Product recommendation method and system based on big data analysis
By dividing the time period into local time periods and using user features and tag vectors to correct similarity, the problem of inaccurate recommendations for users at specific time periods is solved, resulting in more accurate product recommendations and improved user experience.
Patent Information
- Application Number
- CN202511053755.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-07-30
- Publication Date
- 2025-11-04
AI Technical Summary
In existing technologies, users' purchasing behavior at specific times is driven by group effects, leading to inaccurate product recommendations. Furthermore, noise data pollutes the results when calculating user similarity, affecting the user experience.
The current time period is divided into local time periods. Clustering is performed based on user characteristics to obtain the purchase similarity within each local time period. The similarity is then corrected using tag vectors and trend probability to obtain similar users and recommended products for the target user.
It improves the accuracy of matching target users with similar users, ensuring that recommended products better match users' real preferences and enhance user experience.
Smart Images
Figure CN120894104A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of product recommendation technology, and specifically to a product recommendation method and system based on big data analysis. Background Technology
[0002] Product recommendation is the process of efficiently uncovering needs by analyzing user behavior preferences in an environment of information overload, and then accurately recommending products or services to target users.
[0003] Existing methods identify similar users of the target user and include products purchased by similar users but not by the target user as the target user's recommended product set. However, in reality, users' purchasing behavior during specific periods (such as promotional periods or trending topics) may be driven by group effects, resulting in purchased products that do not accurately reflect users' preferences. Furthermore, when calculating user similarity, the weighting of purchasing data across all periods is equal, leading to noise data (herd behavior) contaminating the similarity calculation results. Consequently, it becomes impossible to accurately identify similar users of the target user, resulting in inaccurate product recommendations and negatively impacting the user experience. Summary of the Invention
[0004] To address the technical problem of inaccurate product recommendations for target users due to the inability to accurately identify similar users, the present invention aims to provide a product recommendation method and system based on big data analysis. The specific technical solution adopted is as follows:
[0005] In a first aspect, one embodiment of the present invention provides a product recommendation method based on big data analysis, the method comprising the following steps:
[0006] Get the products purchased by the target user and each reference user within the current time period; get the characteristics of each user;
[0007] Divide the current time period into local time periods, and obtain the purchase similarity between the target user and each reference user within each local time period based on the similarity of the products purchased by the target user and each reference user within each local time period;
[0008] Users who purchase products within each local time period are categorized into user categories based on their characteristics. The extent of purchases of each product within each local time period is obtained based on the purchase status of each product in its user category and the difference in the number of users corresponding to each product and other products.
[0009] According to the difference between the extent of each user's purchase of each product in each local time period and the extent of each user's purchase of each product in each other local time period, the trend probability of each user in each local time period is obtained to correct the similarity of the purchase, and the corrected purchase similarity between the target user and each reference user in each local time period is obtained.
[0010] Based on the corrected product purchase similarity, the similar users of the target user are obtained, and the current recommended product of the target user is obtained based on the products purchased by the similar users.
[0011] Further, the method for obtaining the purchase similarity is:
[0012] Based on the type, function and price of each product purchased, a label vector of each product is constructed;
[0013] For any local time period and any reference user, the cosine value of the label vector of each product purchased by the target user in the local time period and each product purchased by the reference user is obtained, and each is used as a similarity analysis value;
[0014] The average value of all similarity analysis values is used as the purchase similarity between the target user and the reference user in the local time period.
[0015] Further, the method for obtaining the extent is:
[0016] According to the occurrence of each product purchased in each local time period in each user category and the proportion of its purchase users in each user category, a first distribution index of each product purchased in each local time period is obtained;
[0017] The difference between the number of users corresponding to each product purchased in each local time period and the average number of users corresponding to other products is positively correlated, and the second distribution index of each product purchased in each local time period is determined;
[0018] The product of the first distribution index and the second distribution index is normalized, and the result is used as the extent of each product purchased in each local time period.
[0019] Further, the method for obtaining the first distribution index is:
[0020] For any local time period and any product purchased in the local time period, the user category in which the product is purchased in the local time period is used as the reference user category of the product in the local time period;
[0021] The ratio of the number of reference user categories to the number of all user categories in the local time period is used as the first characteristic value;
[0022] an accumulation result of the proportion of users in each reference user category who purchased the product as a second feature value;
[0023] a product of the first feature value and the second feature value as a first distribution index of the product purchased in the local time period.
