion. Core users have shown high-frequency interactive behaviors (such as purchases and in-depth browsing) for the target object or similar objects, and their preference information scores are significantly higher than those of ordinary users. For example, in an e-commerce scenario, core users may be users who have recently clicked on similar products multiple times or have added them to their shopping carts.

[0103] High-potential users are users who have potential interest in the target object but have not yet fully converted. High-potential users have medium preference information and may be converted into actual users due to scenario triggers (such as promotions) or content matching (such as similar object recommendations). For example, on social media platforms, high-potential users may have browsed similar content but not interacted, or have matching interest tags (such as "technology enthusiasts").

[0104] Generalized users are potential user groups that have no obvious direct preference for the target object, but may be interested in it through generalized feature associations. Generalized users have low preference information, but potential needs can be mined through cross-dimensional features (such as brand preference and category association). For example, the target generalized user of a certain sports shoe may have never browsed the product, but may be recommended because of the history of purchasing clothing of the same brand.

[0105] The first preference information threshold is used to screen core users, and can be dynamically adjusted based on historical conversion behavior data. For example, setting the preference score ≥ 0.9 (cosine similarity normalization value) covers 5% of the user group.

[0106] The second preference information threshold is used to screen high-potential users. It can be dynamically adjusted based on historical conversion behavior data to balance accuracy and coverage. For example, if the preference score is set to ≥ 0.7, the coverage of the user group will be expanded to 20%.

[0107] The third preference information threshold is used to screen generalized users, which can expand the user coverage and is usually set based on the long-tail effect. For example, if the preference score is ≥ 0.5, the coverage of the user group is expanded to 50%.

[0108] Optionally, the target object is recommended to the target user. One optional way is to adopt a unified recommendation strategy to recommend the target object to the target user. Another optional way is to adopt a differentiated recommendation strategy to recommend the target object to the target user. There is no limitation here.

[0109] For example, the first preference score threshold is set to ≥0.9, the second preference score threshold is set to ≥0.7, and the third preference score threshold is set to ≥0.5 according to the number of anti-funneling groups (core users, high-potential users, and generalized users) actually required. Based on the preference scores of i candidate users for the target SPU: and three preference score thresholds, the core users, high-potential users and generalized users of each target SPU are screened out from i candidate users in turn.

[0110] In the embodiments of this specification, users are segmented into core, high-potential and generalized groups through stratified thresholds, thereby achieving precise resource allocation, potential demand mining and anti-funnel expansion, and significantly improving the breadth of user reach while ensuring the accuracy of recommendations.

[0111] In an optional embodiment of the present specification, recommending a target object to a target user includes the following specific steps: Execute personalized recommendation strategies to recommend target objects to core users; Implement incentive recommendation strategies to recommend target objects to high-potential users; Execute the coverage recommendation strategy to recommend target objects to generalized users.

[0112] The personalized recommendation strategy is a precise reach strategy customized for the high-interest characteristics of core users, and achieves precise delivery by deeply matching the user's historical behavior and preference characteristics. The personalized recommendation strategy focuses on the clear interests that users have shown, and uses user portraits, real-time behavior data and scenario characteristics to generate customized recommendation content. For example, in an e-commerce scenario, an "exclusive discount page" is pushed to core users to display the matching combinations of target products that they frequently browse. For another example, in a community application, in-site messages are used to push in-depth evaluation content of the target object, and an entrance to the target product "You may be interested in" is attached. Another example is a pop-up reminder triggered by real-time behavior.

[0113] The incentive recommendation strategy is a reward-driven recommendation strategy for high-potential users, which encourages users to complete interest conversion through short-term incentives. The incentive recommendation strategy combines the user's potential interest and conversion cost to design incentive rules, balancing the recommendation effect and resource investment. For example, issuing limited-time coupons to high-potential users. Another example is pushing interactive tasks such as "unlock exclusive content by completing sharing" in community applications. Another example is step-by-step rewards (such as continuous clicks to unlock benefits) and social fission incentives (inviting friends to earn points).

