Personalized commodity recommendation method and device, equipment and medium

Abstract

The invention relates to a personalized commodity recommendation method, a personalized commodity recommendation device, personalized commodity recommendation equipment and a personalized commodity recommendation medium in the technical field of e-commerce. Determining user preference score distribution and re-purchase trend score distribution of a plurality of target categories accessed by the user in the independent stations; based on historical purchase orders and historical conversion paths of all users in the independent station, correspondingly determining co-purchase relation score distribution and conversion relation score distribution of the multiple target categories; the user preference score distribution, the re-purchase trend score distribution, the co-purchase relationship score distribution and the conversion relationship score distribution are weighted and fused into category interest score distribution; and determining recommended commodities pushed to the user based on the category interest score distribution. According to the invention, accurate personalized commodity recommendation is realized, and the user satisfaction is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of e-commerce, and in particular to a personalized commodity recommendation method and a corresponding device, computer equipment and computer readable storage medium. BACKGROUND

[0002] With the rapid development of e-commerce platforms, the number and categories of goods on the platform have increased dramatically, and the information overload problem faced by users is becoming increasingly prominent. How to accurately mine and recommend the commodity categories that users are really interested in from the vast amount of commodities has become a key link to improve user experience and platform conversion rate.

[0003] At present, the common category recommendation method mainly relies on the historical behavior data of individual users, such as click records and purchase records, and models the user preferences through such data to generate recommendation results. However, this method only focuses on the behaviors that have occurred to the user himself, and fails to fully consider the user's potential interest orientation and the internal relevance between different categories, such as the co-purchase rule between categories or the conversion logic implied in the user behavior path. Due to the lack of comprehensive analysis of the multi-dimensional behavior patterns of users and the complex relationships between categories, the existing method often leads to limited personalization and narrow coverage of the recommendation results, making it difficult to adapt to the dynamically changing interest needs of users, thereby limiting the further improvement of the recommendation effect.

[0004] Therefore, there is an urgent need in the art for a personalized recommendation method that can comprehensively analyze multi-dimensional behavior characteristics of users and deeply integrate the association relationships between categories to overcome the above technical limitations. SUMMARY

[0005] The primary purpose of the present application is to solve at least one of the above problems and provide a personalized commodity recommendation method and a corresponding device, computer equipment and computer readable storage medium.

[0006] To meet the various purposes of the present application, the present application adopts the following technical solutions: A personalized commodity recommendation method provided to adapt to one of the purposes of the present application, comprising the following steps: In response to a commodity recommendation event of an independent station user, based on the historical behavior data of the user, determining a user preference score distribution and a repurchase trend score distribution of a plurality of target categories visited by the user in the independent station; Based on the historical purchase orders and historical conversion paths of all users in the independent station, corresponding co-purchase relationship score distribution and conversion relationship score distribution of the plurality of target categories are determined; The user preference score distribution, the repurchase trend score distribution, the co-purchase relationship score distribution and the conversion relationship score distribution are weighted and integrated into a category interest score distribution; determine recommended goods pushed to the user based on the category interest score distribution.

[0007] In another aspect, a personalized goods recommendation device is provided to adapt to one of the purposes of the present application, comprising an event response module, a distribution determination module, a distribution fusion module, and a user recommendation module. The event response module is configured to respond to a goods recommendation event of an independent station user, and determine a user preference score distribution and a repeat purchase trend score distribution of a plurality of target categories visited by the user in the independent station based on historical behavior data of the user. The distribution determination module is configured to determine a co-purchase relationship score distribution and a conversion relationship score distribution of the plurality of target categories based on historical purchase orders and historical conversion paths of all users in the independent station. The distribution fusion module is configured to weight and fuse the user preference score distribution, the repeat purchase trend score distribution, the co-purchase relationship score distribution, and the conversion relationship score distribution into a category interest score distribution. The user recommendation module is configured to determine recommended goods pushed to the user based on the category interest score distribution.

[0008] In yet another aspect, a computer device is provided to adapt to one of the purposes of the present application, comprising a central processing unit and a memory. The central processing unit is configured to invoke a computer program stored in the memory to execute the steps of the personalized goods recommendation method described in the present application.

[0009] In yet another aspect, a computer program product is provided to adapt to another purpose of the present application, comprising a computer program / instruction. When executed by a processor, the computer program / instruction implements the steps of the method described in any one of the embodiments of the present application.

[0010] The technical solution of the present application has multiple advantages, including but not limited to the following aspects: Firstly, the present application responds to a goods recommendation event of an independent station user, and determines a user preference score distribution and a repeat purchase trend score distribution of a plurality of target categories visited by the user in the independent station based on historical behavior data of the user. Therefore, the personal interests and behavior patterns of the user can be more comprehensively and dynamically captured. The user preference score distribution comprehensively reflects the historical interaction preferences of the user for different categories, and the repeat purchase trend score distribution quantifies the repeated purchase tendency of the user for a specific category. This makes the recommendation system not only rely on single click or purchase records, but also more accurately model user interests through multi-dimensional behavior feature analysis, avoids the recommendation rigidity problem caused by only focusing on occurred behaviors, and significantly improves the personalization degree and adaptability of the recommendation.

[0011] Secondly, based on the historical purchase orders and historical conversion paths of all users in the independent station, the co-purchase relationship score distribution and the conversion relationship score distribution of the corresponding multiple target categories are determined, so as to deeply mine the internal correlation between categories and the user behavior rules. The co-purchase relationship score distribution uses group co-purchase data to reveal the co-purchase purchase mode between categories, for example, which categories are frequently purchased by the same user, and the conversion relationship score distribution extracts the conversion probability from browsing to purchase from the historical conversion path, reflecting the potential conversion logic between categories. This analysis based on global user behavior makes up for the shortcomings of relying only on individual historical data, can discover potential interests that users have not expressed, broaden the coverage of recommendations, avoid too narrow or repetitive recommendation results, and thus enhance the diversity and exploration of recommendations.

[0012] In addition, the user preference score distribution, the repeat purchase trend score distribution, the co-purchase relationship score distribution and the conversion relationship score distribution are weighted and fused to form a unified category interest score distribution. This weighted fusion mechanism organically integrates user personal behavior characteristics and group behavior patterns, balances the contribution of different scores, emphasizes user preferences, and integrates category associations and conversion trends, so that the final interest score distribution is more comprehensive and accurate. Based on this distribution, recommended goods are pushed to the user, which not only ensures that the recommendation results fit the user's historical interests, but also introduces the synergistic effect of related categories to promote users to discover new interest points, thereby effectively improving the relevance of recommendations and user experience, while improving the platform conversion rate, overcoming the dynamic interest adaptation problem mentioned in the background art. BRIEF DESCRIPTION OF DRAWINGS

[0013] The above and / or additional aspects and advantages of the present application will become apparent and more readily appreciated from the following description, taken in conjunction with the accompanying drawings, in which: Figure 1 The network architecture of an exemplary e-commerce platform of the present application; Figure 2 The flowchart of a typical embodiment of the personalized commodity recommendation method of the present application; Figure 3 The principle block diagram of the personalized commodity recommendation device of the present application; Figure 4 The structure diagram of a computer device used in the present application. DETAILED DESCRIPTION

[0014] Embodiment 1 of the present application will be described in detail below, examples of which are shown in the accompanying drawings, in which the same or similar reference numerals represent the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by reference to the drawings are exemplary and are only used to explain the present application, and cannot be interpreted as a limitation on the present application.

[0015] It is to be understood that the singular forms "a," "an," and "the" include plural referents unless the context clearly dictates otherwise. It is further understood that the terms "comprise" (and any grammatical variations thereof, such as "comprises" and "comprising"), when used in this specification, specify the presence of stated features, integers, steps, operations, elements, and / or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or groups thereof. It is further understood that when an element is referred to as being "connected" or "coupled" to another element, it can be directly connected or coupled to the other element or intervening elements can be present. Furthermore, "connected" or "coupled" as used herein can include wirelessly connected or wirelessly coupled. As used herein, the term "and / or" includes any and all combinations of one or more of the associated listed items.