[0024] Further, the method for obtaining the trend probability is:
[0025] For any local time period and any user, an average of the extent of popularity of all products purchased by the user in the local time period as an overall purchase popularity analysis value of the user in the local time period;
[0026] obtaining a difference between the overall purchase popularity analysis value of the user in the local time period and the overall purchase popularity analysis value of the user in each other local time period as a first difference;
[0027] normalizing a sum of all first differences as a trend probability of the user in the local time period.
[0028] Further, the method for obtaining the purchase similarity degree is:
[0029] For any local time period and any reference user, a result of negatively correlating and normalizing a product of the trend probability of the target user and the reference user in the local time period as a similarity correction weight of the target user and the reference user in the local time period;
[0030] multiplying the similarity correction weight of the target user and the reference user in the local time period and the purchase similarity degree as a corrected purchase similarity degree of the target user and the reference user in the local time period.
[0031] Further, the method for obtaining the similar user is:
[0032] For any reference user, arranging the corrected purchase similarity degree of the target user and the reference user according to a time sequence from back to front of the corresponding local time period to obtain a corrected purchase similarity degree sequence;
[0033] sequentially numbering the corrected purchase similarity degrees in the corrected purchase similarity degree sequence from left to right starting from 0;
[0034] For any corrected purchase similarity degree in the corrected purchase similarity degree sequence, normalizing a difference between the length of the corrected purchase similarity degree sequence and the number of the corrected purchase similarity degree as an experience weight of the corrected purchase similarity degree;
[0035] The product of the modified purchase similarity degree and the experience weight is taken as a participation value of the modified purchase similarity degree;
[0036] The result of normalizing the sum of the participation values of all the modified purchase similarity degrees in the modified purchase similarity degree sequence is taken as a fusion similarity degree between the target user and the reference user.
[0037] When the fusion similarity degree is greater than a preset fusion similarity degree threshold, the corresponding reference user is taken as a similar user of the target user.
[0038] Further, the method for obtaining the current recommended product of the target user is as follows:
[0039] The product that is not purchased by the target user but purchased by the similar user in the current time period is taken as the current recommended product of the target user, and the recommended order of the current recommended product is from small to large according to the time length between the last time of the current recommended product and the current time.
[0040] Further, the method for obtaining the user category is as follows:
[0041] For any local time period, the user who purchases the product in the local time period is taken as an analysis user.
[0042] The feature vector of each analysis user is constructed based on the feature of each analysis user, the analysis user is unsupervisedly clustered through the DBSCAN density clustering algorithm according to the feature vector of each analysis user, and the user category in the local time period is obtained.
[0043] In the second aspect, another embodiment of the present application provides a product recommendation system based on big data analysis, which comprises a memory, a processor and a computer program stored in the memory and running on the processor, and when the processor executes the computer program, the steps of any one of the above methods are realized.
[0044] The present application has the following beneficial effects:
[0045] The present application firstly divides the current time period into local time periods, which is conducive to more accurately analyzing the similarity of the target user and each reference user in purchasing products, so that the user similar to the target user in purchasing preferences is more accurately obtained subsequently; then, according to the similarity of the target user and each reference user in purchasing products in each local time period, the purchase similarity degree of the target user and each reference user in each local time period is obtained, which preliminarily reflects the similarity of the target user and each reference user in purchasing products in each local time period; in order to reduce the influence of popular products on the analysis of the similarity of the target user and the reference user in purchasing products, then the users purchasing products in each local time period are divided into user categories through the characteristics of the users, and then the popularity of each product purchased in each local time period is obtained according to the purchase of each product in each local time period in its user category and the difference in the number of users corresponding to each product and other products, which accurately reflects the popularity of each product purchased in each local time period; then, according to the difference in the popularity of the products purchased by each user in each local time period and in each other local time period, the trend probability of each user in each local time period is obtained, which accurately reflects the degree of influence of the trend on the products purchased by each user in each local time period, which is conducive to accurately correcting the purchase similarity degree subsequently, and then the corrected purchase similarity degree of the target user and each reference user in each local time period is accurately obtained, which accurately reflects the similarity of the real purchasing preferences of the target user and each reference user in each local time period; then, based on the corrected product purchase similarity degree, the similar user of the target user is accurately obtained, which is conducive to accurately obtaining the current recommended product of the target user; then, based on the products purchased by the similar user, the current recommended product of the target user is accurately obtained, which effectively improves the experience of the target user. BRIEF DESCRIPTION OF DRAWINGS
[0046] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, and the advantages thereof, a brief introduction will be given to the drawings needed in the embodiments or the prior art description. Obviously, the drawings in the following description are only some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor based on these drawings.
