Community Internet Mall Information Recommendation Method Based on Big Data of User Behavior

Abstract

The present invention relates to the technical field of commodity information recommendation, and particularly relates to a method for recommending community Internet mall information based on user behavior big data, including: obtaining the behavior data of each user, dividing the stable time period and the changing time period according to the change trend of the number of commodity orders over time, and screening out the address information of the types to be analyzed; obtaining the recommendation ability index according to the similarity of the behavior data of the user in the stable time period and the user in the changing time period, in combination with the commodity browsing information; obtaining the root node possibility according to the time corresponding to the commodity purchase information and the commodity browsing information of the user in the changing time period, as well as the commodity repurchase data and the recommendation ability index; screening out the users in the stable time period as the root nodes, constructing a recommendation chain tree structure in combination with the behavior data of the users in the changing time period, and determining the mall information recommendation scheme for different users. The present invention provides a more accurate, more personalized and more flexible method for recommending community Internet mall information.

Description

Technical Field

[0001] The present invention relates to the technical field of commodity information recommendation, and particularly relates to a method for recommending community Internet mall information based on user behavior big data. Background Art

[0002] With the rapid development of Internet technology and the popularization of e-commerce, community Internet malls have become an important channel for people's daily shopping. The behavioral data of users on the Internet, including browsing records, purchase histories, evaluation feedback, etc., provides a basis for accurate commodity information recommendation. The method of recommendation based on user behavior big data can analyze users' interests and needs, provide personalized commodity and service recommendations, thereby enhancing users' shopping experience and the sales efficiency of the mall.

[0003] The core technology of community Internet malls is to integrate Internet technology to build an online shopping platform, and at the same time combine offline store operation management technology to realize functions such as commodity display, order placement, distribution, and offline experience and self-pickup; the characteristics are being close to the lives of community residents, providing a convenient shopping method, the types of commodities usually focusing on the daily needs of residents, combining the convenience of online shopping and the intuitive experience of offline shopping, and being able to quickly respond to the consumption demands within the community.

[0004] Currently, the method for recommending community Internet mall information is to perform data analysis on the big data of the purchase behavior of a single user to construct a personalized recommendation plan. However, this method ignores the recommendation behavior between user groups and the diversity of user group purchase behaviors, and cannot divide users into different types of groups, resulting in insufficient personalization of mall information recommendation. Summary of the Invention

[0005] In order to solve the technical problem that the existing method ignores the recommendation behavior between user groups and the diversity of user group purchase behaviors, resulting in insufficient personalization of mall information recommendation, the purpose of the present invention is to provide a method for recommending community Internet mall information based on user behavior big data, and the specific technical solution adopted is as follows:

[0006] Obtain the behavioral data of each user in different dimensions under different types of address information; the behavioral data includes the number of commodity orders, commodity purchase information, and commodity browsing information;

[0007] According to the change trend of the number of commodity orders over time under each type of address information, divide the stable time period and the changing time period; according to the difference in the number of commodity orders between the stable time period and the changing time period under each type of address information, screen out the address information of the type to be analyzed;

[0008] Under the address information of each type to be analyzed, according to the similarity of the behavior data of each user in the stable time period and each user in the changing time period, combining the time information and the product browsing information of the user, the recommendation ability index of each user under each type to be analyzed in the stable time period is obtained;

[0009] According to the time corresponding to the product purchase information and product browsing information of each user in the changing time period and the product repurchase data, combining the recommendation ability index, the root node possibility of each user under each type to be analyzed in the stable time period is obtained;

[0010] According to the root node possibility, each user under each type to be analyzed in the stable time period is screened as a root node, combined with the behavior data of each user under the same type to be analyzed in the changing time period, a recommendation chain tree structure is constructed, and based on the recommendation chain tree structure, a mall information recommendation scheme for different users is determined.

[0011] Preferably, the product purchase information includes the purchase time of the purchased product for each purchase behavior; the product browsing information includes the browsing duration of the product for each purchase behavior and the time interval from the end of browsing to the purchase time for each purchase behavior, which is called the browsing decision-making duration for each purchase behavior; the behavior data also includes the number of product repurchases.

[0012] Preferably, the method for obtaining the recommendation ability index of each user under each type to be analyzed in the stable time period specifically includes:

[0013] Under the address information of any type to be analyzed, all users with the behavior data in the stable time period are regarded as users to be screened, and all users with the behavior data in the changing time period are regarded as users to be recommended; any user to be screened is denoted as the selected user to be screened, and any user to be recommended is regarded as the selected user to be recommended;

[0014] According to the purchase time of the purchased product, the browsing duration of the product for each purchase behavior of the selected user to be screened and the selected user to be recommended, and the similarity index between the selected user to be screened and the selected user to be recommended, the delayed influence degree of each purchase behavior of the selected user to be screened relative to the selected user to be recommended is obtained;

[0015] Calculate the similarity between the behavior data of each purchase behavior of the selected user to be screened in the stable time period and the behavior data of each purchase behavior of the selected user to be recommended in the changing time period, and obtain the corresponding similarity feature factor of the selected user to be recommended for each purchase behavior;

[0016] Taking the degree of delayed influence as a weight, the weighted average of the similar feature factors corresponding to the selected recommended users for each purchase behavior is calculated to obtain the product recommendation power of the selected user to be screened relative to the selected recommended users. The average value of the product recommendation power of the selected user to be screened relative to all recommended users is used as the recommendation ability index of the selected user to be screened.

