A commodity pushing method and system

Abstract

The present invention discloses a product push method and system, which relate to the technical field of product push, and are used to solve the problem of inaccurate recommendations and reduce the exploratory nature of the recommendation system. By identifying the user's product list and user behavior data, the classification tag loading amount and the user's favorite tag coefficient are obtained, and the adaptation index of each classification tag and the current user is comprehensively obtained. The search share and recommendation ratio of different classification tags in the entire platform are retrieved, and the search share and recommendation ratio of different classification tags are integrated and divided into each classification tag using a weighted algorithm. The tags are sorted according to the adaptation index of each classification tag and the current user, and the classification tags recommended to the user are determined according to a preset selected threshold, so as to improve the accuracy of personalized recommendations, avoid recommendation overload, enhance the exploratory nature of recommended products, and meet the personalized needs of users.

Description

Technical Field

[0001] The present invention relates to the technical field of product push, and more specifically, to a product push method and system. Background Art

[0002] Existing product recommendation systems often base recommendations on users' browsing and purchasing history, or use simple collaborative filtering algorithms. These methods rely on historical user behavior data on e-commerce platforms, including browsing history, clicks, add-to-cart purchases, and final purchase decisions. The system analyzes these behavioral patterns to predict user interests and recommend relevant products.

[0003] The existing technology has the following deficiencies:

[0004] Currently, existing recommendation systems primarily rely on historical user behavior data, such as browsing, clicking, and purchasing records. However, these behaviors may not fully reflect a user's true interests. Users may be misclassified due to accidental browsing of certain products, leading to biased recommendations. Furthermore, existing methods often fail to dynamically adjust the weights of classification labels, resulting in inaccurate recommendations, exposing users to a limited number of product categories for extended periods, and reducing the exploratory nature of the recommendation system. Therefore, a product push method and system are proposed.

[0005] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not form the prior art that is already known to a person of ordinary skill in the art. Summary of the Invention

[0006] In order to overcome the above-mentioned defects of the prior art, embodiments of the present invention provide a product push method and system, which solve the problems raised in the above-mentioned background technology by using different product inspection methods.

[0007] To achieve the above-mentioned object, the present invention provides the following technical solution: a product push method, comprising S1: identifying a user's product list to obtain a classification tag loading amount;

[0008] S2: Collect user behavior data to obtain the user's favorite label coefficient, and combine the user's favorite label coefficient with the category label load to obtain the adaptation index of each category label and the current user;

[0009] S3: Search the search share and recommendation ratio of different classification tags across the entire platform, integrate the search share and recommendation ratio of different classification tags, and use a weighted algorithm to divide each classification tag;

[0010] S4: Based on the classification results of each classification label, the labels are sorted according to the adaptation index of each classification label and the current user, and the classification label recommended to the user is determined according to a preset selected threshold.

[0011] In a preferred embodiment, the user commodity list comprises a plurality of commodities, and the commodities comprise a plurality of classification tags;

[0012] The occurrence times of each classification tag in the user commodity list are counted, and the occurrence times of each classification tag are ratio calculated with the total occurrence times of all classification tags to obtain a classification tag loading amount;

[0013] Through a preset analysis time, the total stay duration of the user in the analysis time for browsing commodities of different classification tags is determined, and ratio calculation is performed with the average time interval of the user for browsing commodities of corresponding classification tags adjacent to each other to obtain a user favorite tag coefficient.

[0014] In a preferred embodiment, the classification tag loading amount and the user favorite tag coefficient are standardized and then input into a logistic regression calculation to obtain an adaptation index of each classification tag to the current user.

[0015] In a preferred embodiment, the search proportion and the recommendation proportion of different classification tags in the whole platform are obtained by searching user behavior logs and recommendation system logs.

[0016] In a preferred embodiment, the search records of all users in the analysis time are counted, the search keywords are matched to commodity classification tags through classification mapping to obtain the search times of commodities of classification tags, and ratio calculation is performed with the total search times of all commodities in the whole platform to obtain the search proportion of different classification tags in the whole platform.

[0017] The recommended commodity data pushed by the platform to all users are counted to obtain the recommended display times of commodities of each classification tag, and ratio calculation is performed with the total recommended times of all recommended commodities in the whole platform to obtain the recommendation proportion.