[0047] Figure 1 A schematic flow chart of a product recommendation method based on big data analysis provided by an embodiment of the present application;
[0048] Figure 2 A flow chart of a method for obtaining the extent of popularity provided by an embodiment of the present application;
[0049] Figure 3A product recommendation system structure diagram based on big data analysis provided by an embodiment of the present application;
[0050] Figure 4 A schematic diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION
[0051] In order to further illustrate the technical means and effects taken by the present application to achieve the predetermined purposes, the specific embodiments, structures, features and effects of the product recommendation method and system based on big data analysis according to the present application are described in detail as follows in combination with the drawings and preferred embodiments. 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.
[0052] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which the present application belongs.
[0053] The specific scheme of the product recommendation method and system based on big data analysis provided by the present application is specifically described below in combination with the drawings.
[0054] Embodiment 1:
[0055] The present application proposes a product recommendation method based on big data analysis, please refer to Figure 1 which shows a schematic flowchart of a product recommendation method based on big data analysis provided by an embodiment of the present application, the method comprises the following steps:
[0056] Step S1: obtaining the products purchased by the target user and each reference user in the current time period; obtaining the characteristics of each user.
[0057] Specifically, in order to accurately recommend products to the target user at present and improve the experience of the target user, the present embodiment obtains the purchase records of the target user and each reference user in the current time period, wherein the purchase records include product name, product purchase time, product type, product function and product price and other multi-dimensional information. In order to describe intuitively, the present embodiment obtains the products purchased by the target user and each reference user in the current time period, which contains the name, purchase time, type and function of the product by default. The present embodiment sets the length of the current time period as 1 year, and the number of reference users as 10000. The implementer can set the length of the current time period and the number of reference users according to the actual situation, which is not limited here, but the end time of the current time period must be the current time.
[0058] It should be noted that the target user and the reference user are both users with at least one year of purchase records.
[0059] In order to more accurately obtain similar users with similar purchase product preferences of the target user, so as to accurately obtain the current recommended product of the target user, and then obtain the characteristics of each user, wherein the characteristics of the user include the geographical location, age and gender of the user and the like.
[0060] Step S2: dividing the current time period into local time periods, and obtaining the purchase similarity degree between the target user and each reference user in each local time period according to the similarity of the products purchased by the target user and each reference user in each local time period.
[0061] Specifically, the products purchased by the user generally represent the purchase habits or interests of the user, and therefore, when the products purchased by the target user and a certain reference user are more similar, the reference user is more likely to be a similar user of the target user, and therefore, the products purchased by the reference user but not purchased by the target user are more likely to be the recommended products of the target user. In order to more accurately analyze the similarity of the products purchased by the target user and each reference user, the current time period is divided into multiple local time periods in this embodiment, and the current time period is divided by month in this embodiment. The implementer can set the size of the local time period according to the actual situation, which is not limited here.
[0062] When the products purchased by the target user and a certain reference user in a certain local time period are more similar, it means that the reference user is more likely to be a similar user of the target user in the local time period. Then, the purchase similarity degree between the target user and each reference user in each local time period is obtained according to the similarity of the products purchased by the target user and each reference user in each local time period. The greater the purchase similarity degree, the more likely the corresponding reference user in the corresponding local time period is a similar user of the target user.
[0063] Preferably, in an implementable manner of this embodiment, the method for obtaining the purchase similarity degree is: first, constructing a label vector of each purchased product based on the type, function and price of each product; because the similarity between products generally needs to be reflected in multiple dimensions such as the type, function and price of the product, and then the text type features corresponding to the purchased products are established into a corresponding word library, and the one-hot vector is used to obtain; the numerical type features are directly used as feature values, and then the label vector of each purchased product is obtained. It should be noted that in the process of constructing the label vector, the missing features of the purchased products are supplemented by 0 operation to ensure that the dimensions of the label vectors of each purchased product are consistent; wherein the one-hot vector is a known technology and will not be described in detail.
[0064] Further, for any local time period and any reference user, the cosine value of the label vector of each product purchased by the target user in the local time period and each product purchased by the reference user is obtained as a similarity analysis value; the greater the similarity analysis value is, the more similar the target user and the reference user are in the local time period in terms of the products purchased; and the average of all similarity analysis values is taken as the similarity degree of the target user and the reference user in the local time period.
[0065] Up to now, the similarity degree of the target user and each reference user in each local time period is obtained.
[0066] Step S3: The users who purchase products in each local time period are divided into user categories according to the characteristics of the users, and the popularity of each product purchased in each local time period is obtained according to the purchase of each product in its user category and the difference in the number of users corresponding to each product and other products.