[0017] Preferably, the method for obtaining the degree of delayed influence of the selected user to be screened relative to each purchase behavior of the selected recommended user specifically includes:

[0018] Denote the purchase behavior at any order as the target purchase behavior;

[0019] Based on the cosine similarity between the behavior data of the selected user to be screened in each dimension and the behavior data of the selected recommended user in each dimension, determine the similarity degree between the selected user to be screened and the selected recommended user;

[0020] Take the ratio between the purchase time of the purchased products of the selected recommended user and the selected user to be screened under the target purchase behavior as the first characteristic coefficient; take the ratio between the similarity degree and the browsing time of the browsed products of the selected recommended user under the target purchase behavior as the second characteristic coefficient;

[0021] Calculate the product of the first characteristic coefficient and the second characteristic coefficient to obtain the degree of delayed influence of the selected user to be screened relative to the selected recommended user under the target purchase behavior.

[0022] Preferably, the method for obtaining the root node possibility of each user under each type to be analyzed within a stable time period according to the time corresponding to the product purchase information and product browsing information of each user within a changing time period and the product repurchase data, in combination with the recommendation ability index, specifically includes:

[0023] Obtain the decision-making ability index of the selected user to be screened according to the browsing time of the browsed products, the browsing decision-making time, and the number of product repurchases for each purchase behavior of the selected user to be screened;

[0024] Take the normalized value of the product of the recommendation ability index and the decision-making ability index of the selected user to be screened as the root node possibility of the selected user to be screened.

[0025] Preferably, the method for obtaining the decision-making ability index of the selected user to be screened according to the browsing time of the browsed products, the browsing decision-making time, and the number of product repurchases for each purchase behavior of the selected user to be screened specifically includes:

[0026] Calculate the characteristic ratio of the browsing duration and browsing decision-making duration of each purchase behavior of the selected users to be screened, and take the product of the mean value of the characteristic ratios corresponding to all purchase behaviors of the selected users to be screened within the stable time period and the number of repeated purchases of the products by the selected users to be screened within the stable time period as the decision-making power index of the selected users to be screened.

[0027] Preferably, screening each user under each type to be analyzed within the stable time period as a root node according to the root node possibility, and combining the behavior data of each user under the same type to be analyzed in the changing time period to construct a recommended chain tree structure, specifically including:

[0028] Respectively use the users to be screened with a root node possibility greater than the preset possible threshold as the root nodes of each tree in the recommended chain tree structure;

[0029] Obtain the decision-making power index of each recommended user, and take the product of the corresponding delay influence degree between each recommended user and the user to be screened corresponding to the root node and the decision-making power index of the recommended user as the obedience index of each recommended user relative to the root node;

[0030] For any root node, construct a tree structure in the order of the obedience index of each recommended user relative to the root node from large to small, and the obedience indexes of the recommended users corresponding to the nodes in the same layer of the tree structure are the same;

[0031] The tree structures of all root nodes constitute the recommended chain tree structure.

[0032] Preferably, determining the mall information recommendation scheme for different users based on the recommended chain tree structure specifically includes:

[0033] In the recommended chain tree structure, take the layer where the first common child node of different root nodes is located as the first-stage target layer; after the first-stage target layer in the recommended chain tree structure, take the layer with the largest number of child nodes as the second-stage target layer;

[0034] Use the first type of preset recommendation scheme for the users between the root node and the first-stage target layer, use the second type of preset recommendation scheme for the users between the first-stage target layer and the second-stage target layer, and use the third type of preset recommendation scheme for the users between the third-stage target layer and the bottom layer.

[0035] Preferably, dividing the stable time period and the changing time period according to the change trend of the number of product orders over time under each address information specifically includes:

[0036] For any address information, obtain the order quantity sequence composed of the number of product orders at each moment, calculate the first-order difference value of the order quantity sequence, and screen out the moments with a positive first-order difference value and record them as the increasing trend moments;

[0037] Arrange the growth trend moments in ascending order according to the corresponding first-order difference values to obtain a difference sequence, calculate the second-order difference values of the difference sequence, and take the growth trend moment corresponding to the maximum value of the second-order difference values as the characteristic moment. The time period before the characteristic moment is the stable time period, and the time period after the characteristic moment is the changing time period.

[0038] Preferably, screening the address information of the type to be analyzed according to the difference in the number of commodity orders in the stable time period and the changing time period for each type of address information specifically includes:

[0039] For any type of address information, obtain the first mean value of the number of commodity orders in the stable time period and the second mean value of the number of commodity orders in the changing time period; perform normalization processing on the difference between the second mean value and the first mean value to obtain an order difference coefficient; if the order difference coefficient is greater than a preset difference threshold, then take this type of address information as the address information of the type to be analyzed.