[0114] The coverage recommendation strategy is a broad-spectrum exposure strategy for generalized users, which expands the target object's awareness through low-interference, high-coverage content implantation. The coverage recommendation strategy focuses on brand exposure and long-tail demand mining, and uses cross-platform traffic and pan-interest tags to awaken potential users. For example, inserting advertisements in the information flow. Another example is to display the use scenario video of the target SPU by placing splash screen ads on the platform. Another example is the implantation of interest circle content (such as evaluation videos) and pan-category related exposure (such as "users may also need...").

[0115] For example, when a community content application promotes outdoor sports SPU: For core users: Push SPU combinations (such as tents + sleeping bags) that match their historical purchase records through the "Exclusive Equipment Recommendations" column on the user's homepage, and mark them as "Preferred based on your hiking records."

[0116] Targeting high-potential users: When users browse camping guides, a limited-time discount window "Get 200 off when you purchase 3 or more items" pops up in the sidebar, recommending SPUs associated with the current content.

[0117] For general users: insert a lightweight graphic advertisement "Top 10 Popular Camping Equipment This Summer" into the information flow, and associate it with the topic tag "Weekend Travel" to expand exposure.

[0118] In the embodiments of this specification, while ensuring the accuracy of recommendations, the resource allocation efficiency is optimized to achieve a step-by-step expansion of the anti-funneling user group.

[0119] In an optional embodiment of the present specification, before recommending the target object to the target user in step 108, at least one of the following items is also included: Based on the object recommendation setting of the target user, the target user is screened to obtain the screened target user; Based on the preference information between the target users and the target objects, the target users are sorted to obtain sorted target users.

[0120] Based on the object recommendation settings of the target users, the target users are screened to obtain the screened target users, including but not limited to at least one of the following: User attribute screening: Set user basic attribute conditions according to the promotion needs of the target object, including but not limited to age range, gender, geographic location, device type, etc. For example, in the maternal and infant industry scenario, screen the user group aged 18-40 and female; Behavioral feature screening: Dynamic screening rules are set based on historical user behavior data, including recent click-through rate thresholds, purchase frequency ranges, content interaction depth indicators, etc. For example, in SPU advertising, screen users who have searched for similar products in the past 7 days and whose single browsing time is ≥30 seconds; Interest tag screening: Filter preset interest classification tags through the user portrait system to ensure the relevance of the subject between the target user and the target object. For example, when promoting outdoor equipment SPU, filter users with the tags of "hiking" or "camping enthusiasts"; Time window filtering: Filter the time range of user behavior data to focus on recently active users. For example, in community content recommendations, only users who have logged in ≥ 5 times in the past 30 days and have collected content are retained; Rule screening: Set up a blacklist mechanism to screen out users who are not suitable for push, including users who have purchased, complained, and low-value users. For example, when promoting luxury goods SPU, automatically exclude users who have a return record in the past three months.

[0121] Based on the object recommendation settings of the target users, the target users are screened to obtain the screened target users, including but not limited to at least one of the following: Single-dimension sorting: Sort in descending order based on preference information values, giving priority to recommending users with high preference scores. For example, in an e-commerce platform, sort users from high to low based on weighted conversion rates to ensure that high-potential users are reached first; Multi-dimensional weighted ranking: Combine preference information with other business indicators to build a comprehensive ranking weight. For example, social media platforms use the formula: comprehensive score = preference score × 0.6 + social influence index × 0.3 + content interaction rate × 0.1; Dynamic weight sorting: Automatically optimize sorting strategies based on real-time feedback data. For example, when the click-through conversion rate of a certain category's SPU is detected to be decreasing, the weight of the "deep browsing time" indicator will be automatically increased; Sorting by groups and layers: Implement differentiated sorting strategies based on user value levels. For example, users can be divided into three levels: high net worth users, ordinary users, and new users, and preference ranking can be performed independently within each level.

[0122] In the embodiments of this specification, the dual optimization mechanism of refined screening and intelligent sorting is used to achieve accurate positioning of target user groups and efficient resource allocation. This strategy not only improves the response accuracy and resource utilization of the recommendation system, but also effectively adapts to the personalized needs of different industry scenarios.