[0067] Specifically, in actual situations, some users may have a herd mentality, that is, some users may purchase products based on the influence of popularity in a certain local time period, and the products are relatively not representative, which may cause the similarity degree of the target user and the reference user obtained in step S2 to be biased. In order to accurately obtain the real similarity degree of the target user and each reference user in each local time period, it is necessary to analyze the popularity of each product purchased in each local time period. The popularity of a product is mainly determined by the users who purchase it, and the more users who purchase a product, the greater the popularity of the product. However, if the product is only popular in a certain specific group and the number of users in the specific group is large, the number of users corresponding to the product cannot accurately reflect the popularity of the product.
[0068] In order to accurately analyze the popularity of each product purchased in each local time period, the users who purchase products in each local time period are unsupervisedly clustered into user categories according to the characteristics of the users in this embodiment. When a certain product purchased in a certain local time period appears in different user categories in the local time period more times, it means that the popularity of the product in the local time period is greater. At the same time, when the number of users corresponding to the product is greater than the number of users corresponding to other products purchased in the local time period, it further means that the popularity of the product in the local time period is greater. Therefore, the popularity of each product purchased in each local time period is obtained according to the purchase of each product in its user category and the difference in the number of users corresponding to each product and other products. The greater the popularity is, the more popular the corresponding product purchased in the corresponding local time period is.
[0069] Preferably, in one implementable manner of the present embodiment, the user category acquisition method is that, for any local time period, users who purchase products in the local time period are all analyzed users; in order to more accurately analyze the purchase situation of each product purchased in the local time period in users, the popularity of each product purchased in the local time period is determined, and then the present embodiment constructs a feature vector of each analyzed user based on the characteristics of each analyzed user, and the characteristics of each analyzed user are accurately represented by the feature vector. Wherein, the method for constructing the feature vector of each analyzed user is that: a keyword library is established for each feature of the analyzed user, and each feature word is converted into a numerical value according to a unique coding method, for example, assuming that the feature label of the user is [geographical location, age, gender]; wherein, the keyword library established for gender is male: 0, female: 1, unknown: 2;
[0070] Then, according to the feature vector of each analyzed user, the analyzed users are unsupervised clustered by the DBSCAN density clustering algorithm to obtain the user categories in the local time period. It should be noted that the minimum sample number in the DBSCAN density clustering algorithm is set to the dimension number of the feature vector of the analyzed user plus 1, and the neighborhood radius is set as follows: the length of the difference between the feature vector of each analyzed user and the feature vector of each other analyzed user is obtained as the reference distance corresponding to each analyzed user, and the minimum sample number of the minimum reference distance corresponding to each analyzed user is taken as the specified distance of each analyzed user; for example, assuming that the minimum sample number is 4, the minimum 4 reference distances corresponding to each analyzed user are taken as the specified distance of each analyzed user; all the specified distances of the analyzed users are arranged in descending order and fitted into a curve, and the specified distances corresponding to the points where the first derivative of the curve is 0 are all taken as special distances, and finally the minimum special distance is taken as the neighborhood radius. Wherein, the DBSCAN density clustering algorithm and the method of fitting the curve are both known technologies, and will not be described in detail. It should be noted that any two user categories can be considered as user groups with obvious characteristic differences.
[0071] At this point, each user category in each local time period is obtained.
[0072] Preferably, in one implementable manner of the present embodiment, the acquisition method of the extent of spread is as follows Figure 2 which shows a flowchart of an acquisition method of the extent of spread provided by the present embodiment, and the method comprises the following steps:
[0073] Step S201: according to the purchase situation of each product purchased in each local time period in each user category thereof and the proportion of the purchase users thereof in each user category, a first distribution index of each product purchased in each local time period is obtained.
[0074] The more the users who purchase a certain product in a certain local time period and in different user categories in the local time period, the more likely the certain product purchased in the local time period is popular. Further, the embodiment obtains a first distribution index of each product purchased in each local time period according to the purchase of each product purchased in each local time period in each user category and the proportion of the users who purchase the product in each user category. The greater the first distribution index, the more likely the corresponding product purchased in the corresponding local time period is popular.
[0075] In an implementable manner of the embodiment, the method for obtaining the first distribution index is as follows: for any local time period and any product purchased in the local time period, the user categories in which the product is purchased in the local time period are all reference user categories of the product in the local time period; the ratio of the number of the reference user categories to the number of all user categories in the local time period is a first feature value; the cumulative result of the proportion of the users who purchase the product in each reference user category is a second feature value; the greater the first feature value and the second feature value, the more popular and the more widely distributed the product in the local time period; and the product of the first feature value and the second feature value is the first distribution index of the product purchased in the local time period.