[0040] The embodiments of the present invention have at least the following beneficial effects:

[0041] The present invention first obtains the number of commodity orders within different location information ranges, providing a data basis for subsequent data analysis. Then, by the changing trend of commodity order data over time, divide the time period and analyze the data difference situation to determine the possibility of the existence of a commodity recommendation chain, and screen out the address information of the type to be analyzed, which is also to screen out the range of location information where a recommendation chain may exist. Secondly, conduct an in-depth analysis of the user distribution range where commodity recommendations exist. By the similarity of the behavior data of users in the stable time period and each user in the changing time period, combined with time information, the user's commodity browsing information, and commodity repurchase data, analyze the recommendation power and decision-making power of users to determine the possibility of the root node of each user. Further, screen out the root nodes in the tree structure of the commodity recommendation chain, construct the recommendation tree result, enabling the adaptive formulation of personalized recommendation strategies. The present invention fully combines user behavior data, recommendation chains, and user regional ranges, providing a more accurate, personalized, and flexible community Internet mall information recommendation method, not only improving the recommendation accuracy and user experience, but also bringing higher conversion rates and user satisfaction to the mall. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0043] Figure 1 It is a flowchart of the steps of a method for recommending community Internet mall information based on user behavior big data provided by the present invention;

[0044] Figure 2 It is a schematic diagram of the curve of the order quantity sequence provided by the present invention;

[0045] Figure 3 It is a flowchart of the steps of a method for obtaining recommendation ability indicators provided by the present invention;

[0046] Figure 4 It is a partial schematic diagram of the recommendation chain tree structure provided by the present invention. Detailed implementation manners

[0047] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following combines the accompanying drawings and preferred embodiments to detail the specific implementation manners, structures, features and effects of a method for recommending community Internet mall information based on user behavior big data proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0048] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the technical field to which the present invention belongs.

[0049] The following specifically describes the specific solution of a method for recommending community Internet mall information based on user behavior big data provided by the present invention with reference to the accompanying drawings.

[0050] Please refer to Figure 1 , which shows a flowchart of the steps of a method for recommending community Internet mall information based on user behavior big data provided by an embodiment of the present invention. The method includes the following steps:

[0051] Step S100, obtaining the behavior data of each user in different dimensions under different types of address information; the behavior data includes the number of commodity orders, commodity purchase information, and commodity browsing information.

[0052] In this embodiment, considering that there may be interaction phenomena and mutual recommendation phenomena in the purchase behaviors among users in adjacent geographical locations, the recommendation behaviors of users are analyzed by different types of the address information corresponding to the users. Based on this, in order to provide a data basis for analyzing the performance of the recommended purchase commodity behaviors among users under each type of address information subsequently, first, the behavior data of each user in different dimensions under different types of address information can be obtained.

[0053] It can be understood that there are more users under one type of address information, and one dimension of behavioral data corresponds to one type of data information. Among them, relevant information of users in the community platform is obtained through the background records of the community Internet mall, that is, including the address information of users; in this embodiment, the address information of users belonging to the same community is regarded as the same type of address information.

[0054] Furthermore, in this embodiment, the behavioral data of users may include the number of product orders, product purchase information, product browsing information, and the number of product repurchases. Among them, the product purchase information includes the time of purchasing products for each purchase behavior, the product browsing information includes the browsing duration of products for each purchase behavior and the time interval from the end of browsing to the purchase time for each purchase behavior, and is called the browsing decision-making duration for each purchase behavior. The behavioral data of users also includes the purchased products and product types.

[0055] More specifically, detailed information about users' shopping is obtained from website or application logs, including access time, click events, scrolling, and browsing stay time, etc.; order details of users' final purchases and time information when purchase behaviors occur are obtained through transaction and browsing records. The behavioral data of corresponding users can be obtained through the collected data.

[0056] Step S200, divide the stable time period and the changing time period according to the changing trend of the number of product orders over time under each type of address information; screen out the address information of the type to be analyzed according to the difference in the number of product orders between the stable time period and the changing time period under each type of address information.

[0057] Within a relatively close range of address information, there are mutual recommendation behaviors among different users, which are manifested as a growing trend in the change of the number of orders at a certain time node in terms of product purchase information. By analyzing the degree of change in the number of products purchased by users, the range of address information with recommendation situations can be initially screened out for further analysis of the recommendation chain representation of recommendation behaviors. Based on this, through the analysis results of the changing trends of the number of product orders in two aspects, the address information of the type to be analyzed that needs to be further analyzed for the recommendation chain is screened out.

[0058] The first step is to analyze the changing trend of the number of product orders over time under each type of address information, segment the time period of data collection, and obtain the stable time period and the changing time period.

[0059] It can be understood that the behavior data collected in this embodiment has temporal characteristics, that is, the distribution of behavior data characteristics in multiple aspects of users in the historical database, and the recommendation behavior between users is analyzed to provide a better personalized information recommendation solution. Based on this, in this step, any type of address information is taken as an example for illustration, and the number of commodity orders corresponding to each moment can be obtained to form a time series, that is, the order quantity series. Among them, the length of the time period for collecting data, that is, the number of moments, can be set by the implementer according to the specific implementation scenario, such as Figure 2 , which shows a schematic curve of an order quantity series. In the figure, O is the characteristic moment.

[0060] Specifically, calculate the first-order difference value of the order quantity series, and screen out the moments with positive first-order difference values and record them as growth moments; arrange the growth moments in ascending order according to the corresponding first-order difference values to obtain a difference series, calculate the second-order difference value of the difference series, and take the growth moment corresponding to the maximum value of the second-order difference value as the characteristic moment. The time period before the characteristic moment is the stable time period, and the time period after the characteristic moment is the changing time period.