[0123] With reference to at least one of the above embodiments, Figure 3 FIG. 1 shows a flow chart of an object recommendation method provided by an embodiment of the present specification, such as Figure 3 As shown: Step 1: Generate a list of target objects.

[0124] Step 2: Calculate the feature similarity between the target object and the reference object in three dimensions (object / brand / category).

[0125] Step 3: Mixed recall of similar objects of the target object in three dimensions.

[0126] Step 4: Calculate the candidate user's preference information for similar objects: Calculate the search behavior preference scores of candidate users for similar objects; Calculate the candidate user's interaction behavior preference score for similar objects; Calculate the candidate user's in-depth browsing behavior preference score for similar objects.

[0127] Step 5: Target object collections are elastically grouped and distributed in parallel.

[0128] Step 6: Calculate the candidate user's preference information for the target object.

[0129] Calculation task 1 of the target user anti-funneling user group of the target object; Calculation task 2 of the target user anti-funneling user group of the target object; … The target user anti-funneling user group calculation task P of the target object.

[0130] Step 7: Integrate and generate the target user reverse funnel user group of the target object.

[0131] Step 8: After screening and sorting, apply the target user reverse funnel user group to the one-stop advertising intelligent delivery platform and the commercial data asset platform serving brand customers.

[0132] The following combination Figure 4 Taking the application of the object recommendation method provided in this specification in mining the marketing population of the target SPU as an example, the above object recommendation method is further explained. Figure 4 A flowchart of a process of an object recommendation method for mining a marketing population of a target SPU provided by an embodiment of the present specification is shown, and includes the following specific steps: Step 402: Acquire a user set, SPU features of a target maternal and infant product, and SPU features of candidate maternal and infant products.

[0133] For example, collect a set of all registered users and extract the SPU features of a specific stroller (such as brand name, model, applicable age, etc.) as well as the SPU features of other strollers on the platform.

[0134] Step 404: Based on the user recommendation conditions of the target maternal and infant product, candidate users are determined from the user set.

[0135] For example, you can set the recommendation criteria to users who are interested in the "baby travel" category or users who have searched for related keywords in the past month.

[0136] Step 406: Determine similar products from the candidate maternal and infant products based on the SPU features of the target maternal and infant products and the SPU features of the candidate maternal and infant products and the feature similarities between product dimensions; determine similar products from the candidate maternal and infant products based on the feature similarities between brand dimensions; determine similar products from the candidate maternal and infant products based on the feature similarities between category dimensions.

[0137] For example, recall similar products by comparing features such as the brand, model, and applicable age group of strollers.

[0138] Step 408: sort all similar maternal and infant products in descending order according to feature similarity and perform mixed recall to obtain a set of similar maternal and infant products.

[0139] For example, sort the products from the highest to the lowest similarity to the target stroller to form a list containing the most relevant stroller products.

[0140] Step 410: Based on the user behavior data of the candidate user for similar maternal and infant products, calculate the candidate user's behavior preference score for at least one type of similar maternal and infant products, weight the at least one type of behavior preference score, and obtain the preference score between the candidate user and the similar maternal and infant products.

[0141] For example, analyze users' browsing time, click times, collection behavior, etc. for different baby strollers to evaluate their preference scores.

[0142] Step 412: Determine the number of groups according to the number of similar maternal and infant products, and divide the products into multiple groups according to the number of groups.

[0143] For example, if there are 500 similar baby strollers, you can divide them into 10 groups of 50 products each.

[0144] Step 414: Based on the preference scores between the candidate user and multiple groups of similar maternal and infant products, and the feature similarity between the SPU features of the target maternal and infant product and the SPU features of the candidate maternal and infant product, the preference scores between the candidate user and the target maternal and infant product are calculated in parallel by group.

[0145] For example, data from each group can be processed in parallel through a distributed computing platform to improve efficiency.

[0146] Step 416: Determine the target user from the candidate users based on the preference scores between the candidate users and the target maternal and infant product, and a preset preference score threshold.