[0076] Thus, the first distribution index of each product purchased in each local time period is obtained.
[0077] Step S202: positively correlate the difference between the number of users corresponding to each product purchased in each local time period and the average number of users corresponding to other products to determine the second distribution index of each product purchased in each local time period.
[0078] The more the users who purchase a certain product in a certain local time period and in different user categories in the local time period, the more likely the certain product purchased in the local time period is popular. Further, the embodiment obtains a first distribution index of each product purchased in each local time period according to the purchase of each product purchased in each local time period in each user category and the proportion of the users who purchase the product in each user category. The greater the first distribution index, the more likely the corresponding product purchased in the corresponding local time period is popular.
[0079] The calculation formula of the second distribution index is as follows: ; in the formula, is the second distribution index of the a-th product purchased in the i-th local time period; is the number of users corresponding to the a-th product purchased in the i-th local time period; The mean value of the number of users corresponding to all products other than the a-th product purchased in the i-th local time period; exp is an exponential function with a natural constant as the base number.
[0080] It should be noted that the embodiment of the present application performs positive correlation mapping on the first distribution index and the second distribution index, so that it is more obvious that when the first distribution index is greater than the second distribution index, the a-th product purchased in the i-th local time period is more popular and more widely distributed. The first distribution index and the second distribution index are positively correlated, so that it is more obvious that when the first distribution index is greater than the second distribution index, the a-th product purchased in the i-th local time period is more popular and more widely distributed. The first distribution index and the second distribution index are positively correlated, so that it is more obvious that when the first distribution index is greater than the second distribution index, the a-th product purchased in the i-th local time period is more popular and more widely distributed.
[0081] At this point, the second distribution index of each product purchased in each local time period is obtained.
[0082] Step S203: The product of the first distribution index and the second distribution index is normalized, and the result is used as the extent of the a-th product purchased in the i-th local time period.
[0083] The first distribution index and the second distribution index are positively correlated, so that it is more obvious that when the first distribution index is greater than the second distribution index, the a-th product purchased in the i-th local time period is more popular and more widely distributed.
[0084] At this point, the second distribution index of each product purchased in each local time period is obtained.
[0085] Step S4: According to the difference between the extent of the product purchased by each user in each local time period and the extent of the product purchased by each user in other local time periods, the trend probability of each user in each local time period is corrected according to the purchase similarity, and the corrected purchase similarity between the target user and each reference user in each local time period is obtained.
[0086] Specifically, within a certain local time period, the greater the breadth of products purchased by a user, the more influenced the purchases are by trends, indirectly reflecting the lower the prominence of those products. Simultaneously, considering whether users consistently favor popular products, this allows for a more accurate analysis of the similarity between the target user's and reference users' purchases within each local time period. Therefore, this embodiment obtains the trend probability for each user within each local time period based on the difference in the breadth of product purchases by each user across different local time periods. A lower trend probability indicates that the products purchased by the corresponding user within that local time period are more representative of the user's preferences. This indirectly reflects that the lower the trend probability of both the target user and a reference user within a certain local time period, the more accurate the purchase similarity between them, and the less correction is needed. Therefore, this embodiment corrects the purchase similarity using trend probability, obtaining the corrected purchase similarity between the target user and each reference user within each local time period. This more accurately reflects the similarity in product purchases between the target user and each reference user, facilitating a more accurate subsequent determination of whether each reference user is a similar user to the target user.
[0087] Preferably, in one feasible embodiment of this invention, the method for obtaining the trend probability is as follows: For any local time period and any user, the average of the prevalence of all types of products purchased by the user within that local time period is used as the overall purchase trend analysis value for that local time period. The larger the overall purchase trend analysis value, the more the products purchased by the user within that local time period are influenced by trends, and the less it reflects the user's own preferences. To further highlight the degree to which the user's product purchases are influenced by trends within that local time period, the difference between the overall purchase trend analysis values of the user within that local time period and those within each other local time period is obtained, and these differences are all used as the first difference. The larger the first difference, the greater the degree to which the user's product purchases are influenced by trends within that local time period. The sum of all the first differences is then normalized, and the result is used as the trend probability for the user within that local time period. In this embodiment, the sum of all the first differences is normalized using a normalization function.
[0088] At this point, the trend probability for each user within each local time period is obtained.