[0018] In a preferred embodiment, the search proportion and the recommendation proportion of different classification tags in the whole platform are standardized and then input into a weighting algorithm to obtain a recommendation overload coefficient.

[0019] In a preferred embodiment, the recommendation overload coefficient is compared and analyzed with a preset overload threshold value;

[0020] If the recommendation overload coefficient is greater than or equal to the overload threshold value, the current classification tag is marked as a recommendation overload tag, and an overload signal is generated;

[0021] If the recommendation overload coefficient is less than the overload threshold value, the current classification tag is marked as a regular recommendation tag, and an end signal is generated.

[0022] In a preferred embodiment, the classification results of each classification label include a recommendation overload label and a regular recommendation label; the recommendation overload label is removed, the regular recommendation label is retained, and the regular recommendation label is counted and set as a set;

[0023] Retrieve the adaptation index of each category tag corresponding to each conventional recommended tag in the conventional recommended tag set and the current user, and sort the category tags in descending order according to the adaptation index values ​​of the category tags and the current user;

[0024] A selected threshold is set, and each classification label that is greater than or equal to the selected threshold is marked with the current user's adaptation index to obtain the current marked adaptation index.

[0025] In a preferred embodiment, the recommendation is made according to the ratio of the current marker fitness index to the sum of all marker tag fitness indices as the recommendation ratio.

[0026] A product push system includes an identification and collection module, an adaptation tag module, a retrieval and collection module, a recommendation overload module, and a tag recommendation module; each module is connected by signals;

[0027] The recognition and collection module is used to identify the user's product list and collect user behavior data, obtain the classification label load and the user's favorite label coefficient, and send it to the adaptation label module;

[0028] The adaptation tag module is used to calculate the adaptation index of each category tag and the current user based on the category tag loading amount and the user's favorite tag coefficient using the logistic regression algorithm, and send it to the tag recommendation module;

[0029] The retrieval and collection module is used to retrieve the search share and recommendation ratio of different classification tags in the entire platform and send them to the recommendation overload module;

[0030] The recommendation overload module is used to divide each category tag into recommendation overload tags and regular recommendation tags based on the search share and recommendation ratio of different category tags in the entire platform using a weighted algorithm, and send them to the tag recommendation module;

[0031] The tag recommendation module is used to filter out common recommendation tags, sort the tags according to the compatibility index between each category tag and the current user, select and sort the tags according to the preset selected threshold, and then recommend products to the user.

[0032] Technical effects and advantages of the present invention:

[0033] 1. The present invention obtains the classification tag loading amount and the user's favorite tag coefficient by identifying the user's product list and user behavior data, comprehensively obtains the adaptation index of each classification tag with the current user, and retrieves the search share and recommendation ratio of different classification tags in the entire platform. After integrating the search share and recommendation ratio of different classification tags and using a weighted algorithm to divide each classification tag, the tags are sorted according to the adaptation index of each classification tag with the current user, and the classification tags recommended to the user are determined according to the preset selected threshold, thereby improving the accuracy of personalized recommendations, avoiding recommendation overload, enhancing the exploratory nature of recommended products, and meeting the personalized needs of users. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] Figure 1 The present invention is a method flow chart of a product push method.

[0035] Figure 2 This is a module diagram of a product push system of the present invention. DETAILED DESCRIPTION

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0037] Example 1

[0038] See also Figure 1 , a product push method, the specific operation process is as follows:

[0039] S1: Identify the user's product list, determine the category label of the merchant corresponding to the product, and calculate the proportion of each category label in the user's product list as the category label loading volume;

[0040] The user's product list is obtained through the following data sources, such as browsing history (such as the URL of the webpage visited by the user, the time the user stays on the link, the number of times the user clicks on the link, etc.), data on products that the user has added but not purchased, user purchase records (products that have been successfully traded), and products that the user has actively collected, etc. The specific data source channels are not limited and will not be detailed here;

[0041] Set user The product list is:

[0042] ;

[0043] Where j = 1, 2, …, n; For the Products;

[0044] Specifically, in order to assign classification labels to products and establish a product-merchant-category mapping relationship, classification is usually based on the existing product classification system of the e-commerce platform. The specific classification method is not limited, but is based on the number of products and types of merchants on the actual product operation platform. It will not be detailed here.