[0147] For example, set a certain click-through rate or conversion rate threshold to screen out the user group with the greatest potential to purchase the target stroller.

[0148] Step 418: Based on the maternal and infant product recommendation settings of the target users, the target users are screened to obtain the screened target users, and based on the preference scores between the target users and the target maternal and infant products, the target users are sorted to obtain the sorted target users.

[0149] For example, exclude users who have already purchased similar products and sort the remaining users according to the strength of their preference.

[0150] Step 420: Recommend target maternal and infant products to the target user in order according to the ranking results of the target user.

[0151] For example, information about target baby strollers can be pushed to the top ranked users in turn to ensure that the users with the greatest potential interest are reached first.

[0152] In the embodiment of this specification, a series of steps are used to realize the application of an SPU-based anti-funnel crowd mining technology in the maternal and infant industry, which significantly improves the accuracy and efficiency of target population acquisition and positioning as a whole. Using a multi-dimensional (product, brand, category) feature similarity calculation method, similar products are identified and recalled from candidate maternal and infant products, and the most relevant maternal and infant product list is formed by descending sorting, which enhances the accuracy and diversity of similarity matching. Based on the user behavior data of candidate users for similar maternal and infant products, different types of behavioral preference scores are calculated and weighted, and the user's interest in similar products and potential purchase intention are quantified, thereby further mining the user's potential needs. Using a distributed elastic parallel computing method, similar maternal and infant products are divided into multiple groups for parallel processing, which greatly improves the computing efficiency under large-scale data volume and ensures that the task execution performance meets the requirements of online applications. Based on the preference score between the candidate user and the target maternal and infant product and the preset preference score threshold, the target user group is finally determined, and the fine screening and priority sorting of the target user are achieved through the user's personalized settings and preference score sorting. It not only solves the problems of black-box nature and high cost of understanding of traditional audience expansion methods, but also supports flexible adjustment of strategies according to the characteristics of different industries, promoting the growth of advertisers' audience assets within the platform. It is especially suitable for areas such as the maternal and infant industry that require highly customized marketing plans. It ultimately achieves accurate and efficient identification and recommendation of target audiences, greatly improving user experience and satisfaction.

[0153] Corresponding to the above method embodiment, this specification also provides an object recommendation device embodiment, Figure 5 FIG. 1 is a schematic diagram showing the structure of an object recommendation device provided by an embodiment of the present specification. Figure 5 As shown, the device comprises: An acquisition module 502 is configured to acquire object features of a target object and object features of a reference object; A similarity determination module 504 is configured to determine similar objects from the reference objects based on feature similarities between object features of the target object and object features of the reference object; A user matching module 506 is configured to determine preference information between the candidate user and the target object based on preference information between the candidate user and the similar objects and feature similarity between object features of the target object and object features of the similar objects; The recommendation module 508 is configured to determine a target user from the candidate users based on preference information between the candidate users and the target object, and recommend the target object to the target user.

[0154] Optionally, the similarity determination module 504 is further configured to: Based on the feature similarity between the object feature of the target object and the object feature of the reference object, and a preset feature similarity threshold, similar objects are determined from the reference objects.

[0155] Optionally, the similarity determination module 504 is further configured to perform at least one of the following: Determining similar objects from the reference objects based on object features of the target object and object features of the reference object and feature similarities between object dimensions; Determine similar objects from the reference objects based on the object characteristics of the target object and the object characteristics of the reference object and the characteristic similarity between the brand dimensions; Based on the object features of the target object and the object features of the reference object, and the feature similarity between the category dimensions, similar objects are determined from the reference objects.

[0156] Optionally, the device further comprises: The preference determination module is configured to determine preference information between the candidate user and the similar object based on the user behavior data of the candidate user with respect to the similar object.

[0157] Optionally, the user behavior data includes at least one type of user behavior data of search, interaction and in-depth browsing; Correspondingly, the preference determination module is further configured as follows: Based on the user behavior data of the candidate user for the similar objects, at least one type of behavior preference score of the candidate user for the similar objects is calculated; at least one type of behavior preference score is weighted to obtain preference information between the candidate user and the similar objects.