[0089] Preferably, in one implementable manner of the present embodiment, the acquisition method of the modified purchase similarity degree is that, for any local time period and any reference user, the result of the negative correlation and normalization of the product of the trend probability of the target user and the reference user in the local time period is taken as the similarity modification weight of the target user and the reference user in the local time period; in one specific embodiment of the present application, the inverse of the product is taken as the power of an exponential function with a natural constant as the base, and the output result of the exponential function is the negative correlation and normalized result. Other basic mathematical methods can be selected by those skilled in the art for implementation, which are not limited and elaborated here; the greater the similarity modification weight is, the more accurate the purchase similarity degree of the target user and the reference user in the local time period is, and the smaller the degree of modification is required, and then the product of the similarity modification weight and the purchase similarity degree of the target user and the reference user in the local time period is taken as the modified purchase similarity degree of the target user and the reference user in the local time period.
[0090] Up to now, the modified purchase similarity degree of the target user and each reference user in each local time period is acquired.
[0091] Step S5: acquiring similar users of the target user based on the modified product purchase similarity degree, and acquiring the current recommended product of the target user based on the products purchased by the similar users.
[0092] Specifically, the overall modified purchase similarity degree of the target user and each reference user in the current time period can be acquired through the modified purchase similarity degree of the target user and each reference user in all local time periods, and the similarity of the target user and each reference user in purchasing products can be accurately analyzed, so that the similar users of the target user can be accurately acquired, and then the current recommended product of the target user can be accurately acquired based on the products purchased by the similar users, thereby effectively improving the experience of the target user.
[0093] Preferably, in one implementable manner of the present embodiment, the acquisition method of the similar user is that, for any reference user, the closer the local time period corresponding to the modified purchase similarity degree of the target user and the reference user is to the current time, the more reference value the corresponding modified purchase similarity degree has, and then the modified purchase similarity degrees of the target user and the reference user are arranged according to the time order from back to front of the corresponding local time period to acquire a modified purchase similarity degree sequence; it should be noted that the modified purchase similarity degree corresponding to the local time period in which the reference user does not purchase products is supplemented by 0. The modified purchase similarity degrees in the modified purchase similarity degree sequence are numbered in sequence from left to right starting from 0, that is, the numbers corresponding to the modified purchase similarity degrees in the modified purchase similarity degree sequence are 0, 1, 2, 3, … in sequence.
[0094] The smaller the number is, the more reference significance the modified purchase similarity degree has. Then, for any modified purchase similarity degree in the modified purchase similarity degree sequence, the embodiment takes the result of normalizing the difference between the length of the modified purchase similarity degree sequence and the number of the modified purchase similarity degree as the experience weight of the modified purchase similarity degree. The greater the experience weight is, the greater the degree of participation of the modified purchase similarity degree in the analysis of the similarity of the products purchased by the target user and the reference user should be. The embodiment normalizes the difference between the length of the modified purchase similarity degree sequence and the number of the modified purchase similarity degree by using the norm normalization function. In order to accurately represent the participation part of the modified purchase similarity degree, the embodiment takes the product of the modified purchase similarity degree and the experience weight as the participation value of the modified purchase similarity degree. Finally, the embodiment takes the result of normalizing the sum of the participation values of all the modified purchase similarity degrees in the modified purchase similarity degree sequence as the fusion similarity degree of the target user and the reference user. The embodiment normalizes the sum of the participation values of all the modified purchase similarity degrees in the modified purchase similarity degree sequence by using the norm normalization function.
[0095] The greater the fusion similarity degree is, the more likely the corresponding reference user is a similar user of the target user. Then, the embodiment sets the preset fusion similarity degree threshold value to 0.9. The implementer can set the size of the preset fusion similarity degree threshold value according to the actual situation, which is not limited here. When the fusion similarity degree is greater than the preset fusion similarity degree threshold value, the corresponding reference user is taken as a similar user of the target user.
[0096] Preferably, in an implementable manner of the embodiment, the method for obtaining the current recommended product of the target user is as follows: taking the product that is not purchased by the target user but purchased by the similar user in the current time period as the current recommended product of the target user; and the recommendation order of the current recommended product is in the order from small to large according to the time length between the last occurrence time of the current recommended product and the current time, because the closer to the current time the recommended product appears, the more reference significance the recommended product has for the target user.
[0097] To sum up, the embodiment acquires products purchased by a target user and reference users, acquires a purchase similarity degree of the target user and the reference users according to a similar situation of the products purchased by the target user and the reference users, acquires a popular degree of each product according to a distribution situation of the products purchased in a local time period in the users, corrects the purchase similarity degree according to a difference of the popular degree of the products purchased by each user in each local time period and other local time periods, acquires a corrected purchase similarity degree and further acquires similar users of the target user, and acquires a current recommended product of the target user based on products purchased by the similar users. The application effectively improves the accuracy of acquiring the similar users by correcting the purchase similarity degree, and further accurately acquires the current recommended product of the target user, thereby ensuring the experience of the target user in real time.