[0045] Through the above classification method, a product information database is constructed;

[0046] Generally speaking, the product information database contains: the unique identifier of the product ID, the merchant ID to which it belongs, and the product category label (which may include multi-level classification);

[0047] Set up the first Products Has the following classification labels:

[0048] ;

[0049] in, Representative products The set of classification labels to which it belongs, k=1, 2, ..., m, is the k-th classification label;

[0050] For example, if the product name is ipone15, its corresponding category label can be: ;

[0051] Get each product by querying the product database Category label ;

[0052] Calculate the user classification label set and The product category label set is defined as:

[0053] ;

[0054] This formula expresses the union of the category labels of all products in the user's product list;

[0055] The category tag load refers to the proportion of a certain category tag in the user's product list. It can be calculated by counting the frequency of all products belonging to different category tags. Its acquisition logic is to first count the number of occurrences of each category tag in the user's product list, that is:

[0056] ;

[0057] Where, Indicates product Whether it belongs to the classification label If it belongs to the category label, the value is 1, otherwise the value is 0;

[0058] The occurrence frequency of each category label is divided by the total occurrence frequency of all category labels to obtain the category label load ;

[0059] It should be noted that when calculating the above-mentioned category label load, the cumulative value of all category label loads is ensured to be 1, that is, the category label load can be considered as the proportion of each category label in the user's commodity list;

[0060] For example, assuming that a user's commodity list is as follows: commodity A (electronic product, mobile phone), commodity B (electronic product, home appliance), commodity C (home appliance), commodity D (home appliance, smart device), and commodity E (electronic product);

[0061] The occurrence frequency of each category label is counted, that is, electronic product 3 times, mobile phone 1 time, home appliance 3 times, and smart device 1 time;

[0062] The calculation of the category label load shows that the electronic product category label load is 0.375, the mobile phone category label load is 0.125, the home appliance category label load is 0.375, and the smart device category label load is 0.125;

[0063] Through the above steps, the relative weight of each category label in the user's commodity list can be obtained, which provides a basis for subsequent commodity recommendation;

[0064] S2: Collecting user behavior data to obtain a user favorite label coefficient, and comprehensively using a logistic regression algorithm to calculate the adaptation index of each category label and the current user according to the user favorite label coefficient and the category label load;

[0065] The user behavior data is collected through log analysis and database query. Generally, when a user browses a commodity, different behavior data will be generated, including but not limited to browsing time, which refers to the time that the user stays in a commodity detail page or a category page, access frequency, which refers to the user's multiple access to the same category of commodities within a certain time, click times, which refers to the user's click times on a category of commodities, add to cart / collection / purchase behavior, which refers to the user's interest in the category of commodities, and browsing time interval, which refers to the time interval between two times of browsing the same category;

[0066] The user favorite label coefficient is obtained by analyzing the user behavior data;

[0067] The user like label coefficient refers to the interest degree of the user to a certain classification label, the coefficient is used to quantify the preference of the user to different commodity classifications, is calculated through the behavior data of the user such as the browsing time, the access frequency, the click number, and is mainly used to measure the attention degree of the user to a certain commodity, and the acquisition logic is to determine the total stay time of the user in browsing different classification label commodities in the analysis time through the preset analysis time, and the average time interval of the user adjacent two times of browsing the corresponding classification label commodities is calculated by ratio, to obtain the user like label coefficient ;

[0068] The preset analysis time is obtained through user behavior log analysis and business requirement setting, and the analysis time window and the statistical period can be determined by the experimenters according to the specific implementation mode, specifically, the active mode of the user (such as daily activity, weekly activity, etc.) is analyzed through historical data, a reasonable analysis time window is set, and the like, which will not be repeated here;

[0069] It should be noted that it is assumed that the classification data browsed by the user is as follows: the classification label mobile phone, the total browsing time is 300 seconds, the average adjacent browsing time interval is 600 seconds, the classification label home appliance, the total browsing time is 200 seconds, the average adjacent browsing time interval is 500 seconds, and the classification label home appliance electronic product, the total browsing time is 500 seconds, and the average adjacent browsing time interval is 800 seconds;

[0070] It can be obtained that the user like mobile phone label coefficient is 0.5, the user like home appliance label coefficient is 0.4, and the user like electronic product label coefficient is 0.625, so it can be obtained that the user frequently browses the electronic product classification label in a short time, and the user like label coefficient of the electronic product classification will increase, indicating that the user is more interested in the electronic product classification;