[0061] Among them, calculate the difference between the number of commodity orders at each moment in the order quantity series and the number of commodity orders at the previous adjacent moment as the first-order difference value. When the value of the first-order difference value is positive, it indicates that the number of commodity orders at the corresponding moment shows an increasing trend. When the value of the first-order difference value is negative, it indicates that the number of commodity orders at the corresponding moment shows a decreasing trend. When there is a mutual recommendation purchase behavior between users, it should be shown in the data as an increasing trend. Therefore, screen out the moments with a first-order difference value greater than 0 for further analysis of the growth degree.

[0062] Furthermore, calculate the difference between each first-order difference value in the difference series and the previous adjacent first-order difference value as the second-order difference value. The second-order difference value reflects the magnitude of the increase in the number of commodity orders. The larger its value, the greater the growth trend. Therefore, screen out the moment corresponding to the maximum value of the second-order difference value, that is, the characteristic moment. The characteristic moment indicates that the change in the number of commodity orders was relatively stable before this moment, which may be due to the mutual recommendation purchase behavior between users. At the characteristic moment, a large increase in the number of commodity orders occurred.

[0063] Second step, by analyzing the difference in the number of commodities between the stable time period where stable data distribution may exist and the changing time period where increasing trend change distribution may exist, determine whether there is a recommendation interaction behavior between different users under the current type of address information.

[0064] Specifically, obtain the first mean value of the number of commodity orders within a stable time period and the second mean value of the number of commodity orders within a changing time period; perform normalization processing on the difference between the second mean value and the first mean value to obtain an order difference coefficient, which can be expressed by the formula: , where represents the order difference coefficient, represents the mean value of the number of all commodity orders within the changing time period, that is, the second mean value; represents the mean value of the number of all commodity orders within the stable time period, that is, the first mean value; r represents the changing time period, and l represents the stable time period, is a normalization function.

[0065] The larger the value of

[0066] , the greater the data difference of the time segmentation based on the characteristic moment, indicating that the increment of the number of commodity orders is greater. The larger the value of the corresponding order difference coefficient, the more likely there is a mutual recommendation purchase behavior among users within the current address information range.

[0067] Step S300, under each type of address information to be analyzed, according to the similarity of the behavior data of each user within the stable time period and each user within the changing time period, combined with the time information and the commodity browsing information of the user, obtain the recommendation ability index of each user under each type of address information to be analyzed within the stable time period.

[0068] Within the range of the same type of address information, the commodity recommendation chain formed by the recommendation behavior among users can be represented in the form of a multi-way tree. In the tree structure of the multi-way tree, the root node can be regarded as the source user in the commodity recommendation chain, that is, the user who first generates a purchase behavior and may generate a recommendation behavior to other users. During the commodity purchase process, different user groups may have independent recommendation starting points, which means that there may be multiple users who can be used as the root nodes in the tree structure of the commodity recommendation chain. Therefore, it is necessary to screen whether a user can be used as a root node through the analysis results of the two aspects of whether the user who has made a purchase behavior earlier has a certain recommendation ability and a high decision-making ability.

[0069] In this embodiment, take any type of address information to be analyzed as an example for illustration. Such as Figure 3As shown in the figure, the method for obtaining the recommendation ability index can be implemented by steps S301 to S304.

[0070] Step S301: All users with the behavior data during the stable time period are regarded as users to be screened, and all users with the behavior data during the changing time period are regarded as recommended users.

[0071] Specifically, users with purchase behavior during the stable time period are called users to be screened, and users with purchase behavior during the changing time period are called recommended users. The users to be screened represent users with earlier purchase behavior, that is, these users may have recommendation behavior and need to further analyze whether they have recommendation power and decision-making power. The recommended users represent users with later purchase behavior, that is, the purchase behavior of these users may be due to the recommendation behavior of the users to be screened.

[0072] In this embodiment, taking any user to be screened and any recommended user as examples for feature analysis, that is, any user to be screened is denoted as the selected user to be screened, and any recommended user is taken as the selected recommended user.

[0073] Step S302: According to the purchase time of the purchased goods, the browsing duration of the goods, and the similarity index between the selected user to be screened and the selected recommended user for each purchase behavior, obtain the degree of delayed influence of the selected user to be screened on the selected recommended user for each purchase behavior.

[0074] If the user to be screened has recommended a product, then after the time node of the purchase behavior of the user to be screened, there is a purchase behavior of the recommended user, which can effectively quantify the time delay information between the purchase behaviors of the two types of users. In this embodiment, the purchase behavior in any order is denoted as the target purchase behavior. It can be understood that the order refers to the order of all purchase behaviors of the user. For example, the order of the first purchase behavior of a certain user is 1, and the i-th purchase behavior is taken as the target purchase behavior in this embodiment.

[0075] Specifically, based on the cosine similarity between the behavior data of the selected user to be screened in each dimension and the behavior data of the selected recommended user in each dimension, determine the similarity degree between the selected user to be screened and the selected recommended user; take the ratio between the purchase time of the purchased goods of the selected recommended user and the selected user to be screened under the target purchase behavior as the first feature coefficient; take the ratio between the similarity degree and the browsing duration of the purchased goods of the selected recommended user under the target purchase behavior as the second feature coefficient; calculate the product of the first feature coefficient and the second feature coefficient to obtain the degree of delayed influence of the selected user to be screened on the selected recommended user under the target purchase behavior.