[0016] Those skilled in the art will appreciate that unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. It will be further understood that terms, such as those defined in commonly used dictionaries, should be interpreted as having a meaning that is consistent with their meaning in the context of the relevant art and will not be interpreted in an idealized or overly formal sense unless expressly so defined herein.

[0017] As Figure 1 In the network architecture as shown, the e-commerce platform 82 is deployed in the Internet to provide corresponding services to its users, and the devices 80 of the merchant users and the devices 81 of the consumer users of the e-commerce platform 82 are also connected to the Internet to use the services provided by the e-commerce platform.

[0018] An exemplary e-commerce platform 82 provides a supply and demand matching of products and / or services to the public through the Internet infrastructure, in which the products and / or services are provided as commodity information. For the sake of simplicity of description, the concepts of commodity, product, etc. are used in the present application to refer to the products and / or services in the e-commerce platform 82, which can be physical products, digital products, tickets, service subscriptions, other offline services, etc.

[0019] Real-world entities can access the e-commerce platform 82 in the identity of users to use various online services provided by the e-commerce platform 82 to achieve the purpose of participating in the business activities implemented by the e-commerce platform 82. These entities can be natural persons, legal persons, or social organizations, etc. Corresponding to the two types of entities of merchants and consumers in the business activities, the e-commerce platform 82 correspondingly exists two types of users of merchant users and consumer users. The entities of the product circulation chain in the business activities, including manufacturers, sellers, retailers, logistics providers, etc., can use online services in the e-commerce platform 82 in the identity of merchant users, while the consumers in the business activities, including real or potential consumers, can use online services in the e-commerce platform 82 in the identity of their corresponding consumer users. In actual business activities, the same entity can act as a merchant user and a consumer user, and flexible and variable understanding should be made accordingly.

[0020] The infrastructure for deploying the e-commerce platform 82 mainly includes a backend architecture and front-end devices. The backend architecture runs various online services through a service cluster, including middleware or front-end services for platform parties, services for consumers, services for merchants, etc., to enrich and perfect its service functions. The front-end devices mainly cover terminal devices used by users as clients to access the e-commerce platform 82, including but not limited to various mobile terminals, personal computers, point-of-sale devices, etc. For example, a merchant user can use his terminal device 80 to input product information for his online store, or use the interface provided by the e-commerce platform to generate product information; a consumer user can access the web page of the online store implemented by the e-commerce platform 82 through his terminal device 81, trigger the shopping process through the shopping button provided on the web page, and call various online services provided by the e-commerce platform 82 in the shopping process, so as to achieve the purpose of shopping and ordering.

[0021] In some embodiments, the e-commerce platform 82 can be implemented by a processing facility including a processor and a memory, which stores a set of instructions that, when executed, cause the e-commerce platform 82 to perform the electronic commerce and support functions involved in this application. The processing facility can be part of a server, a client, a network infrastructure, a mobile computing platform, a cloud computing platform, a fixed computing platform, or other computing platforms, and provides electronic components of the e-commerce platform 82, merchant devices, payment gateways, application developers, marketing channels, transportation providers, customer devices, point-of-sale devices, etc.

[0022] The e-commerce platform 82 can be implemented as a cloud computing service, a software as a service (SaaS), an infrastructure as a service (IaaS), a platform as a service (PaaS), a desktop as a service (DaaS), a hosted software as a service, a mobile backend as a service (MBaaS), an information technology management as a service (ITMaaS), and the like online services. In some embodiments, various functional components of the e-commerce platform 82 can be implemented to be suitable for operation on various platforms and operating systems, for example, an administrator user of an online store enjoys the same or similar functions regardless of various embodiments such as iOS, Android, HomonyOS, or web pages.

[0023] The e-commerce platform 82 can implement respective independent stations for various merchants to run their respective online stores, and provide respective business management engine instances for the merchants to establish, maintain, and run one or more online stores in one or more independent stations. The business management engine instance can be used for content management, task automation, and data management of one or more online stores, and can configure various specific business processes of the online store through an interface or a built-in component to support the implementation of business activities. The independent station is the infrastructure of the e-commerce platform 82 with cross-border service functions, and the merchant can maintain its online store based on the independent station in a relatively centralized and autonomous manner. The independent station usually has a domain name and storage space dedicated to the merchant, and different independent stations have relative independence. The e-commerce platform 82 can provide standardized or personalized technical support for a large number of independent stations, so that the merchant user can customize a business management engine instance suitable for itself, and use this business management engine instance to maintain one or more online stores owned by the merchant.

[0024] The online store can be configured and maintained by the merchant user logging in as an administrator to its business management engine instance. With the support of various online services provided by the infrastructure of the e-commerce platform 82, the merchant user can configure various functions in its online store and review various data as an administrator, for example, the merchant user can manage various aspects of its online store, such as viewing recent activities of the online store, updating product catalog of the online store, managing orders, recent visit activities, total order activities, and the like; the merchant user can also view more detailed information about business and visitors to the online store of the merchant by obtaining reports or metrics, such as displaying sales summary of the overall business of the merchant, specific sales and engagement data of the activity sales marketing channel, and the like.

[0025] The e-commerce platform 82 can provide a communication facility and associated merchant interface for providing electronic communications and marketing, such as collecting and analyzing communication interactions between merchants, consumers, merchant devices, customer devices, point-of-sale devices, etc. using an electronic message aggregation facility, aggregating and analyzing communications, such as for increasing the potential for providing product sales, etc. For example, a consumer can have a question about a product that can generate a conversation between the consumer and a merchant (or an automated processor-based agent representing the merchant), where the communication facility is responsible for the interaction and provides the merchant with analysis on how to improve the probability of a sale.

[0026] In some embodiments, an application suitable for installation to a terminal device can be provided to serve the access needs of different users, so as to enable various users to access the e-commerce platform 82, such as a merchant back-end module of an online store in the e-commerce platform 82, etc. in the terminal device by running the application. In the process of implementing business activities through these functions, the e-commerce platform 82 can implement various functions related to the implementation of business activities as middleware or online services and open corresponding interfaces, and then implant the interface access functions into the application to realize function extension and task implementation. The business management engine can include a series of basic functions, and expose these functions to online services and / or application calls through APIs. Online services and applications use corresponding functions by remotely calling corresponding APIs.

[0027] With the support of various components of the business management engine instance, the e-commerce platform 82 can provide online shopping functions, enabling merchants to establish contact with customers in a flexible and transparent manner, and enabling consumer users to select and purchase goods online, create a product order, provide a delivery address for the goods in the product order, and complete payment confirmation of the product order. Then, the merchant can review and complete or cancel the order.

[0028] The personalized product recommendation method of the present application can be programmed as a computer program product, deployed in a client or server for running and implementation, for example, in the exemplary application scenario of the present application, it can be implemented in the server of the e-commerce customer service platform. Therefore, the interface opened after the running of the computer program product can be accessed, and the process of the computer program product can be interacted with through the graphical user interface to execute the method.

[0029] Referring to Figure 2 In a typical embodiment of the personalized product recommendation method of the present application, the following steps are included: Step S1100, in response to a product recommendation event of an independent station user, determining a user preference score distribution and a repurchase trend score distribution of a plurality of target categories visited by the user in the independent station based on historical behavior data of the user; The e-commerce platform can provide users with product recommendation services in multiple different scenarios in the independent station, such as the homepage recommendation column, the “Guess You Like” area below the product detail page, the additional recommendation column on the shopping cart page, the recommendation expansion area at the end of the search result page, and the exclusive recommendation area of the user personal center. Each recommendation scenario has its specific business goal and user intention; for example, the homepage recommendation column focuses on exploring the user's potential interest and improving exploration, and the shopping cart additional recommendation aims to make complementary or alternative recommendations based on the current selected products. When a user accesses a specific page or area of a recommendation scenario in the independent station, a product recommendation event is triggered. The event can carry the user's unique identifier and the unique identifier of the recommendation scenario. The user's unique identifier is used to uniquely refer to the user, to distinguish from other users, which can be a user ID, and can be configured by a person skilled in the art as needed.