[0071] The classification label load quantity and the user like label coefficient are standardized, all input variables are converted to the same range, and the contribution of each input to the model is balanced, specifically, the standardization method is Z-score normalization, and the specific formula is expressed as:

[0072] ;

[0073] In the formula, is the original value (such as the classification label load quantity or the user like label coefficient), is the standardized value, the data distribution keeps the original form, but is centralized to 0, and the fluctuation range is within the standard deviation scale, is the mean value of the variable, is the standard deviation of the variable;

[0074] After standardization, the classification label load and the user favorite label coefficient are in the range of [0, 1], and the system can compare and make decisions on a unified scale, thereby improving the accuracy and efficiency of the logistic regression model;

[0075] After standardization, the classification label load and the user favorite label coefficient are in the range of [0, 1], and the system can compare and make decisions on a unified scale, thereby improving the accuracy and efficiency of the logistic regression model;

[0076] ;

[0077] In the formula, is the logistic regression calculation result, that is, the adaptation index of each classification label to the current user, e is the natural base, y is the linear combination term of the logistic regression model, and y can be set as:

[0078] ;

[0079] In the formula, is the bias term, and are the regression coefficients of the classification label load and the user favorite label coefficient, respectively;

[0080] S3: Search for the search proportion and the recommendation proportion of different classification labels in the whole platform, and divide the classification labels by using a weighted algorithm according to the search proportion and the recommendation proportion of different classification labels;

[0081] Specifically, the search proportion and the recommendation proportion of different classification labels in the whole platform are obtained by searching the user behavior log and the recommendation system log.

[0082] Among them, in the recommendation system, the search proportion and the recommendation proportion need to rely on the user behavior log and the recommendation system log, which respectively record the active search behavior and the system recommendation behavior of the user on the platform. Specifically, the user behavior log refers to the interactive data of the user on the platform, which contains the search, click, browse, collect, purchase and other behaviors of the user, and is used to analyze the user preference.

[0083] Specifically, the recommendation system log refers to the record of the platform pushing recommended goods to the user, which is mainly used to track the recommended goods category, exposure situation, user feedback and the like to optimize the recommendation algorithm.

[0084] The search proportion of different classification labels in the whole platform refers to a search proportion measuring the active search interest of a user for a classification label commodity, indicating the proportion of the classification label commodity in all user search behaviors. The obtaining logic is to count the search records of all users in the analysis time, match the search keywords to the commodity classification labels through classification mapping, obtain the search times of the classification label related commodities, and perform ratio calculation with the total search times of all commodities in the whole platform to obtain the search proportion of different classification labels in the whole platform ;

[0085] The analysis time has been described in the embodiment and will not be repeated here.

[0086] It should be noted that the classification mapping is understood by those skilled in the art as a process of converting user search keywords into standardized commodity classification labels, that is, matching unstructured search words with structured commodity classification systems to ensure that search behavior can be accurately attributed to the corresponding commodity category. Through unified search data format, it is ensured that the search keywords can be correctly classified into the existing commodity classification system, improving the accuracy of search data analysis and making the search proportion calculation more accurate.

[0087] Specifically, the classification mapping can be based on rule matching. First, a commodity classification dictionary is established: the commodity classification system is defined in advance, including main classification, sub-classification, synonyms, etc. For example: the corresponding keywords of mobile phones are (iPhone, Android phone, smart phone), the corresponding keywords of mobile phone accessories are (phone case, protective film, charger), and the corresponding keywords of home appliances are (refrigerator, washing machine, air conditioner). Then, keyword matching is performed: for example, when a user searches for “iPhone 15”, the dictionary is matched and classified into “mobile phone”. When a user searches for “charger”, the dictionary is matched and classified into “mobile phone accessories”. Optionally, fuzzy matching is performed using regular expressions or word segmentation algorithms, for example: “Apple mobile phone” can be classified as “mobile phone”, “double-door refrigerator” can be classified as “home appliance”, and the like, which will not be repeated here.

[0088] The recommendation proportion refers to the recommendation strength of the platform for a classification label commodity, that is, the proportion of the commodity in all recommended content. The obtaining logic is to count the recommended commodity data pushed by the platform to all users to obtain the recommended display times of each classification label commodity, and perform ratio calculation with the total recommended times of all recommended commodities in the whole platform to obtain the recommendation proportion .