[0076] As a specific example, if the m-th user to be screened is taken as the selected user to be screened and the n-th recommended user is taken as the selected recommended user, then the calculation formula for the delay influence degree of the selected user to be screened relative to the selected recommended user under the target purchase behavior can be expressed as:

[0077]

[0078] Among them, represents the delay influence degree of the selected user to be screened relative to the selected recommended user under the target purchase behavior, m represents the m-th user to be screened, n represents the n-th recommended user, and i represents the i-th purchase behavior; represents the purchase time of the selected user to be screened for the purchased commodity under the target purchase behavior, represents the purchase time of the selected recommended user for the purchased commodity under the target purchase behavior; represents the browsing time of the selected recommended user for the commodity under the target purchase behavior, represents the similarity degree between the selected user to be screened and the selected recommended user.

[0079] It should be noted that the similarity degree is obtained by calculating the cosine similarity between the data vector composed of the behavior data of the selected user to be screened in all dimensions and the data vector composed of the behavior data of the selected recommended user in all dimensions. It reflects the similarity situation between all purchase behavior information of the selected user to be screened and the selected recommended user.

[0080] The first characteristic coefficient The larger the value of, it indicates that after the recommendation behavior occurs, the purchase behavior occurs after the purchase behavior of the user to be screened. At the same time, the greater the similarity of purchase information between the recommended user and the user who may have the recommendation behavior, and the shorter the commodity browsing time of the recommended user, indicating a higher trust in the recommendation behavior, that is, the second characteristic coefficient The larger the value of, the greater the delay of the corresponding user's purchase behavior, that is, the greater the delay of the recommended user under the influence of the purchase behavior of the user to be screened, that is, the larger the value of the delay influence degree.

[0081] Step S303: Calculate the similarity between the behavior data of each purchase behavior of the selected user to be screened in the stable time period and the behavior data of each purchase behavior of the selected recommended user in the changing time period, and obtain the corresponding similarity characteristic factor of the selected recommended user for each purchase behavior.

[0082] Considering that some of the time nodes of the purchase behaviors of the users to be screened may be within the changing time period, and some of the time nodes of the purchase behaviors of the recommended users may also be within the stable time period. Based on this, in order to more accurately quantify the recommendation possibility degree of the users to be screened for the commodities, on the basis of the delayed influence degree of the purchase behaviors, the relevance between the behavior information of the users to be screened within the corresponding stable time period and the behavior information of the recommended users within the corresponding changing time period is combined to comprehensively obtain the commodity recommendation power of the final users to be screened.

[0083] Specifically, in this embodiment, the cosine similarity between the vector composed of the behavior data of each purchase behavior of the selected user to be screened in all dimensions within the stable time period and the vector composed of the behavior data of the corresponding purchase behavior of the same order of the selected recommended user in all dimensions within the changing time period is used as the similarity feature factor under the same order of purchase behavior.

[0084] Step S304, using the delayed influence degree as the weight, perform weighted averaging on the similarity feature factors corresponding to the selected recommended user for each purchase behavior to obtain the commodity recommendation power of the selected user to be screened relative to the selected recommended user, and use the average value of the commodity recommendation powers of the selected user to be screened relative to all recommended users as the recommendation ability index of the selected user to be screened.

[0085] Users who have purchase behaviors in the changing time period after the commodity order quantity shows an increasing trend are all likely to be recommended users. The more similar the behavior data between the users who have purchase behaviors in the stable changing time period before the commodity order quantity shows an increasing trend and the recommended users, and the more delayed the purchase time and purchase behavior of the recommended users are, the greater the possibility that the users to be screened have recommendation behaviors. And this phenomenon appears in multiple purchase behaviors, which further indicates that the higher the recommendation power of the users with recommendation behaviors.

[0086] As a specific example, the calculation formula of the recommendation ability index of the selected user to be screened can be expressed as:

[0087]

[0088] Wherein, represents the recommendation ability index of the selected user to be screened, m represents the m-th user to be screened, n represents the n-th recommended user, i represents the i-th purchase behavior, represents the delayed influence degree of the selected user to be screened relative to the selected recommended user for the i-th purchase behavior, represents the similarity feature factor of the selected user to be screened relative to the selected recommended user for the i-th purchase behavior, represents the minimum value of the total number of purchase behaviors, Indicates the total number of recommended users.

[0089] Indicates the result of weighted average. The larger its value, the more similar the purchase behaviors are and the more obvious the delay in purchase behaviors in multiple purchase behaviors. The larger the value of the recommended ability index of the selected user to be screened, the more likely it is that there are certain recommendation behaviors for the selected user to be screened, and more recommended users generate purchase behaviors. The higher the recommendation power of this user, the better the effect of constructing the product recommendation chain as the root node.

[0090] Step S400: According to the time corresponding to the product purchase information and product browsing information of each user within the changing time period and the product repurchase data, and in combination with the recommended ability index, obtain the possibility of each user as the root node under each type to be analyzed within the stable time period.

[0091] For users with recommendation power, their purchase behaviors have relatively high decision-making power, that is, such users have relatively high action power and decision-making power for purchasing products. Therefore, in this embodiment, when screening the root node, the recommendation power and decision-making power performance of the user to be screened are further combined to quantify the possibility of the user to be screened as the root node.

[0092] In this embodiment, taking the location information of any type to be analyzed as an example, in the first step, the decision-making power index of the selected user to be screened is obtained according to the browsing product duration, browsing decision duration of each purchase behavior of the selected user to be screened, and the product repurchase times.