[0030] When the server responds to the user recommendation event, it obtains the unique identifier of the recommendation scenario and the user identifier carried in the event. The identifier is used to uniquely distinguish different recommendation service scenarios, which can be a pre-defined scenario number or descriptive text, and can be configured by a person skilled in the art as needed. Then, the user's historical behavior data is extracted from the user behavior database according to the user identifier. The historical behavior data includes multiple interaction behavior information generated by the user's interaction with the products in the independent station. Each interaction behavior information includes: the target category of the user's current interaction (i.e. the product category), the interaction timestamp, and the interaction behavior type.

[0031] For constructing the user preference score distribution, the user's historical interaction behavior data is obtained, and a rolling time window mechanism is used to divide the behavior data in time sequence, ensuring that the dynamic changes of the user's interest are captured. For each target category in each time window, the adjusted preference score is calculated according to different interaction behavior types (such as purchase weight higher than browsing) and their frequencies, and a time decay algorithm (such as exponential decay) is used to calibrate the score to improve the weight of recent behavior, and finally form a user preference score distribution reflecting the user's real-time interest preference.

[0032] For constructing the repeat purchase trend score distribution, the historical purchase frequency of the user's historical behavior data corresponding to each target category is counted, and the category popularity bias is eliminated by normalization processing (such as global proportion algorithm) to generate the repeat purchase trend score distribution, to quantify the user's repeated purchase tendency for a specific category.

[0033] Step S1200, based on the historical purchase orders and historical conversion paths of all users in the independent station, the co-purchase relationship score distribution and the conversion relationship score distribution of the plurality of target categories are determined correspondingly; In one embodiment, for the co-purchase relationship score distribution, co-purchase patterns between categories (such as "beer and diapers" type of association) are mined by analyzing historical order data of global users, and co-purchase strength is quantified in the form of conditional probability. When calculating the co-purchase relationship score of each target category in the distribution, the co-purchase relationship score is obtained by summing up each co-purchase conditional probability corresponding to the target category, so that the score reflects the group co-purchase rule. The co-purchase conditional probability corresponding to the target category is specifically determined by the conditional probability of purchasing the goods of the target category on the premise of purchasing the goods of other categories, which is determined by analyzing the co-purchase performance of all users in the independent station. Those skilled in the art can flexibly modify and implement it according to the disclosure herein, or implement it according to the disclosure of the subsequent embodiments.

[0034] In one embodiment, for the conversion relationship score distribution, the conversion conditional probability between categories (such as the probability of purchasing pants after browsing shirts) is mined based on the conversion path of the user from browsing to purchasing. When calculating the conversion relationship score of each target category in the distribution, the conversion relationship score is obtained by summing up each conversion conditional probability corresponding to the target category, to predict the potential purchase intention of the group. The conversion conditional probability corresponding to the target category is specifically determined by the conditional probability of purchasing the goods of the target category on the premise of browsing the goods of other categories, which is determined by analyzing the conversion performance of all users in the independent station. Those skilled in the art can flexibly modify and implement it according to the disclosure herein, or implement it according to the disclosure of the subsequent embodiments.

[0035] Step S1300, weighting and fusing the user preference score distribution, the repeat purchase trend score distribution, the co-purchase relationship score distribution and the conversion relationship score distribution into a category interest score distribution; The user preference score distribution, the repeat purchase trend score distribution, the co-purchase relationship score distribution and the conversion relationship score distribution are weighted and fused to form a unified category interest score distribution. The fusion mechanism organically integrates user personal behavior characteristics (preference and repeat purchase) and group behavior patterns (co-purchase and conversion), and by measuring the contribution of different scores to configure the weight of the corresponding distribution, the user's own historical interest is emphasized, and the category association and conversion trend are integrated, so that the final score is more comprehensive and accurate, and can dynamically adapt to user interest drift. Those skilled in the art can configure the weight corresponding to each distribution according to the disclosure herein.

[0036] Step S1400, determining recommended goods pushed to the user based on the category interest score distribution.

[0037] According to the unique identifier of the recommended scene, the specific recommended scene corresponding to the current response recommendation event (such as the home page recommendation column, the shopping cart additional recommendation, etc.) is determined, and then the recommended sampling strategy preset by the scene (such as the exploration and utilization trade-off strategy for the home page and the deterministic screening strategy for the shopping cart) is called to sample the goods from the category interest score distribution. For example, high-score category goods can be selected by using the inverted screening strategy, and non-highest-score category goods are randomly selected by using the exploration strategy to broaden the diversity of the recommended goods. Finally, the server pushes the recommended goods determined to the terminal of the user, and the terminal displays the recommended goods in the corresponding scene interface after receiving the recommended goods, thereby realizing the context-aware accurate push.

[0038] As can be understood from the above embodiments, the present application has many advantages over the prior art, at least including: Firstly, by responding to the goods recommendation event of an independent in-site user and based on the historical behavior data of the user, the present application determines the user preference score distribution and the repeat purchase trend score distribution of a plurality of target categories visited by the user in the independent site, so as to more comprehensively and dynamically capture the personal interests and behavior patterns of the user. The user preference score distribution comprehensively reflects the historical interaction preferences of the user for different categories, and the repeat purchase trend score distribution quantifies the repeated purchase tendency of the user for a specific category, which makes the recommendation system not only rely on single click or purchase records, but also more accurately model the user interest through multi-dimensional behavior feature analysis, avoids the recommendation rigidity problem caused by only focusing on the occurred behavior, and significantly improves the individualization degree and adaptability of the recommendation.

[0039] Secondly, based on the historical purchase orders and historical conversion paths of all users in the independent site, the co-purchase relationship score distribution and the conversion relationship score distribution corresponding to a plurality of target categories are determined, so as to deeply mine the internal correlation between categories and the user behavior rules. The co-purchase relationship score distribution uses group co-purchase data to reveal the co-purchase mode between categories, for example, which categories are frequently purchased by the same user, and the conversion relationship score distribution extracts the conversion probability from browsing to purchase from the historical conversion path, reflecting the potential conversion logic between categories. This global user behavior-based analysis makes up for the shortcomings of relying only on individual historical data, can discover potential interests that have not been expressed by the user, broaden the coverage of the recommendation, avoid the recommendation results being too narrow or repetitive, and thus enhance the diversity and exploration of the recommendation.

[0040] In addition, the user preference score distribution, the repurchase trend score distribution, the co-purchase relationship score distribution and the conversion relationship score distribution are weighted and fused to form a unified category interest score distribution. This weighted fusion mechanism organically integrates user personal behavior characteristics and group behavior patterns, balances the contribution of different scores, emphasizes user preferences, and integrates category association and conversion trends, so that the final interest score distribution is more comprehensive and accurate. Based on this distribution, recommended goods are pushed to the user, which not only ensures that the recommendation result fits the user's historical interest, but also introduces the synergistic effect of related categories to promote the user to discover new interest points, thereby effectively improving the relevance of the recommendation and the user experience, and improving the platform conversion rate, overcoming the dynamic interest adaptation problem mentioned in the background technology.

[0041] In a further embodiment, step S1100, based on the historical behavior data of the user, determining a user preference score distribution of a plurality of target categories visited by the user in the independent station, comprises the following steps: Step S1110, obtaining the historical behavior data of the user, the historical behavior data comprising a plurality of interaction behavior information, the interaction behavior information comprising a target category of the user's current interaction, an interaction timestamp, an interaction behavior type; The user behavior database stores the historical behavior data of all users in the independent station, and each historical behavior data is associated with the unique identifier of the corresponding user for storage. Therefore, according to the unique identifier of the user, the historical behavior data of the user can be obtained from the historical behavior database.