[0089] The recommended commodity data pushed includes home page recommendation, personalized recommendation, advertisement recommendation, etc., which will not be repeated here.

[0090] Specifically, the recommended display times of each classification label commodity can be obtained based on the recommendation system log. The specific obtaining method has been described in the embodiment.

[0091] The search proportion and the recommendation proportion of different classification labels in the whole platform are standardized according to the above standardization processing formula, and then the recommendation overload coefficient is obtained by using the weighted algorithm, and the specific formula is as follows:

[0092]

[0093] In the formula, is the recommendation overload coefficient, and are preset proportion coefficients of the search proportion and the recommendation proportion of different classification labels in the whole platform, and and are greater than 0;

[0094] According to the formula, the higher the search proportion and the recommendation proportion of different classification labels in the whole platform, the more overloaded the corresponding classification label recommendation is, and the higher the recommendation overload coefficient is. Conversely, the lower the search proportion and the recommendation proportion of different classification labels in the whole platform, the lower the recommendation overload coefficient is.

[0095] The recommendation overload coefficient is compared and analyzed with the preset overload threshold value;

[0096] If the recommendation overload coefficient is greater than or equal to the overload threshold value, the current classification label is marked as a recommendation overload label, and an overload signal is generated;

[0097] If the recommendation overload coefficient is less than the overload threshold value, the current classification label is marked as a regular recommendation label, and an end signal is generated;

[0098] The overload threshold value is obtained by analyzing historical recommendation data and user interaction feedback data. Specifically, the recommendation proportion of each classification label and user click, purchase and other feedback data in history are counted, a reasonable threshold range is set in combination with the recommendation effect, and the critical point of user interest decline caused by too high recommendation intensity is identified in combination with the response of users to different recommendation intensities, such as click-through rate (CTR), conversion rate (CVR) and other indicators, and the overload threshold value is further set. The specific selection of historical period and interaction feedback times is not limited, but is set by the experimenters according to the characteristics of specific classification labels, which is not described here;

[0099] S4: According to the classification label division result, the classification labels marked with selection are recommended to the user according to the label sorting of each classification label and the adaptation index of the current user, and the selection of the classification label is marked according to the preset selection threshold value.

[0100] The classification label division result includes a recommendation overload label and a regular recommendation label.

[0101] ​The recommended overload labels are removed from the classification labels, and the regular recommended labels are reserved for subsequent recommendation;

[0102] The regular recommended labels are counted and set as a set, and the regular recommended label set is denoted as:

[0103] ;

[0104] Among them, represents the i-th classification label that meets the recommendation condition;

[0105] The regular recommended label set is sorted in descending order according to the adaptation index of each classification label and the current user;

[0106] That is, the adaptation index of each classification label corresponding to each regular recommended label in the regular recommended label set is retrieved, and the adaptation index of each classification label and the current user is sorted in descending order according to the value of the adaptation index of each classification label and the current user;

[0107] A set of classification labels corresponding to the regular recommended label set is obtained:

[0108] ;

[0109] At the same time, the descending order is met, and the number of regular recommended labels in the regular recommended label set is corresponding:

[0110] ;

[0111] A selected threshold is set to screen the final recommended label, and the total number of regular recommended labels is greater than the selected threshold number;

[0112] The adaptation index of each classification label and the current user greater than or equal to the selected threshold is marked to obtain the current marked adaptation index, and the adaptation index of each classification label and the current user less than the selected threshold is excluded;

[0113] It should be noted that the selected threshold can be set according to business requirements, for example: taking the top N adaptation index highest label (such as N=5), or setting the minimum value of the adaptation index of each classification label and the current user (such as the adaptation index of each classification label and the current user is not less than 0.35), etc., which will not be repeated here;

[0114] The recommendation ratio is recommended according to the ratio of the current marked adaptation index to the sum of all marked label adaptation indexes;

[0115] ​Specifically, if the recommended ratio is 0.74 and the recommended ratio calculated for all the fitting indexes of the labels is the highest value, the corresponding block ratio of the user page or the platform advertising area is put according to the recommended ratio, and the way of recommendation according to the recommended ratio is not limited, but is determined by the experimenters according to the specific platform operation plan;

[0116] The application obtains the classification label loading amount and the user favorite label coefficient by identifying the user commodity list and the user behavior data, comprehensively obtains the fitting index of each classification label and the current user, searches the search proportion and the recommended proportion of different classification labels in the whole platform, divides the classification labels by using the weighting algorithm according to the search proportion and the recommended proportion of different classification labels, sorts the labels according to the fitting index of each classification label and the current user, determines the classification label recommended to the user according to the preset selected threshold, improves the accuracy of personalized recommendation, avoids recommendation overload, enhances the exploratory of recommended commodities, and meets the personalized needs of the user.