[0093] Specifically, calculate the characteristic ratio of the browsing product duration and browsing decision duration of each purchase behavior of the selected user to be screened, and take the product of the mean value of the characteristic ratios corresponding to all purchase behaviors of the selected user to be screened within the stable time period and the product repurchase times of the selected user to be screened within the stable time period as the decision-making power index of the selected user to be screened. In this embodiment, the decision-making power index can be expressed by the formula: , where represents the decision-making power index of the selected user to be screened, m represents the mth user to be screened, represents the total number of purchase behaviors of the selected user to be screened within the stable time period, represents the browsing product duration of the selected user to be screened in the i-th purchase behavior, represents the browsing decision duration of the selected user to be screened in the i-th purchase behavior, represents the product repurchase times of the selected user to be screened within the stable time period.

[0094] It is a characteristic ratio that reflects the length of time considered by the user to be screened under the current purchase behavior. The smaller its value, the higher the action and decision-making power of the user. Corresponding to the repeat purchase behavior that occurs multiple times within a stable time period, it indicates that the user to be screened has a relatively high degree of trust in certain products. Furthermore, the larger the value of the decision-making power indicator, the higher the decision-making power of the user to be screened.

[0095] In the second step, comprehensively obtain the likelihood of the user to be screened as the root node based on the recommendation power and decision-making power of the user to be screened. Specifically, use the normalized value of the product of the recommended ability indicator and the decision-making power indicator of the selected user to be screened as the likelihood of the selected user to be screened as the root node. The likelihood of the root node characterizes the likelihood and the quality of the effect of the corresponding user to be screened as the root node in the tree structure of the multi-fork tree of the product recommendation chain. The larger its value, the greater the likelihood of the user to be screened as the root node in the tree structure of the multi-fork tree of the product recommendation chain, and the better the effect.

[0096] Step S500: According to the likelihood of the root node, screen each user in each type to be analyzed within a stable time period as the root node, and combine the behavior data of each user in the same type to be analyzed during the changing time period to construct a recommendation chain tree structure, and determine the mall information recommendation scheme for different users based on the recommendation chain tree structure.

[0097] First, according to the likelihood of the root node of each user to be screened, screen all users to be screened by setting a threshold to screen out users with high recommendation power and decision-making power who have recommendation behavior as the source users in the product recommendation chain, which can effectively construct a relatively complete recommendation chain structure.

[0098] Specifically, use the users to be screened whose likelihood of the root node is greater than the preset possible threshold as the root nodes of each tree in the recommendation chain tree structure. In this embodiment, the value of the possible threshold is 0.7, and the implementer can set it according to the specific implementation scenario.

[0099] Then, there is a corresponding delayed influence degree between each recommended user and each user to be screened, which characterizes that there is a high degree of similar purchase behavior between the recommended user and the user to be screened, and the purchase behavior of the recommended user has a certain delay. Furthermore, combined with the decision-making power performance of the recommended user itself, it can reflect the compliance performance of the recommended user's recommendation behavior relative to a certain user to be screened.

[0100] Specifically, obtain the decision-making power index of each recommended user, and take the product of the corresponding delay influence degree between each recommended user and the user to be screened corresponding to the root node and the decision-making power index of the recommended user as the compliance index of each recommended user relative to the root node. The larger the value of the compliance index, the higher the decision-making power and action power of the recommended user's purchase behavior during the change period, and the greater the influence of the user to be screened on the recommended user before the recommended user, which further indicates that the recommended user has a greater compliance with the user to be screened.

[0101] It should be noted that according to the method for obtaining the decision-making power index of the user to be screened, the decision-making power index corresponding to the recommended user during the change period can be obtained. More specifically, calculate the characteristic ratio of the browsing duration of the goods and the browsing decision duration of each purchase behavior of the recommended user, and take the product of the mean value of the characteristic ratios corresponding to all purchase behaviors of the recommended user during the change period and the number of repeated purchases of the goods by the recommended user during the change period as the decision-making power index of the recommended user.

[0102] Furthermore, there is a corresponding compliance index between each recommended user and the user to be screened corresponding to each root node. The larger the compliance index, the greater the possibility that the purchase behavior of the recommended user is affected by the recommendation behavior of the user to be screened. Therefore, when constructing the tree structure, it is more necessary to divide the recommended user into the tree structure where the root node of the corresponding user to be screened is located.

[0103] Specifically, for any root node, construct a tree structure in the order of the compliance index of each recommended user relative to the root node from large to small. The compliance indexes of the recommended users corresponding to the nodes in the same layer of the tree structure are the same; the tree structures of all root nodes form a recommended chain tree structure.

[0104] Finally, hierarchically divide the user group through the recommended chain tree structure, and then personalized recommendation schemes can be adaptively set for different levels and user groups with different purchase behaviors. Specifically, in the recommended chain tree structure, the layer where the first common child node of different root nodes is located is used as the first-stage target layer; after the first-stage target layer in the recommended chain tree structure, the layer with the largest number of child nodes is used as the second-stage target layer.