[0042] It can be understood that the user's interaction with the goods in the independent station on his terminal can be monitored through the front end. Whenever an interaction occurs, the category to which the goods that interact with the user belong is taken as the target category, the current timestamp is taken as the interaction timestamp, and the specific behavior type of the current interaction is taken as the interaction behavior type. The specific behavior type can include browsing, clicking, adding to the shopping cart, purchasing, collecting, sharing, etc. The behavior type can be configured by the person skilled in the art according to the actual business needs. Subsequently, the historical behavior data of the user is queried from the user behavior database according to the unique identifier of the user, the target category of the current interaction, the interaction timestamp, and the interaction behavior type are taken as the latest interaction behavior information, and inserted into the first position of the historical behavior data. If the interaction behavior information in the historical behavior data reaches the upper limit of the information amount at this time, the oldest interaction behavior information, i.e., the last interaction behavior information, is removed to maintain the timeliness of the historical behavior data. The upper limit of the information amount can be configured by the person skilled in the art as needed.

[0043] Step S1120, dividing the historical behavior data using a rolling time window mechanism, and determining the interaction behavior information corresponding to each time window according to the interaction timestamp in the interaction behavior information; The rolling time window mechanism sequentially slides along the time axis with windows of fixed time length, covering the entire time range of user historical behavior data. Each time window has the same preset time length, for example, one day, one week, or one month, which can be configured by those skilled in the art according to actual business needs. The system divides each interaction behavior information into a specific time window according to the interaction timestamp in the interaction behavior information. For example, if the time window length is configured as one day, all interaction behavior information with interaction timestamps within the same natural day is classified into the same time window; if the time window length is configured as one week, all interaction behavior information with interaction timestamps within the same natural week is classified into the same time window. This mechanism ensures that each interaction behavior information belongs to only one time window, and there is no overlap between time windows and the entire data history period is continuously covered.

[0044] Step S1130, for each time window, determining the target category's to-be-adjusted preference score according to all interaction behavior types and their interaction frequencies of the target category in the time window; For different interaction behavior types including browsing, clicking, adding to cart, purchasing, collecting, sharing, etc., those skilled in the art can configure specific behavior types according to actual business needs. Each interaction behavior type can be assigned a preset weight value, and the size of the weight value reflects the importance of the behavior type to the user preference representation. For example, the purchasing behavior can be assigned the highest weight value (such as 1.0), the adding-to-cart behavior is assigned a higher weight value (such as 0.8), the clicking behavior is assigned a medium weight value (such as 0.5), and the browsing behavior is assigned a lower weight value (such as 0.3). The weight value is dynamically adjusted by those skilled in the art according to business goals. For a single target category in a specific time window, the interaction frequencies of each interaction behavior type corresponding to the target category are accumulated, which is equivalent to the total number of interaction behavior information containing the target category and the interaction behavior type in the time window, and then the interaction frequencies of each behavior type are multiplied by their corresponding weight values and summed to obtain the to-be-adjusted preference score of the target category in the time window. For example, a user has 2 purchases (weight 1.0), 3 additions to cart (weight 0.8), and 5 clicks (weight 0.5) for a category in a time window, and the to-be-adjusted preference score of the category is calculated as (2x1.0) + (3x0.8) + (5x0.5) = 2 + 2.4 + 2.5 = 6.9.

[0045] Step S1140, using a time decay algorithm to determine the preference adjustment coefficient of each time window, and adjusting the to-be-adjusted preference score of each target category in the corresponding time window to a calibrated preference score; The time decay algorithm assigns different decay factors to the time windows based on their distance to the current time. The closer the time window is to the current time, the smaller the decay factor, indicating that the ability of the time window to reduce the preference score to be adjusted is weaker. The farther the time window is to the current time, the larger the decay factor, indicating that the ability of the time window to reduce the preference score to be adjusted is stronger. The time decay algorithm includes any one of an exponential decay algorithm, a Gaussian decay algorithm, a linear decay algorithm, a natural logarithm decay algorithm, etc. A person skilled in the art can select one according to the business and the function characteristics of matching the business.

[0046] In an embodiment, the exponential decay algorithm is used as the time decay algorithm, and the preference adjustment coefficient is calculated using the exponential decay algorithm, and the formula is: decay factor = e^(-λ × Δt), where Δt represents the time interval of the time window from the current time (in units of time window length), and λ is the decay rate parameter, which is configured by a person skilled in the art. For example, λ can be configured as 0.1, indicating that each time window interval decays by 10% of the influence. Thus, the decay factor is used as the preference adjustment coefficient to amplify the contribution of the recent window. Accordingly, the adjusted preference score in each time window is multiplied by its corresponding preference adjustment coefficient to obtain the calibrated preference score.

[0047] For ease of understanding, by way of example, a certain adjusted preference score 3 time window intervals away from the current time is 6.9, and the preference adjustment coefficient is e^(-0.1×3)≈0.74, so the calibrated preference score is 6.9 × 0.74 = 5.106.

[0048] Step S1150, for each target category appearing in the historical behavior data, the user preference score is determined according to the calibrated preference score of the target category in the corresponding time window, and the user preference score distribution is formed.

[0049] For each target category in the historical behavior data, the calibrated preference score in at least one time window of the target category is summed to obtain the user preference score. Finally, each target category is associated with its user preference score to form the user preference score distribution.

[0050] In this embodiment, based on the multi-dimensional interaction information of target categories, interaction timestamps and behavior types in the user's detailed historical behavior data, a solid data foundation is laid for subsequent accurate quantification of user preferences. The rolling time window mechanism is adopted to divide the massive behavior data in time sequence, ensuring the dynamic capture ability of user interest changes and effectively distinguishing the differentiation and timeliness characteristics of user behavior in different periods. Further, for a specific category in each time window, the weighted aggregation of different interaction behavior types and their frequencies is used to calculate the adjusted preference score, so that the contribution difference of different behaviors to preference representation is fully considered, such as the purchase weight being greater than browsing, making the preference measurement more in line with the actual business scenario. The time decay algorithm is introduced to calibrate the above score, which significantly improves the weight of recent behavior, so that it can sensitively reflect the latest drift of user interest and enhance the adaptability of the recommendation system to the dynamic evolution of user preferences, thereby providing accurate and time-sensitive quantification of user interest for highly personalized product recommendation.

[0051] In a further embodiment, step S1100, based on the user's historical behavior data, determines the repurchase trend score distribution of a plurality of target categories visited by the user in the independent station, comprising the following steps: Step S1101, obtaining the historical behavior data of the user, the historical behavior data comprising a plurality of interaction behavior information, the interaction behavior information comprising the target category of the user's current interaction, the interaction timestamp, and the interaction behavior type; With reference to step S1110, this step is not described in detail.

[0052] Step S1102, for each different target category in the historical behavior data, determining the interaction frequency of the corresponding purchase behavior type of the target category; For different target categories appearing in the historical behavior data, the total number of interaction behavior types that are purchase in all interaction behavior information containing the target category is accumulated, which is the interaction frequency of the purchase behavior type corresponding to the target category. For example, the user has purchased five times of goods of the "smartphone" category, and the purchase interaction frequency corresponding to the "smartphone" category is 5.

[0053] Step S1103, normalizing the interaction frequency corresponding to each target category to obtain the repurchase trend score of each target category to form a repurchase trend score distribution.

[0054] Subsequently, the purchase interaction frequency of each target category is normalized to obtain a repurchase trend score. The purpose of normalization is to eliminate the bias between different categories due to the difference in absolute purchase frequency, so that the score can more fairly reflect the user's repurchase tendency for each category. The normalization operation can be realized by using any one of the normalization algorithms such as softmax function, z-score standardization, linear scaling normalization (Min-Max Scaling), truncation, binning, logarithmic transformation, etc.

[0055] In an embodiment, a global proportion algorithm is used as the normalization algorithm. First, the interaction frequencies of all target categories appearing in the historical behavior data are summed to obtain the total interaction frequency. Then, for each target category, the interaction frequency of the target category is divided by the total interaction frequency to complete the normalization of the interaction frequency. The quotient obtained in this way is used as the repurchase trend score of the target category. Finally, each target category is associated with its repurchase trend score to form a repurchase trend distribution.