[0117] Embodiment 2

[0118] Please refer to Figure 2 A commodity pushing system, comprising an identification and collection module, a fitting label module, a search and collection module, a recommendation overload module and a label recommendation module; the modules are signal connected;

[0119] The identification and collection module is used for identifying the user commodity list and collecting the user behavior data, obtaining the classification label loading amount and the user favorite label coefficient, and sending to the fitting label module;

[0120] The fitting label module is used for calculating the fitting index of each classification label and the current user according to the classification label loading amount and the user favorite label coefficient by using the logistic regression algorithm, and sending to the label recommendation module;

[0121] The search and collection module is used for searching the search proportion and the recommended proportion of different classification labels in the whole platform, and sending to the recommendation overload module;

[0122] The recommendation overload module is used for dividing each classification label into recommended overload labels and regular recommended labels according to the search proportion and the recommended proportion of different classification labels in the whole platform by using the weighting algorithm, and sending to the label recommendation module;

[0123] The label recommendation module is used for screening the regular recommended labels, sorting the labels according to the fitting index of each classification label and the current user, selecting the label sorting and marking according to the preset selected threshold, and recommending the commodities to the user;

[0124] The above formulas are all dimensionless values calculated, and the formulas are obtained by collecting a large amount of data to simulate a formula of the most recent real situation, and the preset parameters in the formula are set by a person skilled in the art according to the actual situation.

[0125] The above embodiments can be implemented wholly or partially by software, hardware, firmware, or any other combination. When implemented by software, the above embodiments can be implemented wholly or partially in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the processes or functions described in the embodiments of the present application are wholly or partially generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions can be stored in a computer-readable storage medium or transferred from one computer-readable storage medium to another, for example, the computer instructions can be transferred from one website, computer, server, or data center to another website, computer, server, or data center through a wired or wireless (such as infrared, wireless, microwave, etc.) manner. The computer-readable storage medium can be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. containing one or more available medium collections. The available medium can be a magnetic medium (such as a floppy disk, a hard disk, a magnetic tape), an optical medium (such as a DVD), or a semiconductor medium. The semiconductor medium can be a solid-state disk.

[0126] It should be understood that the term "and / or" herein merely describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone, where A and B can be singular or plural. In addition, the character " / " herein generally represents that the associated objects before and after it are in an "or" relationship, but it can also represent an "and / or" relationship, which can be understood according to the context before and after it.

[0127] In this application, "at least one" means one or more, and "multiple" means two or more. "At least one of the following" or similar expressions means any combination of these items, including any combination of single or multiple items. For example, at least one of a, b, or c can represent a, b, c, a-b, a-c, b-c, or a-b-c, where a, b, and c can be single or multiple.

[0128] It should be understood that the size of the sequence number of each process described above does not mean the order of execution, and the execution order of each process should be determined according to its function and inherent logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.

[0129] Those skilled in the art can realize that the units and algorithm steps of each example described in combination with the embodiments disclosed herein can be realized in electronic hardware or a combination of computer software and electronic hardware. Whether the functions are realized in hardware or software depends on the specific application and design constraints of the technical solution. The skilled person can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

[0130] Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the system, device and unit described above can refer to the corresponding process in the foregoing method embodiments, which will not be repeated here.

[0131] In several embodiments provided in the present application, it should be understood that the disclosed system, device and method can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the units is only a logical function division, and actual implementation can have another division manner, for example, a plurality of units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the units shown or discussed can be indirect coupling or communication connection through some interface, device or unit, and can be electrical, mechanical or other forms.

[0132] The units described as separate components can or can not be physically separated, and the components shown as units can or can not be physical units, that is, they can be located in one place, or can be distributed on a plurality of network units. Part or all of the units can be selected according to actual needs to achieve the purpose of the embodiment.

[0133] In addition, each functional unit in each embodiment of the present application can be integrated into a processing unit, or each unit can exist physically independently, or two or more units can be integrated into one unit.