[0158] Optionally, the device further comprises: The reference determination module is configured to obtain a user set; and determine a candidate user from the user set based on a user recommendation condition of the target object.

[0159] Optionally, the device further comprises: The grouping module is configured to determine the number of groups according to the number of similar objects, and divide the objects into multiple groups according to the number of groups; Correspondingly, the user matching module 506 is further configured to: Based on the preference information between the candidate user and the similar objects of the plurality of groups and the feature similarity between the object feature of the target object and the object feature of the reference object, the preference information between the candidate user and the target object is calculated in parallel by group.

[0160] Optionally, the recommendation module 508 is further configured to: Based on the preference information between the candidate users and the target object, and a preset preference information threshold, a target user is determined from the candidate users.

[0161] Optionally, the target users include core users, high-potential users and generalized users, the preset preference information thresholds include a preset first preference information threshold, a second preference information threshold and a third preference information threshold, the first preference information threshold, the second preference information threshold and the third preference information threshold decrease in sequence; the recommendation module 508 is further configured to: When the preference information between the candidate user and the target object exceeds the first preference information threshold, the candidate user is screened out as a core user; when the preference information between the candidate user and the target object exceeds the second preference information threshold, the candidate user is screened out as a high-potential user; when the preference information between the candidate user and the target object exceeds the third preference information threshold, the candidate user is screened out as a generalized user.

[0162] Optionally, the recommendation module 508 is further configured to: Implement personalized recommendation strategies to recommend target objects to core users; implement incentive recommendation strategies to recommend target objects to high-potential users; implement coverage recommendation strategies to recommend target objects to generalized users.

[0163] In the embodiments of the present specification, similar objects are used as a bridge, and the preference information between the candidate users and the target object is determined by reference and evaluation. Based on the preference information between the candidate users and the target object, the target user is determined from the candidate users, and the target object is recommended to the target user, thereby discovering the target users who have potential preferences for the target object. The potential user recommendation of the target object completes the user circle of the target object, and realizes accurate and flexible object recommendation that can fully tap potential users.

[0164] The above is a schematic scheme of an object recommendation device of this embodiment. It should be noted that the technical scheme of the object recommendation device and the technical scheme of the object recommendation method described above belong to the same concept, and the details not described in detail in the technical scheme of the object recommendation device can be referred to the description of the technical scheme of the object recommendation method described above.

[0165] Figure 6 The block diagram of a computing device provided by an embodiment of the present specification is shown. The components of the computing device 600 include but are not limited to a memory 610 and a processor 620. The processor 620 is connected to the memory 610 via a bus 630, and the database 650 is used to store data.

[0166] The computing device 600 also includes an access device 640 that enables the computing device 600 to communicate via one or more networks 660. Examples of these networks include a Public Switched Telephone Network (PSTN), a Local Area Network (LAN), a Wide Area Network (WAN), a Personal Area Network (PAN), or a combination of communication networks such as the Internet. The access device 640 may include one or more of any type of network interface (e.g., a Network Interface Controller (NIC)) of wired or wireless type, such as an IEEE 802.11 Wireless Local Area Network (WLAN) wireless interface, a Worldwide Interoperability for Microwave Access (Wi-MAX) interface, an Ethernet interface, a Universal Serial Bus (USB) interface, a cellular network interface, a Bluetooth interface, and a Near Field Communication (NFC).

[0167] In one embodiment of the present specification, the above components of the computing device 600 and Figure 6 Other components not shown in the figure may also be connected to each other, for example, via a bus. It should be understood that Figure 6 The computing device structure block diagram shown is only for the purpose of illustration, and is not intended to limit the scope of this specification. Those skilled in the art can add or replace other components as needed.

[0168] The computing device 600 may be any type of stationary or mobile computing device, including a mobile computer or mobile computing device (e.g., a tablet computer, a personal digital assistant, a laptop computer, a notebook computer, a netbook, etc.), a mobile phone (e.g., a smart phone), a wearable computing device (e.g., a smart watch, smart glasses, etc.), or other types of mobile devices, or a stationary computing device such as a desktop computer or a personal computer (PC). The computing device 600 may also be a mobile or stationary server.