[0056] In this embodiment, the focus is on the repeated purchase behavior of users for specific categories. By counting the historical purchase frequency of each target category, a core indicator that directly reflects user loyalty and consumption habits is obtained. Subsequently, the absolute frequency is normalized to effectively eliminate the bias caused by the difference in category popularity or the difference in overall user activity, so that the repurchase tendency between different categories is comparable. The final repurchase trend score distribution can clearly reveal which categories the user shows a stable repeated consumption willingness, providing key data support for the recommendation system in maintaining the user's existing interest and promoting loyalty improvement, and helping to timely integrate the user's frequently purchased categories in the recommendation list, improving user satisfaction and shopping convenience.

[0057] In a further embodiment, step S1200, based on the historical purchase orders of all users in the independent station, determines the co-purchase relationship score distribution of the plurality of target categories, including the following steps: Step S1201, based on the historical purchase orders of all users in the independent station, determines a corresponding co-purchase relationship set, the co-purchase relationship set contains a plurality of co-purchase relationship items and their co-purchase condition probability, the co-purchase relationship item includes a first category and a second category, and the co-purchase condition probability is the probability of purchasing the first category as a premise while purchasing the second category; It can be understood that the server of the independent station generally needs to store the order formed by each user in the independent station to purchase the goods. At this time, all orders are obtained as historical purchase orders, and all different categories to which the goods purchased by the user belong in the historical purchase orders are determined. For each two different categories in the categories, it is determined whether there are historical purchase orders corresponding to the purchase of the goods of the two categories at the same time or in a short time in succession. When there are, the total number of different users is determined from the users who output these historical purchase orders as the co-purchase user number of the two categories. In addition, the two categories are divided into a former (for example: a) and a latter (for example: b), and the former (a) is taken as a first category and the latter (b) is taken as a second category to form a co-purchase relationship item (a, b). At the same time, the latter (b) is taken as a first category and the former (a) is taken as a second category to form another co-purchase relationship item (b, a). When there are not, it means that the corresponding two different categories are not co-purchased by any user, and the corresponding co-purchase relationship item cannot be formed. The purchase of two different categories in a short time in succession is manifested as the time difference between the order timestamps of the corresponding two historical purchase orders being less than a preset time length. The preset time length can be configured by a person skilled in the art as needed, for example, 3 minutes.

[0058] Further, for each co-purchase relationship item obtained above, all historical purchase orders for purchasing the goods of the first category in the co-purchase relationship item are determined, and the total number of different users is determined from the users outputting these historical purchase orders as the purchase user number of the first category. Subsequently, the co-purchase user number of the two categories in each co-purchase relationship item is divided by the purchase user number of the first category in the co-purchase relationship item to obtain the co-purchase condition probability of each co-purchase relationship item. Each co-purchase relationship item and its co-purchase condition probability constitute a co-purchase relationship set.

[0059] Step S1202, for each accessed target category, the co-purchase condition probability of the co-purchase relationship item with the target category as the second category is queried in the co-purchase relationship set; Based on the constructed co-purchase relationship set, for each target category accessed by the current evaluation user, the co-purchase condition probability of all co-purchase relationship items with the target category as the second category is queried. The co-purchase condition probability represents the probability of the user purchasing the goods of the target category on the premise of purchasing the goods of the first category in the corresponding co-purchase relationship item. For example, if there is a co-purchase relationship item (fruit knife, apple) in the co-purchase relationship set, and the co-purchase condition probability is 0.15, then among all users who have purchased the fruit knife, 15% of the users have also purchased the apple.

[0060] Step S1203, when the co-purchase condition probability is not found, the co-purchase relationship score of the accessed target category is determined to be zero; If no co-purchase relationship item with the target category as the second category is found in the co-purchase relationship set, it indicates that there is no evidence in the historical purchase records of the independent station that other categories have a significant co-purchase pattern with the target category, and the co-purchase relationship score of the target category is directly determined to be zero.

[0061] Step S1204, when the co-purchase condition probability is found, the co-purchase relationship score of the target category is determined according to each co-purchase condition probability and the user preference score of the first category in the user preference score distribution in the co-purchase relationship item of the co-purchase condition probability. On the contrary, if one or more co-purchase relationship items are found, the final co-purchase relationship score of the target category needs to be calculated by integrating these information. The calculation process needs to combine two aspects of data: one is the co-purchase condition probability of each co-purchase relationship item itself, and the other is the user preference score of the first category in the user preference score distribution of the current user corresponding to these co-purchase relationship items. The user preference score reflects the historical interest intensity of the user to the category. A specific calculation method is weighted summation: the co-purchase condition probability of each found co-purchase relationship item is multiplied by the user preference score of the first category in the co-purchase relationship item, and then all the product results are added to obtain the sum, which is the co-purchase relationship score of the target category. Its calculation formula can be expressed as: co-purchase relationship score = Σ(co-purchase condition probability i × user preference score (first category i)). For example, assuming that the target category is "diapers", two related items are found from the co-purchase relationship set: 1. (beer, diapers), the co-purchase condition probability is 0.25; 2. (wet wipes, diapers), the co-purchase condition probability is 0.18. At the same time, it is found from the user's preference score distribution that the user's preference score for "beer" is 8.5, and the user's preference score for "wet wipes" is 6.0. Then the co-purchase relationship score of "diapers" = (0.25 × 8.5) + (0.18 × 6.0) = 2.125 + 1.08 = 3.205.

[0062] Step S1205, forming a co-purchase relationship score distribution with the co-purchase relationship score of each target category.

[0063] Each target category is associated with its co-purchase relationship score to form a co-purchase relationship score distribution.

[0064] In this embodiment, by deeply analyzing the historical purchase order data of all users in the independent station, the co-occurrence purchase pattern between categories is extracted, and the correlation strength is quantified in the form of conditional probability. It can be seen that potential association patterns such as "beer and diapers" can be found from group consumption behavior. When calculating the category co-purchase relationship score of a specific user, the global co-purchase conditional probability and the user's own preference score for associated categories are combined and weighted to make the score contain both the general consumption law of the public and the unique interest preference of the individual. Therefore, the recommendation result not only conforms to the general trend, but also further reflects the personalized needs of the user, effectively guiding the user to discover novel goods related to his historical interests, broadening the exploration space and alleviating the information cocoon effect.

[0065] In further embodiments, step S1200, based on the historical conversion paths of all users in the independent station, determines the conversion relationship score distribution of the plurality of target categories, including the following steps: Step S1210, based on the historical conversion paths of all users in the independent station, determines the corresponding conversion relationship set, the conversion relationship set includes a plurality of conversion relationship items and their conversion conditional probability, the conversion relationship item includes a third category and a fourth category, and the conversion conditional probability is the probability of purchasing the fourth category based on browsing the third category as a premise; For each user in the independent station, the historical behavior data of the user is obtained, a plurality of interaction behavior information arranged in the order of new to old according to the interaction occurrence is traversed, the interaction behavior information of each interaction behavior type as purchase is determined, and the first interaction behavior information corresponding to the target category with the interaction behavior type as browsing and not containing the interaction behavior information is determined after the interaction behavior information of each interaction behavior type as purchase. Subsequently, for each interaction behavior information of the interaction behavior type as purchase, the target category in the interaction behavior information is taken as the end node, the first interaction behavior information after the interaction behavior information with the interaction behavior type as browsing is taken as the starting node, the target category in the interaction behavior information is taken as the starting node, and the starting node and the end node form a historical conversion path. Accordingly, all historical conversion paths corresponding to the user can be obtained.

[0066] Further, first, different individual historical conversion paths are filtered out from all historical conversion paths of all users, and then the starting node and the ending node in each historical conversion path are taken as the third category and the fourth category in the conversion relationship item, thereby obtaining the corresponding individual conversion relationship item. For each conversion relationship item, it is determined whether the historical conversion path corresponding to the conversion relationship item exists in all historical conversion paths of each user, and the total number of different users corresponding to the existing is accumulated as the conversion user number corresponding to the purchase of the fourth category of goods after the browsing of the third category of goods in the conversion relationship item; in addition, it is queried whether the corresponding interaction behavior information containing the third category in the conversion relationship item and the interaction behavior type of browsing exists in the historical behavior data of each user, and the total number of different users corresponding to the existing is accumulated as the browsing user number of the third category in the conversion relationship item. Finally, the conversion user number is divided by the browsing user number to obtain the conversion condition probability of the conversion relationship item. Each conversion relationship item and the conversion condition probability thereof constitute a conversion relationship set.