[0134] If the functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the present application essentially or the parts that contribute to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes a plurality of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM), a random access memory (RAM), a magnetic disk or an optical disk, and various media that can store program codes.

[0135] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any person skilled in the art can easily think of changes or replacements within the technical scope disclosed in the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A product push method, characterized by: include: S1: Identify the user's product list and obtain the classification label loading amount; S2: Collect user behavior data to obtain the user's favorite label coefficient, and combine the user's favorite label coefficient with the category label load to obtain the adaptation index of each category label and the current user; S3: Search the search share and recommendation ratio of different classification tags across the entire platform, integrate the search share and recommendation ratio of different classification tags, and use a weighted algorithm to divide each classification tag; S4: Based on the classification results of each category label, the labels are sorted according to the adaptation index of each category label and the current user, and the category label recommended to the user is determined according to the preset selected threshold; The user's product list contains multiple products, and the products contain multiple category tags; Count the number of occurrences of each category tag in the user's product list, calculate the ratio of the number of occurrences of each category tag to the total number of occurrences of all category tags, and obtain the category tag loading amount; The total duration of time users spend browsing products with different category labels during the analysis time is determined by the preset analysis time, and the ratio is calculated with the average time interval between two consecutive browsing times of the user browsing products with the corresponding category labels to obtain the user's favorite label coefficient. Normalize the search share and recommendation ratio of different category tags across the entire platform and substitute them into the weighted algorithm to obtain the recommendation overload coefficient; Compare and analyze the recommended overload coefficient with the preset overload threshold; If the recommended overload coefficient is greater than or equal to the overload threshold, the current classification label is marked as the recommended overload label and an overload signal is generated; If the recommendation overload coefficient is less than the overload threshold, the current classification label is marked as a regular recommendation label and an end signal is generated; The classification results of each category label include recommended overload labels and regular recommended labels; Remove the recommended overload tags, retain the regular recommended tags, and count the regular recommended tags and set them as a collection; Retrieve the adaptation index of each category tag corresponding to each conventional recommended tag in the conventional recommended tag set and the current user, and sort the category tags in descending order according to the adaptation index values ​​of the category tags and the current user; A selected threshold is set, and each classification label that is greater than or equal to the selected threshold is marked with the current user's adaptation index to obtain the current marked adaptation index.

2. A product push method according to claim 1, characterized in that: The classification label loading amount and the user's favorite label coefficient are normalized and then input into the logistic regression calculation to obtain the adaptation index between each classification label and the current user.

3. A product push method according to claim 1, characterized in that: By retrieving user behavior logs and recommendation system logs, we can obtain the search share and recommendation ratio of different category tags in the entire platform.

4. A product push method according to claim 3, characterized in that: Count all user search records during the analysis period, match search keywords to product category tags through category mapping, obtain the number of searches for products related to the category tags, and calculate the ratio of the number of searches to the total number of searches for all products on the platform to obtain the search share of different category tags on the entire platform; The recommended product data pushed by the statistical platform to all users is obtained to obtain the recommended display times of each category label product, and the ratio is calculated with the total recommendation times of all recommended products on the entire platform to obtain the recommendation ratio.

5. A product push method according to claim 1, characterized in that: The recommendation is made according to the ratio of the current marker fitness index to the sum of all marker label fitness indexes.

6. A product push system, configured to implement the product push method according to any one of claims 1 to 5, characterized in that: It includes recognition and acquisition module, adaptation label module, retrieval and acquisition module, recommendation overload module and tag recommendation module; Signal connection between modules; The recognition and collection module is used to identify the user's product list and collect user behavior data, obtain the classification label load and the user's favorite label coefficient, and send it to the adaptation label module; The adaptation tag module is used to calculate the adaptation index of each category tag and the current user based on the category tag loading amount and the user's favorite tag coefficient using the logistic regression algorithm, and send it to the tag recommendation module; The retrieval and collection module is used to retrieve the search share and recommendation ratio of different classification tags in the entire platform and send them to the recommendation overload module; The recommendation overload module is used to divide each category tag into recommendation overload tags and regular recommendation tags based on the search share and recommendation ratio of different category tags in the entire platform using a weighted algorithm, and send them to the tag recommendation module; The tag recommendation module is used to filter out common recommendation tags, sort the tags according to the compatibility index between each category tag and the current user, select and sort the tags according to the preset selected threshold, and then recommend products to the user.

Images