[0098] Embodiment 2
[0099] The application further provides a product recommendation system based on big data analysis, please refer to Figure 3 , which shows a structure diagram of a product recommendation system based on big data analysis provided by one embodiment of the application, and the system comprises an acquisition module 10, a purchase similarity degree acquisition module 20, a popular degree acquisition module 30, a corrected purchase similarity degree acquisition module 40 and a data processing module 50.
[0100] The acquisition module 10 is used for acquiring products purchased by a target user and each reference user in a current time period, and acquiring features of each user.
[0101] The purchase similarity degree acquisition module 20 is used for dividing the current time period into local time periods, and acquiring a purchase similarity degree of the target user and each reference user in each local time period according to a similar situation of the products purchased by the target user and each reference user in each local time period.
[0102] The popular degree acquisition module 30 is used for dividing users who purchase products in each local time period into user categories according to the features of the users, and acquiring a popular degree of each product purchased in each local time period according to a purchase situation of each product in the user category and a difference of the number of users corresponding to each product and other products.
[0103] The corrected purchase similarity degree acquisition module 40 is used for correcting the purchase similarity degree according to a difference of the popular degree of the products purchased by each user in each local time period and other local time periods, acquiring a trend probability of each user in each local time period, and acquiring a corrected purchase similarity degree of the target user and each reference user in each local time period.
[0104] The data processing module 50 is configured to acquire similar users of the target user based on the modified product purchase similarity, and acquire the current recommended product of the target user based on the product purchased by the similar users.
[0105] It should be noted that the system provided in the above embodiments is only exemplified by the division of the above functional modules. In actual applications, the above functions can be completed by different functional modules according to needs, that is, the internal structure of the computer device is divided into different functional modules to complete all or part of the above described functions. In addition, the product recommendation system and the product recommendation method based on big data analysis provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be repeated here.
[0106] Embodiment 3
[0107] The application further provides a product recommendation device based on big data analysis. The device comprises a memory and a processor. The memory stores executable program codes. The processor is configured to call and execute the executable program codes to execute the product recommendation method based on big data analysis provided in the embodiments. The device can be a chip, a component or a module. The chip can comprise a processor and a memory connected to each other. The memory is configured to store instructions. When the processor calls and executes the instructions, the chip can execute the product recommendation method based on big data analysis provided in the above embodiments.
[0108] In addition, the embodiments of the present application also protect a computer device. Please refer to Figure 4 The computer device comprises a memory 401, a processor 402 and a computer program 403 stored in the memory 401 and running on the processor 402. When the processor 402 executes the computer program 403, the computer device can execute any product recommendation method based on big data analysis introduced above.
[0109] Embodiment 4
[0110] The application further provides a computer readable storage medium, which stores computer program codes. When the computer program codes run on a computer, the computer executes the above related method steps to implement the product recommendation method based on big data analysis provided in the above embodiments.
[0111] Embodiment 5
[0112] The application further provides a computer program product. When the computer program product runs on a computer, the computer executes the above related steps to implement the product recommendation method based on big data analysis provided in the above embodiments.
[0113] Among them, the device, computer readable storage medium, computer program product or chip provided by the embodiment are used for executing the corresponding method provided above, so the beneficial effects that can be achieved are referable to the beneficial effects of the corresponding method provided above, which will not be described here.
[0114] It should be noted that the above-mentioned sequence of the embodiments of the present application is only for description, and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multi-task processing and parallel processing are also possible or can be advantageous.
[0115] Each of the embodiments in the specification is described in a progressive manner, and the same or similar parts between each of the embodiments can be referred to each other. Each embodiment mainly describes the difference from other embodiments.
Claims
1. A product recommendation method based on big data analysis, characterized in that, The method includes the following steps: Get the products purchased by the target user and each reference user within the current time period; get the characteristics of each user; Divide the current time period into local time periods, and obtain the purchase similarity between the target user and each reference user within each local time period based on the similarity of the products purchased by the target user and each reference user within each local time period; Users who purchase products within each local time period are categorized into user categories based on their characteristics. The extent of purchases of each product within each local time period is obtained based on the purchase status of each product in its user category and the difference in the number of users corresponding to each product and other products. Based on the difference in the breadth of products purchased by each user in each local time period from each other local time period, the trend probability of each user in each local time period is obtained to correct the purchase similarity, and the corrected purchase similarity between the target user and each reference user in each local time period is obtained. Based on the modified product purchase similarity, obtain similar users of the target user; based on the products purchased by similar users, obtain the target user's current recommended products.