[0067] Step S1220, for each accessed target category, the conversion condition probability of the conversion relationship item in which the target category is the fourth category is queried in the conversion relationship library; Based on the constructed conversion relationship set, for each target category accessed by the current evaluation user, the conversion condition probability of all conversion relationship items in which the target category is the fourth category is queried. The conversion condition probability represents the probability of the user purchasing the fourth category of goods on the premise of browsing the third category of goods corresponding to the conversion relationship item. For example, if the conversion relationship set contains the item (top, trousers) and the conversion condition probability thereof is 0.15, it is indicated that 15% of the users who have purchased trousers have browsed tops before purchasing.

[0068] Step S1230, when the conversion condition probability is not found, the conversion relationship score of the accessed target category is determined to be zero; If no conversion relationship item in which a target category is the fourth category is found in the conversion relationship set for the target category, it is indicated that the historical conversion record of the independent station lacks evidence that other categories and the target category have a significant conversion mode, and the conversion relationship score of the target category is directly determined to be zero.

[0069] Step S1240, when the conversion condition probability is found, the conversion relationship score of the accessed target category is determined according to each conversion condition probability and the user preference score of the third category in the conversion relationship item in the user preference score distribution of the user preference score distribution; Conversely, if one or more conversion relationship items are queried, the information needs to be integrated to calculate the final conversion relationship score of the target category. The calculation process needs to combine two aspects of data: one is the conversion condition probability of each conversion relationship item itself, and the other is the user preference score corresponding to the third category in the user preference score distribution of the current user in these conversion relationship items. The user preference score reflects the historical interest intensity of the user to the category. A specific calculation method is weighted summation: multiply the conversion condition probability of each queried conversion relationship item with the user preference score of the third category in the conversion relationship item, and then add all the product results to obtain the sum, which is the conversion relationship score of the target category. The calculation formula can be expressed as: conversion relationship score = Σ(conversion condition probability i × user preference score(third category i)). For example, assuming that the target category is "pants", two relevant items are queried from the conversion relationship set: 1. (top, pants), the conversion condition probability is 0.15; 2. (shoes, pants), the conversion condition probability is 0.08. At the same time, it is known from the user's preference score distribution that the user's preference score for "top" is 10, and the user's preference score for "shoes" is 15. Then the conversion relationship score of "pants" = (0.15 × 10) + (0.08 × 15.0) = 1.5 + 1.2 = 2.7.

[0070] Step S1250, forming a conversion relationship score distribution with the conversion relationship score of each target category.

[0071] Associating each target category with its conversion relationship score forms a conversion relationship score distribution.

[0072] In this embodiment, the behavior conversion path of the user from browsing to purchasing is analyzed in depth, the conversion condition probability between categories is constructed based on the historical behavior of all users on the site, and the migration rule of the user interest from preliminary cognition to final decision is accurately described. When calculating the conversion relationship score of a certain category for a specific user, the weighted strategy of combining global conversion probability with personal preference of the user to the source category is also adopted. This makes the final score not only capture the general conversion trend of "people who have seen this also buy that", but also sensitively reflect the interest intensity of the user itself to the potential lead category, so as to more accurately predict the next purchase intention of the user. The recommended goods generated accordingly are more likely to meet the potential but not explicitly expressed consumption demand of the user, effectively guide the user to complete the decision-making closed loop, and improve the conversion efficiency and business value of the recommendation.

[0073] In a further embodiment, step S1400, determining the recommended goods pushed to the user based on the category interest score distribution, includes the following steps: Step S1410, determining the recommended scene corresponding to the goods recommendation event; The server, in response to a user recommendation event, acquires a unique identifier of a recommendation scenario carried in the event, which is used to uniquely distinguish different recommendation service scenarios, which can be a predefined scenario number or descriptive text, which can be configured by those skilled in the art according to actual needs. Thus, the recommendation scenario uniquely indicated by the identifier can be determined.

[0074] Step S1420, calling a preset recommendation sampling strategy corresponding to the recommendation scenario to sample corresponding recommended goods from the category interest score distribution; For each recommendation scenario, the platform pre-configures a suitable recommendation sampling strategy to filter out the recommended goods most suitable for the scenario from the category interest score distribution. The recommendation sampling strategy includes but is not limited to: deterministic filtering strategy, inverted filtering strategy, and exploration and utilization trade-off strategy (such as ε-greedy strategy or Thompson sampling strategy), etc., which can be selected and configured by those skilled in the art according to actual business needs.

[0075] In an embodiment, for the home page recommendation column scenario, the preset strategy is an exploration and utilization trade-off strategy, which is specifically configured as follows: with a probability of 90% (utilization), select the goods corresponding to the top K categories in the category interest score distribution in terms of score, and with a probability of 10% (exploration), randomly select goods from the top M categories in terms of score but not the highest, where K and M are both preset positive integers, which can be configured by those skilled in the art. For the shopping cart additional recommendation scenario, the preset strategy is a deterministic filtering strategy, which is specifically configured as follows: select categories that have high co-purchase relationships or high conversion relationships with the categories to which the goods in the current shopping cart belong, and further select goods with the highest category interest scores from these categories. High co-purchase relationship can be a category with high co-purchase frequency with the current category, and high conversion relationship can be a category with high conversion frequency with the current category. For the exclusive recommendation area of the user personal center, the preset strategy is an inverted filtering strategy, which sorts the category interest scores in the category interest score distribution in descending order, and selects at least one category corresponding to the top category interest scores. Part of the goods in these categories are selected as recommended goods.

[0076] Step S1430, pushing the recommended goods to the user for display in the corresponding recommendation scenario.

[0077] The server returns the sampled recommended commodities to the terminal device of the user. After receiving, the terminal device dynamically renders and displays the recommended commodities in a specific area (such as a "Guess You Like" module) corresponding to the recommended scene in the original visual interface triggering the commodity recommendation event. The user can further interact with the recommended commodities through clicking, adding to cart, purchasing and other interactive behaviors. These behavior data will be recorded and fed back to the system for subsequent user behavior data analysis, thereby forming a recommendation closed loop.

[0078] In this embodiment, by introducing a differentiated sampling strategy deeply bound to the recommended scene, it is realized that the business objectives and user psychological expectations of different recommended scenes are essentially different. By presetting or dynamically configuring the most suitable sampling strategy for different scenes, the high compatibility of the recommended content and the scene context is ensured. Then, the commodities sampled according to the accurate category interest score distribution are accurately delivered to the specific scene interface currently occupied by the user for display. It can be seen that the relevance, situational awareness and user experience immersion of the recommended results are greatly improved, so that the diversified business demands can be flexibly and intelligently met.

[0079] In a further embodiment, after step S1400, determining the recommended commodities pushed to the user based on the category interest score distribution, the following steps are included: Step S1500, determining the feedback performance score of recommending at least one recommended commodity to multiple users; After the server returns the recommended commodities to the user terminal, it continuously collects the actual feedback data of each recommended commodity at the user terminal. The feedback data includes but is not limited to one or more of the number of exposures, the number of clicks, the number of adds to cart, the number of purchases, the number of collections, the number of shares, the page dwell time and the conversion rate. The feedback performance score is calculated based on a preset scoring model, which can be configured to assign different weights to different feedback indicators and perform weighted summation. In one embodiment, the feedback performance score obtained by the preconfigured scoring model = 0.3 x click rate + 0.4 x purchase conversion rate + 0.2 x collection rate + 0.1 x average dwell time. Wherein, the click rate is the ratio of the number of clicks to the number of exposures, the purchase conversion rate is the ratio of the number of purchases to the number of exposures, and the collection rate is the ratio of the number of collections to the number of exposures. The feedback data is collected in real time through client-side tracking or server-side logs, and is stored in a feedback database in association with the unique identifier of the commodity, the unique identifier of the user and the unique identifier of the recommended scene, for subsequent periodic or real-time calculation of the feedback performance score of each recommended commodity.