[0169] The processor 620 is used to execute the following computer program / instructions, which implement the steps of the above-mentioned object recommendation method when executed by the processor.

[0170] The above is a schematic scheme of a computing device of this embodiment. It should be noted that the technical scheme of the computing device and the technical scheme of the object recommendation method described above are of the same concept, and the details not described in detail in the technical scheme of the computing device can be found in the description of the technical scheme of the object recommendation method described above.

[0171] An embodiment of the present specification further provides a computer-readable storage medium storing a computer program / instruction, which implements the steps of the above-mentioned object recommendation method when executed by a processor.

[0172] The above is a schematic scheme of a computer-readable storage medium of this embodiment. It should be noted that the technical scheme of the storage medium and the technical scheme of the object recommendation method described above are of the same concept, and the details not described in detail in the technical scheme of the storage medium can be found in the description of the technical scheme of the object recommendation method described above.

[0173] An embodiment of the present specification also provides a computer program product, including a computer program / instruction, which implements the steps of the above-mentioned object recommendation method when executed by a processor.

[0174] The above is a schematic scheme of a computer program product of this embodiment. It should be noted that the technical scheme of the computer program product and the technical scheme of the object recommendation method described above are of the same concept, and the details not described in detail in the technical scheme of the computer program product can be found in the description of the technical scheme of the object recommendation method described above.

[0175] The above is a description of a specific embodiment of the specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0176] The computer instructions include computer program codes, which may be in source code form, object code form, executable files or some intermediate forms, etc. The computer readable medium may include: any entity or device capable of carrying the computer program code, recording medium, USB flash drive, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electric carrier signal, telecommunication signal and software distribution medium, etc. It should be noted that the contents contained in the computer readable medium may be appropriately increased or decreased according to the requirements of patent practice. For example, in some regions, according to patent practice, computer readable media do not include electric carrier signals and telecommunication signals.

[0177] It should be noted that, for the above-mentioned method embodiments, for the sake of simplicity of description, they are all expressed as a series of action combinations, but those skilled in the art should be aware that the embodiments of this specification are not limited by the order of the actions described, because according to the embodiments of this specification, some steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the embodiments of this specification.

[0178] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0179] The preferred embodiments of this specification disclosed above are only used to help explain this specification. The optional embodiments do not describe all the details in detail, nor do they limit the invention to only the specific implementation methods described. Obviously, many modifications and changes can be made according to the content of the embodiments of this specification. This specification selects and specifically describes these embodiments in order to better explain the principles and practical applications of the embodiments of this specification, so that technicians in the relevant technical field can well understand and use this specification. This specification is only limited by the claims and their full scope and equivalents.

Claims

1. An object recommendation method, characterized in that: include: Obtaining object features of the target object and object features of the reference object; determining similar objects from the reference objects based on feature similarities between object features of the target object and object features of the reference objects; determining preference information between the candidate user and the target object based on preference information between the candidate user and the similar object and feature similarity between an object feature of the target object and an object feature of the similar object; Based on the preference information between the candidate users and the target object, a target user is determined from the candidate users, and the target object is recommended to the target user.

2. The method according to claim 1, characterized in that The determining a similar object from the reference object based on the feature similarity between the object feature of the target object and the object feature of the reference object comprises: Based on the feature similarity between the object feature of the target object and the object feature of the reference object, and a preset feature similarity threshold, similar objects are determined from the reference objects.

3. The method according to claim 1 or 2, characterized in that: The determining of similar objects from the reference objects based on the feature similarity between the object feature of the target object and the object feature of the reference object comprises at least one of the following: Determining similar objects from the reference objects based on the object features of the target object and the object features of the reference objects, and feature similarities between object dimensions; Determining similar objects from the reference objects based on the object characteristics of the target object and the object characteristics of the reference object and the characteristic similarity between the brand dimensions; Based on the object features of the target object and the object features of the reference object, and the feature similarity between the category dimensions, similar objects are determined from the reference objects.