[0105] As Figure 4 shown, it shows a partial schematic diagram of the recommended tree structure, taking the root node b, the root node c, and the root node d as examples for display. Figure 4The numbers in it refer to the serial numbers of the child nodes corresponding to the recommended users with a high degree of compliance with the corresponding root nodes. Then, the child node labeled 9 in the figure represents the child node with a sharing phenomenon. Therefore, the layer where the child node labeled 9 is located is the first-stage target layer, which characterizes that for the first time, there may be recommended users who are recommended by multiple users. In the figure, the number of child nodes in the layer below the first-stage target layer is the largest. Therefore, it is the second-stage target layer, which characterizes that affected by the user recommendation behavior, there may also be mutual recommendation behaviors among the recommended users.

[0106] Further, a first type of preset recommendation scheme is used for the users between the root node and the first-stage target layer, a second type of preset recommendation scheme is used for the users between the first-stage target layer and the second-stage target layer, and a third type of preset recommendation scheme is used for the users between the third-stage target layer and the bottom layer.

[0107] In Figure 4 In the tree structure taking [example] as an example, all users from the first layer to the second layer use the first type of preset recommendation scheme, and all users from the third layer and the fourth layer use the second type of preset recommendation scheme. In this tree structure, there are no child nodes in other layers except the second-stage target layer. Therefore, there is no need to use the third type of preset recommendation scheme.

[0108] In this embodiment, the first type of preset recommendation scheme is that the community Internet mall can recommend similar products to the products that users often buy or have already bought. In the first type, users are the source of recommendation. It is easier for them to recommend and try products that are relatively similar to the ones they have bought. When new useful products are screened, they may also be recommended to other users. The second type of preset recommendation scheme is that the community Internet mall can recommend new or hot-selling products so that users can feel the freshness of the products and the diversity of the platform. In the second type, users are relatively easy to accept recommendations. Therefore, the diversity of products can be recommended through the official. The third type of preset recommendation scheme is that the community Internet mall can recommend the product information that they have browsed but finally did not buy through recommendation based on user behavior tracing. In the third type, users are relatively less likely to accept recommendation behaviors, and their decision-making power and action ability are also relatively low. Therefore, products that they did not buy can be recommended to such users to increase exposure.

[0109] In summary, the present invention first obtains the number of commodity orders within different location information ranges, and determines the possibility of the existence of a commodity recommendation chain through the change trend; deeply analyzes the user distribution range where commodity recommendations exist; screens out the root nodes of the tree structure in the recommendation chain through the behavior information of users in the time period before the order increment occurs and the behavior information such as purchase, browsing, and time delay of users in the time period after the order increment occurs, and then divides all the remaining users, thereby generating a complete recommendation network, and formulates corresponding recommendation strategies according to the recommendation progress; the present invention fully combines user behavior data, recommendation chains, and user regional ranges, realizing a more accurate, personalized, and flexible information recommendation method for community Internet shopping malls, not only improving the recommendation accuracy and user experience, but also bringing higher conversion rates and user satisfaction to the shopping mall.

[0110] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the various embodiments of the present application, and should all be included within the protection scope of the present application.

Claims

1. A community Internet mall information recommendation method based on user behavior big data, characterized in that: The method comprises the following steps: Obtaining behavioral data of each user in different dimensions under different types of address information; the behavioral data includes the number of product orders, product purchase information, and product browsing information; According to the changing trend of the number of product orders under each address information over time, the stable time period and the changing time period are divided; according to the difference in the number of product orders in the stable time period and the changing time period under each address information, the address information of the type to be analyzed is screened out; Under the address information of each category to be analyzed, according to the similarity of the behavior data of each user in the stable time period and each user in the changing time period, combined with the time information and the user's product browsing information, the recommendation ability index of each user under each category to be analyzed in the stable time period is obtained; According to the time corresponding to the product purchase information and product browsing information of each user in the changing time period and the product repurchase data, combined with the recommendation ability index, the root node possibility of each user under each category to be analyzed in the stable time period is obtained; According to the possibility of the root node, each user under each category to be analyzed in a stable time period is selected as a root node, and the behavior data of each user under the same category to be analyzed in a changing time period is combined to build a recommendation chain tree structure, and the mall information recommendation plan for different users is determined based on the recommendation chain tree structure.

2. According to claim 1, a method for recommending community Internet shopping mall information based on user behavior big data is characterized in that: The product purchase information includes the time of purchase of each purchase behavior; the product browsing information includes the browsing time of each purchase behavior and the time interval from the end of browsing to the purchase time under each purchase behavior, which is called the browsing decision time of each purchase behavior; the behavior data also includes the number of repurchases of products.

3. A method for recommending community Internet shopping mall information based on user behavior big data according to claim 2, characterized in that: The method for obtaining the recommendation capability index of each user under each category to be analyzed within the stable time period specifically includes: Under any address information of the type to be analyzed, all users with the behavior data in the stable time period are regarded as users to be screened, and all users with the behavior data in the changing time period are regarded as recommended users; any user to be screened is recorded as a selected user to be screened, and any recommended user is recorded as a selected recommended user; According to the purchase time and browsing time of each purchase behavior of the selected user to be screened and the selected recommended user, and the similarity index between the selected user to be screened and the selected recommended user, the delayed influence degree of each purchase behavior of the selected user to be screened relative to the selected recommended user is obtained; Calculate the similarity between the behavior data of each purchase behavior of the selected to-be-screened user in the stable time period and the behavior data of each purchase behavior of the selected recommended user in the changing time period, and obtain the similarity feature factor corresponding to the selected recommended user under each purchase behavior; Taking the degree of delayed influence as the weight, the weighted average of the similar feature factors corresponding to the selected recommended user under each purchase behavior is performed to obtain the product recommendation power of the selected user to be screened relative to the selected recommended user, and the average of the product recommendation power of the selected user to be screened relative to all recommended users is taken as the recommendation ability indicator of the selected user to be screened.