[0080] Step S1510, constructing a preselected recommended commodity set by preferentially selecting recommended commodities with high feedback performance scores, for determining the recommended commodities pushed to new users.

[0081] According to the preset scoring threshold and / or the inverted list, the recommended commodities with high feedback performance scores or exceeding the scoring threshold are filtered from the feedback database, and are aggregated into a preselected recommended commodity set. In specific implementation, one or more of the following manners can be included: filtering commodities with feedback performance scores greater than a preset threshold, which can be 0.7, and can be set by a person skilled in the art as needed; and / or selecting commodities with top N feedback performance scores in each category, where N is a preset positive integer and can be set by a person skilled in the art as needed. In other embodiments, commodities with feedback performance scores continuously higher than the average level in a recent period of time and with cumulative exposure reaching a certain scale can also be selected. The preselected recommended commodity set is stored in a cache or a database in a structured data form, and a timing task is enabled to update the preselected recommended commodity set according to the latest feedback performance scores. When recommending for a new user, all or part of commodities in the preselected recommended commodity set can be selected as the recommended result, thereby improving the stability and effectiveness of the recommended result.

[0082] In this embodiment, user feedback data of recommended commodities are collected and quantified, comprehensive performance scores of the recommended commodities are calculated, high-quality commodities are filtered and a preselected recommended commodity set is constructed, which is used for recommendation for new users. This mechanism can dynamically evaluate the actual performance of commodities, fully utilize historical recommendation effectiveness data, filter commodities with high feedback scores, high conversion rates or high interaction rates, effectively alleviate the new user cold start problem, and improve the reliability and user experience of the recommended result. Meanwhile, the preselected recommended commodity set is continuously updated, dynamic optimization of recommended content is realized, and the adaptability and effectiveness of the recommendation system are further improved.

[0083] Please refer to Figure 3 , which is a personalized commodity recommendation device provided for one of the purposes of the present application, and is a functional embodiment of the personalized commodity recommendation method of the present application. On the other hand, the personalized commodity recommendation device provided for one of the purposes of the present application includes an event response module 1100, a distribution determination module 1200, a distribution fusion module 1300 and a user recommendation module 1400. The event response module 1100 is configured to respond to a commodity recommendation event of an independent in-store user, determine a user preference score distribution and a repurchase trend score distribution of a plurality of target categories visited by the user in the independent store based on historical behavior data of the user. The distribution determination module 1200 is configured to determine a co-purchase relationship score distribution and a conversion relationship score distribution of the plurality of target categories based on historical purchase orders and historical conversion paths of all users in the independent store. The distribution fusion module is configured to weight and fuse the user preference score distribution, the repurchase trend score distribution, the co-purchase relationship score distribution and the conversion relationship score distribution into a category interest score distribution. The user recommendation module 1400 is configured to determine recommended commodities pushed to the user based on the category interest score distribution.

[0084] In a further embodiment, the event response module 1100 comprises: a first data acquisition submodule for acquiring historical behavior data of the user, the historical behavior data comprising a plurality of interaction behavior information, the interaction behavior information comprising a target category of the user's current interaction, an interaction timestamp, and an interaction behavior type; an information determination submodule for dividing the historical behavior data using a rolling time window mechanism and determining interaction behavior information corresponding to each time window according to the interaction timestamp in the interaction behavior information; a first score determination submodule for determining, for each time window, a to-be-adjusted preference score of a target category according to all interaction behavior types corresponding to the target category in the time window and interaction frequencies of the interaction behavior types; a score adjustment submodule for determining a preference adjustment coefficient of each time window using a time decay algorithm and adjusting the to-be-adjusted preference score of each target category in the corresponding time window to a calibrated preference score; and a first distribution formation submodule for determining, for each different target category in the historical behavior data, a user preference score according to the calibrated preference score of the target category in the corresponding time window and forming a user preference score distribution.

[0085] In a further embodiment, the event response module 1100 comprises: a second data acquisition submodule for acquiring historical behavior data of the user, the historical behavior data comprising a plurality of interaction behavior information, the interaction behavior information comprising a target category of the user's current interaction, an interaction timestamp, and an interaction behavior type; a frequency determination submodule for determining, for each different target category in the historical behavior data, an interaction frequency of a purchase behavior type corresponding to the target category; and a second distribution formation submodule for normalizing the interaction frequency corresponding to each target category to obtain a repurchase trend score of each target category and form a repurchase trend score distribution.

[0086] In a further embodiment, the distribution determining module 1200 comprises: a first set determining submodule, configured to determine a corresponding co-purchase relationship set based on historical purchase orders of all users in the independent station, the co-purchase relationship set comprising a plurality of co-purchase relationship items and co-purchase condition probabilities thereof, the co-purchase relationship item comprising a first product category and a second product category, and the co-purchase condition probability being a probability of purchasing the second product category on the premise of purchasing the first product category; a first probability querying submodule, configured to query, for each accessed target product category, the co-purchase condition probability of the co-purchase relationship item of the second product category in the co-purchase relationship set; a second score determining submodule, configured to determine a co-purchase relationship score of the accessed target product category as zero when the co-purchase condition probability is not found; a third score determining submodule, configured to determine the co-purchase relationship score of the accessed target product category according to each co-purchase condition probability and a user preference score of the first product category in the co-purchase relationship item in the user preference score distribution; and a third distribution forming submodule, configured to form a co-purchase relationship score distribution based on the co-purchase relationship score of each target product category.

[0087] In a further embodiment, the distribution determining module 1200 comprises: a second set determining submodule, configured to determine a corresponding conversion relationship set based on historical conversion paths of all users in the independent station, the conversion relationship set comprising a plurality of conversion relationship items and conversion condition probabilities thereof, the conversion relationship item comprising a third product category and a fourth product category, and the conversion condition probability being a probability of purchasing the fourth product category on the premise of browsing the third product category; a second probability querying submodule, configured to query, for each accessed target product category, the conversion condition probability of the conversion relationship item of the fourth product category in the conversion relationship set; a fourth score determining submodule, configured to determine a conversion relationship score of the accessed target product category as zero when the conversion condition probability is not found; a fifth score determining submodule, configured to determine the conversion relationship score of the accessed target product category according to each conversion condition probability and a user preference score of the third product category in the conversion relationship item in the user preference score distribution; and a fourth distribution forming submodule, configured to form a conversion relationship score distribution based on the conversion relationship score of each target product category.

[0088] In a further embodiment, the user recommendation module 1400 comprises: a scenario determining submodule, configured to determine a recommendation scenario corresponding to the commodity recommendation event; a commodity determining submodule, configured to call a preset recommendation sampling strategy corresponding to the recommendation scenario to sample a corresponding recommended commodity from the product category interest score distribution; and a commodity pushing submodule, configured to push the recommended commodity to the user for display in the corresponding recommendation scenario.

[0089] In a further embodiment, the user recommendation module 1400 further comprises a feedback determination submodule, configured to determine a feedback performance score of the at least one recommended commodity corresponding to the plurality of users; and a commodity optimization submodule, configured to optimize the recommended commodity with the feedback performance score to construct a preselected recommended commodity set, and determine the recommended commodity pushed to the new user.

[0090] To solve the above technical problems, the embodiments of the present application further provide a computer device. As shown in Figure 4 the internal structure diagram of the computer device. The computer device comprises a processor, a computer readable storage medium, a memory and a network interface connected through a system bus. The computer readable storage medium of the computer device stores an operating system, a database and computer readable instructions. The database can store a control information sequence. The computer readable instructions, when executed by the processor, can enable the processor to implement a personalized commodity recommendation method. The processor of the computer device is configured to provide computing and control capabilities to support the operation of the entire computer device. The memory of the computer device can store computer readable instructions. The computer readable instructions, when executed by the processor, can enable the processor to execute the personalized commodity recommendation method of the present application. The network interface of the computer device is configured to communicate with the terminal. Those skilled in the art can understand that Figure 4 the structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0091] In the present embodiment, the processor is configured to execute the specific functions of each module and its submodules in Figure 3 . The memory stores the program codes and various data required to execute the above modules or submodules. The network interface is configured to transmit data between the user terminal or the server. The memory in the present embodiment stores the program codes and data required to execute all modules / submodules in the personalized commodity recommendation device of the present application. The server can call the program codes and data of the server to execute the functions of all submodules.