2. The product recommendation method based on big data analysis as described in claim 1, characterized in that, The method for obtaining the purchase similarity is as follows: Based on the type, function, and price of each purchased product, construct a tag vector for each product; For any local time period and any reference user, obtain the cosine value of the tag vector of each product purchased by the target user and each product purchased by the reference user within that local time period, and use them as similarity analysis values. The mean of all similarity analysis values is used as the purchase similarity between the target user and the reference user within this local time period.
3. The product recommendation method based on big data analysis as described in claim 1, characterized in that, The method for obtaining the aforementioned breadth is as follows: Based on the purchase occurrence of each product in each user category within each local time period and the percentage of users who purchased it in each user category, obtain the first distribution index for each product purchased within each local time period. By positively mapping the difference between the number of users corresponding to each product purchased in each local time period and the mean number of users corresponding to other products, a second distribution index for each product purchased in each local time period is determined. The product of the first distribution index and the second distribution index is normalized to represent the breadth of each product purchased within each local time period.
4. The product recommendation method based on big data analysis as described in claim 3, characterized in that, The method for obtaining the first distribution index is as follows: For any given local time period and any product purchased within that local time period, the user category that purchased that product within that local time period shall be used as the reference user category for that product within that local time period. The ratio of the number of reference user categories to the number of all user categories within that local time period is used as the first feature value; The sum of the percentage of users who purchased this type of product in each reference user category is used as the second feature value; The product of the first eigenvalue and the second eigenvalue is used as the first distribution index of the product purchased within this local time period.
5. The product recommendation method based on big data analysis as described in claim 1, characterized in that, The method for obtaining the power flow probability is as follows: For any given local time period and any given user, the average of the breadth of all types of products purchased by that user within that local time period is used as the overall purchase popularity analysis value for that user within that local time period. The difference between the user's overall purchase popularity analysis value in this local time period and that in each other local time period is used as the first difference. The sum of all the first differences and the result of normalization are taken as the power flow probability of the user in this local time period.
6. The product recommendation method based on big data analysis as described in claim 1, characterized in that, The method for obtaining the corrected purchase similarity is as follows: For any local time period and any reference user, the product of the power flow probabilities of the target user and the reference user within that local time period is negatively correlated and normalized, and used as the similarity correction weight between the target user and the reference user within that local time period. The product of the similarity adjustment weight and the purchase similarity between the target user and the reference user within the local time period is used as the adjusted purchase similarity between the target user and the reference user within the local time period.
7. The product recommendation method based on big data analysis as described in claim 1, characterized in that, The method for obtaining similar users is as follows: For any reference user, the modified purchase similarity between the target user and the reference user is arranged in chronological order from back to front according to the corresponding local time period to obtain the modified purchase similarity sequence; In the modified purchase similarity sequence, the modified purchase similarity is numbered sequentially from left to right, starting from 0; For any modified purchase similarity in the modified purchase similarity sequence, the result of normalizing the difference between the length of the modified purchase similarity sequence and the number of the modified purchase similarity is used as the empirical weight of the modified purchase similarity. The product of the modified purchase similarity and the experience weight is used as the participation value of the modified purchase similarity. The result of summing all the participating values of the modified purchase similarity in the modified purchase similarity sequence and normalizing them is used as the fusion similarity between the target user and the reference user. When the degree of similarity in fusion is greater than the preset threshold for similarity in fusion, the corresponding reference user will be regarded as a similar user of the target user.
8. The product recommendation method based on big data analysis as described in claim 1, characterized in that, The method for obtaining the target user's currently recommended products is as follows: Products that the target user did not purchase within the current time period but that similar users have purchased are considered as the target user's current recommended products. The order of the current recommended products is based on the duration between the last time a product appeared and the current time, from shortest to longest.
9. The product recommendation method based on big data analysis as described in claim 1, characterized in that, The method for obtaining the user category is as follows: For any given local time period, all users who purchase products within that local time period are included in the analysis. Based on the characteristics of each user being analyzed, a feature vector is constructed for each user. Then, the user is unsupervised clustered using the DBSCAN density clustering algorithm based on the feature vector of each user to obtain the user category within the local time period.
10. A product recommendation system based on big data analysis, comprising a memory, a processor, and a computer program stored in the memory and running on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the product recommendation method based on big data analysis as described in any one of claims 1-9.