4. The method according to claim 1, characterized in that: Before determining the preference information between the candidate user and the target object based on the preference information between the candidate user and the similar object and the feature similarity between the object feature of the target object and the object feature of the similar object, the method further includes: Based on the user behavior data of the candidate user with respect to the similar object, preference information between the candidate user and the similar object is determined.

5. The method according to claim 4, characterized in that The user behavior data includes at least one type of user behavior data of search, interaction and in-depth browsing; The determining, based on the user behavior data of the candidate user with respect to the similar object, preference information between the candidate user and the similar object comprises: Calculating the at least one type of behavior preference score of the candidate user for the similar object based on the user behavior data of the candidate user for the similar object; The at least one type of behavior preference score is weighted to obtain preference information between the candidate user and the similar object.

6. The method according to claim 1, characterized in that Before determining the preference information between the candidate user and the target object based on the preference information between the candidate user and the similar object and the feature similarity between the object feature of the target object and the object feature of the similar object, the method further includes: Get the user collection; Based on the user recommendation condition of the target object, a candidate user is determined from the user set.

7. The method according to claim 1, characterized in that Before determining the preference information between the candidate user and the target object based on the preference information between the candidate user and the similar object and the feature similarity between the object feature of the target object and the object feature of the similar object, the method further includes: Determining the number of groups according to the number of similar objects, and dividing the groups into multiple groups according to the number of groups; The determining the preference information between the candidate user and the target object based on the preference information between the candidate user and the similar object and the feature similarity between the object feature of the target object and the object feature of the similar object comprises: Based on the preference information between the candidate user and the multiple groups of similar objects and the feature similarity between the object feature of the target object and the object feature of the reference object, the preference information between the candidate user and the target object is calculated in parallel by group.

8. The method according to claim 1, characterized in that The step of determining a target user from the candidate users based on the preference information between the candidate users and the target object includes: Based on the preference information between the candidate users and the target object, and a preset preference information threshold, a target user is determined from the candidate users.

9. The method according to claim 8, characterized in that The target users include core users, high-potential users and generalized users, the preset preference information thresholds include a preset first preference information threshold, a second preference information threshold and a third preference information threshold, the first preference information threshold, the second preference information threshold and the third preference information threshold decreasing in sequence; The determining the target user from the candidate users based on the preference information between the candidate users and the target object and a preset preference information threshold comprises: When the preference information between the candidate user and the target object exceeds the first preference information threshold, screening the candidate user as a core user; When the preference information between the candidate user and the target object exceeds the second preference information threshold, screening the candidate user as a high-potential user; When the preference information between the candidate user and the target object exceeds the third preference information threshold, the candidate user is screened out as a generalized user.

10. The method according to claim 9, characterized in that The recommending the target object to the target user comprises: Execute a personalized recommendation strategy to recommend the target object to the core user; Execute an incentive recommendation strategy to recommend the target object to the high-potential user; The coverage recommendation strategy is executed to recommend the target object to the generalized user.

11. An object recommendation device, characterized in that: include: An acquisition module, configured to acquire object features of a target object and object features of a reference object; a similarity determination module configured to determine similar objects from the reference objects based on feature similarities between object features of the target object and object features of the reference objects; A user matching module, configured to determine the preference information between the candidate user and the target object based on the preference information between the candidate user and the similar object and the feature similarity between the object feature of the target object and the object feature of the similar object; The recommendation module is configured to determine a target user from the candidate users based on preference information between the candidate users and the target object, and recommend the target object to the target user.

12. A computing device, characterized in that: include: Memory and processor; The memory is used to store computer programs / instructions, and the processor is used to execute the computer programs / instructions. When the computer programs / instructions are executed by the processor, the steps of the method described in any one of claims 1 to 10 are implemented.

13. A computer-readable storage medium, characterized in that: It stores a computer program / instruction, which implements the steps of the method described in any one of claims 1 to 10 when executed by a processor.

14. A computer program product, characterized in that The invention comprises a computer program / instruction, which, when executed by a processor, implements the steps of the method according to any one of claims 1 to 10.

Images