[0092] The present application further provides a storage medium storing computer readable instructions. The computer readable instructions, when executed by one or more processors, enable the one or more processors to perform the steps of the personalized commodity recommendation method of any embodiment of the present application.

[0093] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiments of the method of the present application can be completed by a computer program instructing relevant hardware. The computer program can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the above-mentioned embodiments of the method. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0094] In summary, based on the historical behavior data, the present application dynamically and multi-dimensionally calculates the user preference score distribution and the repeat purchase trend score distribution of the user for multiple target categories, effectively capturing the real-time interest and repeat purchase tendency of the user. At the same time, by combining the historical purchase orders and conversion paths of the global users of the independent station, the co-purchase relationship and conversion relationship between categories are mined to generate the corresponding co-purchase relationship score distribution and conversion relationship score distribution. The precise category interest score distribution is formed by weighted integration of the above distributions, thereby fusing individual behavior characteristics and group behavior patterns. Differentiated sampling strategies can be adapted according to different recommendation scenarios, and finally the context-aware personalized product recommendation is realized, which significantly improves the accuracy, diversity and adaptability of the recommendation, and enhances the user shopping experience and platform conversion rate. In addition, the prior goods with good recommendation effectiveness can be collected for recommendation to new users, effectively alleviating the problem of sparse behavior data and cold start.

[0095] Those skilled in the art can understand that the steps, measures, and schemes in the various operations, methods, and processes discussed in the present application can be alternated, changed, combined, or deleted. Further, other steps, measures, and schemes in the various operations, methods, and processes discussed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted. Further, the steps, measures, and schemes in the prior art with the various operations, methods, and processes discussed in the present application can also be alternated, changed, rearranged, decomposed, combined, or deleted.

[0096] The above only describes some embodiments of the present application. It should be noted that those skilled in the art can make several improvements and refinements without departing from the principles of the present application, and these improvements and refinements should also be considered within the scope of protection of the present application.

Claims

1. A personalized product recommendation method, characterized in that, Includes the following steps: In response to product recommendation events by users within the independent website, based on the users' historical behavior data, determine the distribution of user preference scores and repurchase trend scores for multiple target product categories accessed by the users on the independent website; Based on the historical purchase orders and historical conversion paths of all users within the independent website, the distribution of co-purchase relationship scores and conversion relationship scores for the multiple target product categories are determined accordingly. The user preference score distribution, repurchase trend score distribution, co-purchase relationship score distribution, and conversion relationship score distribution are weighted and merged into a category interest score distribution; Recommended products are determined and pushed to the user based on the category interest score distribution.

2. The personalized product recommendation method according to claim 1, characterized in that, Based on the user's historical behavior data, the distribution of user preference scores for multiple target product categories accessed by the user on the independent website is determined, including the following steps: Acquire the user's historical behavior data, which includes multiple interaction behavior information, including the user's target product category, interaction timestamp, and interaction behavior type for the current interaction. The historical behavior data is divided using a rolling time window mechanism, and the interaction behavior information within each time window is determined based on the interaction timestamp in the interaction behavior information. For each time window, the adjustment preference score of the target product category is determined based on all interaction behavior types and their interaction frequencies corresponding to the target product category within the time window. The time decay algorithm is used to determine the preference adjustment coefficient for each time window, and the preference score to be adjusted for each target category in the corresponding time window is adjusted to the calibration preference score. For each different target category in the historical behavior data, the user preference score is determined based on the calibration preference score of the target category in each corresponding time window, thus forming a user preference score distribution.

3. The personalized product recommendation method according to claim 1, characterized in that, Based on the user's historical behavior data, the distribution of repurchase trend scores for multiple target product categories accessed by the user on the independent website is determined, including the following steps: Acquire the user's historical behavior data, which includes multiple interaction behavior information, including the user's target product category, interaction timestamp, and interaction behavior type for the current interaction. For each different target category in the historical behavior data, determine the interaction frequency of the corresponding purchase behavior type for that target category; The interaction frequency corresponding to each target product category is normalized to obtain the repurchase trend score of each target product category, forming the repurchase trend score distribution.

4. The personalized product recommendation method according to claim 1, characterized in that, Based on the historical purchase orders of all users within the independent website, the distribution of co-purchase relationship scores for the multiple target product categories is determined, including the following steps: Based on the historical purchase orders of all users within the independent website, a corresponding co-purchase relationship set is determined. The co-purchase relationship set contains multiple co-purchase relationship items and their co-purchase condition probabilities. The co-purchase relationship items include a first category and a second category. The co-purchase condition probability is the probability of purchasing the second category at the same time as the first category. For each target product category visited, query the co-purchase condition probability of the co-purchase relationship item in the co-purchase relationship set where the target product category is the second product category; When the probability of the co-purchase condition is not found, the co-purchase relationship score of the target product category is determined to be zero. When the co-purchase condition probability is found, the co-purchase relationship score of the target category being accessed is determined based on each co-purchase condition probability and the user preference score of the first category within the co-purchase relationship item in the user preference score distribution. A co-purchase relationship score distribution is formed based on the co-purchase relationship score for each of the target product categories.

5. The personalized product recommendation method according to claim 1, characterized in that, Based on the historical conversion paths of all users within the independent website, the conversion relationship score distribution of the multiple target product categories is determined, including the following steps: Based on the historical conversion paths of all users within the independent website, a corresponding conversion relationship set is determined. The conversion relationship set includes multiple conversion relationship items and their conversion condition probabilities. The conversion relationship items include a third category and a fourth category. The conversion condition probability is the probability of purchasing the fourth category given that browsing the third category is a prerequisite. For each target product category visited, query the conversion condition probability of the conversion relationship item in the conversion relationship database where the target product category is the fourth product category; When the conversion condition probability is not found, the conversion relationship score of the target product category is determined to be zero. When the conversion condition probability is found, the conversion relationship score of the target category being accessed is determined based on the user preference score of the third category within each conversion condition probability and the user preference score of the third category in the user preference score distribution. A conversion relationship score distribution is formed based on the conversion relationship score of each target product category.

6. The personalized product recommendation method according to claim 1, characterized in that, Determining recommended products to be pushed to the user based on the category interest score distribution includes the following steps: Determine the recommendation scenario corresponding to the product recommendation event; The preset recommendation sampling strategy corresponding to the recommendation scenario is invoked to sample the corresponding recommended products from the category interest score distribution; The recommended products are pushed to the user for display in the appropriate recommendation scenario.

7. The personalized product recommendation method according to claim 1, characterized in that, After determining the recommended products to be pushed to the user based on the category interest score distribution, the following steps are included: Determine the feedback performance score for recommending at least one corresponding recommended product to multiple users; The recommended products based on the feedback performance scores are used to construct a pre-selected set of recommended products, which is used to determine the recommended products to be pushed to new users.

8. A personalized product recommendation device, characterized in that, include: The event response module is used to respond to product recommendation events by users within the independent website. Based on the user's historical behavior data, it determines the distribution of user preference scores and repurchase trend scores for multiple target product categories accessed by the user on the independent website. The distribution determination module is used to determine the co-purchase relationship score distribution and conversion relationship score distribution of the multiple target product categories based on the historical purchase orders and historical conversion paths of all users within the independent site. The distribution fusion module is used to weightedly fuse the user preference score distribution, repurchase trend score distribution, co-purchase relationship score distribution, and conversion relationship score distribution into a category interest score distribution; The user recommendation module is used to determine recommended products to be pushed to the user based on the category interest score distribution.

9. A computer device comprising a central processing unit and a memory, characterized in that, The central processing unit is used to invoke and run a computer program stored in the memory to perform the steps of the method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, It stores, in the form of computer-readable instructions, a computer program implemented according to any one of claims 1 to 7, which, when invoked by a computer, executes the steps included in the corresponding method.

Images