4. A method for recommending community Internet shopping mall information based on user behavior big data according to claim 3, characterized in that: The method for obtaining the delayed influence degree of each purchase behavior of the selected to-be-screened user relative to the selected recommended user specifically includes: Record the purchase behavior in any order as the target purchase behavior; Determine the similarity between the selected user to be screened and the selected recommended user based on the cosine similarity between the behavior data of the selected user to be screened in each dimension and the behavior data of the selected recommended user in each dimension; The ratio between the time when the selected recommended user and the selected to-be-screened user purchased the product under the target purchase behavior is used as the first characteristic coefficient; the ratio between the similarity and the browsing time of the selected recommended user under the target purchase behavior is used as the second characteristic coefficient; The product of the first characteristic coefficient and the second characteristic coefficient is calculated to obtain the degree of delayed influence of the selected to-be-screened user relative to the selected recommended user under the target purchase behavior.

5. The method for recommending community Internet shopping mall information based on user behavior big data according to claim 3 is characterized in that: The method of obtaining the root node possibility of each user under each category to be analyzed in a stable time period based on the time corresponding to the product purchase information and product browsing information of each user in the changing time period and the product repurchase data in combination with the recommendation capability index specifically includes: The decision-making power index of the selected user to be screened is obtained according to the browsing time and browsing decision time of each purchase behavior of the selected user to be screened and the number of product repurchases; The normalized value of the product of the recommendation ability index and the decision-making ability index of the selected user to be screened is used as the root node possibility of the selected user to be screened.

6. A method for recommending community Internet shopping mall information based on user behavior big data according to claim 5, characterized in that: The decision-making power index of the selected user to be screened is obtained according to the browsing time and browsing decision time of each purchase behavior of the selected user to be screened and the number of product repurchases, specifically including: Calculate the characteristic ratio of product browsing time and browsing decision time for each purchase behavior of the selected user to be screened, and take the product of the mean of the characteristic ratios corresponding to all purchase behaviors of the selected user to be screened within a stable time period and the number of product repurchases of the selected user to be screened within the stable time period as the decision-making power index of the selected user to be screened.

7. The method for recommending community Internet shopping mall information based on user behavior big data according to claim 5, characterized in that: The step of selecting each user of each category to be analyzed in a stable time period as a root node according to the root node possibility, and combining the behavior data of each user of the same category to be analyzed in a variable time period to construct a recommendation chain tree structure specifically includes: The users to be screened whose root node probability is greater than the preset possibility threshold are used as the root nodes of each tree in the recommendation chain tree structure; Obtain the decision-making power index of each recommended user, and multiply the corresponding delayed influence degree between each recommended user and the to-be-screened user corresponding to the root node by the decision-making power index of the recommended user as the compliance index of each recommended user relative to the root node; For any root node, a tree structure is constructed in descending order of the compliance index of each recommended user relative to the root node. The compliance index of the recommended users corresponding to the nodes at the same level in the tree structure has the same size. The tree structure of all root nodes constitutes the recommendation chain tree structure.

8. A method for recommending community Internet shopping mall information based on user behavior big data according to claim 7, characterized in that: The determining of the mall information recommendation scheme for different users based on the recommendation chain tree structure specifically includes: In the recommendation chain tree structure, the layer where the first common child node of different root nodes is located is used as the first stage target layer; after the first stage target layer in the recommendation chain tree structure, the layer where the largest number of child nodes is located is used as the second stage target layer; The first type of preset recommendation scheme is used for users between the root node and the first-stage target layer, the second type of preset recommendation scheme is used for users between the first-stage target layer and the second-stage target layer, and the third type of preset recommendation scheme is used for users between the third-stage target layer and the bottom layer.

9. The method for recommending community Internet shopping mall information based on user behavior big data according to claim 1, characterized in that: The method of dividing the stable time period and the changing time period according to the changing trend of the number of commodity orders under each address information over time specifically includes: For any address information, obtain the order quantity sequence consisting of the number of commodity orders at each moment, calculate the first-order difference value of the order quantity sequence, and select the moment with a positive first-order difference value as the increasing moment; Arrange the increasing moments in ascending order according to the corresponding first-order difference values ​​to obtain a difference sequence, calculate the second-order difference value of the difference sequence, and take the increasing moment corresponding to the maximum value of the second-order difference value as the characteristic moment. The time period before the characteristic moment is the stable time period, and the time period after the characteristic moment is the changing time period.

10. A method for recommending community Internet shopping mall information based on user behavior big data according to claim 9, characterized in that: The address information to be analyzed is selected based on the difference in the number of commodity orders in the stable time period and the variable time period under each type of address information, specifically including: For any type of address information, obtain the first mean of the number of commodity orders in a stable time period and the second mean of the number of commodity orders in a changing time period; normalize the difference between the second mean and the first mean to obtain the order difference coefficient; if the order difference coefficient is greater than the preset difference threshold, then use this type of address information as the address information of the type to be analyzed.

Images