Intelligent product recommendation method based on collaborative filtering algorithm

Abstract

The invention relates to the technical field of product recommendation, in particular to an intelligent product recommendation method based on a collaborative filtering algorithm. According to the invention, the popular style of each product is obtained according to the product sales volume change trend of each product at different transaction moments; obtaining the conformity of each user in combination with the transaction time characteristics of the corresponding historically purchased products; for any user as a target user, according to the similarity between each emerging product and different historically purchased products, obtaining the product similarity between the emerging products and the historically purchased products, and screening emerging recommended products; according to the transaction times and transaction time distribution of different historically purchased products, obtaining transaction performance of the user corresponding to each historically purchased product; the interest tendency of the target user corresponding to the historical purchased product is obtained; and obtaining a recommendation index of the target user corresponding to the emerging recommended product. According to the method, the accurate recommendation index of each emerging product to the user is obtained, and the product recommendation accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the technical field of product recommendation, and in particular to a product intelligent recommendation method based on a collaborative filtering algorithm. Background Art

[0002] Product recommendation algorithm collaborative filtering, in order to make more accurate product recommendations to users, is usually based on the user's historical behavior data, and then recommends a collection of similar items. However, with the emergence of a large number of emerging products, there is a lack of sufficient historical data, making it difficult for collaborative filtering algorithms to effectively form product recommendations, which is the cold start problem of collaborative filtering algorithms. In the prior art, the recommendation of emerging products is initialized by using a pre-trained model on a large data set, that is, a large amount of data is previewed to provide preliminary recommendations for the algorithm in the form of improving product ratings and exposure. However, the use of big data preview can only improve product exposure and consumes a lot of resources. It is impossible to analyze the correlation between emerging products and historical products, and the accuracy of product recommendations is poor. Summary of the invention

[0003] In order to solve the technical problem that the correlation between emerging products and historical products cannot be analyzed and the accuracy of product recommendations is poor, the purpose of the present invention is to provide a product intelligent recommendation method based on a collaborative filtering algorithm. The technical solution adopted is as follows: The present invention proposes a product intelligent recommendation method based on a collaborative filtering algorithm, the method comprising: Based on the product trading platform, obtain the product sales of emerging products and users' historically purchased products at each transaction time, as well as the number of transactions and transaction time of each historically purchased product by the user; According to the sales trend of each product at different transaction times, the popularity of each product is obtained; according to the popularity of each user's different historical purchase products and the transaction time characteristics of the corresponding historical purchase products, the conformity of each user is obtained; For any user as the target user, the product similarity between each emerging product and each corresponding historically purchased product is obtained based on the similarity between each emerging product and each corresponding historically purchased product, and the emerging recommended products are screened out; According to the transaction times and transaction time distribution of different historically purchased products by the user, the transaction performance of each historically purchased product by the user is obtained; according to the transaction performance of the target user corresponding to the products purchased by different users, multiple highly similar product clusters are obtained; according to the transaction feature distribution between different users for each historically purchased product in the highly similar product cluster, the interest tendency of the target user corresponding to each historically purchased product is obtained; According to the popularity of the emerging recommended products, the corresponding conformity, interest tendencies and product similarities of the target users, the recommendation index of the emerging recommended products corresponding to the target users is obtained.

[0004] Furthermore, the method for obtaining the explosiveness includes: For any product, select the transaction moment with the largest sales value as the hot selling moment; Obtain the difference between the product sales of the corresponding product and the product sales of other products at each transaction moment in the neighborhood of the hot-selling moment as the relative sales difference; The cumulative sum of the relative sales volume differences of each product at all transaction times within the neighborhood of the hot-selling time is obtained as the hot-selling nature of each product.

[0005] Furthermore, the method for obtaining the conformity includes: For each historically purchased product of each user, obtain the time difference between each transaction time in the neighborhood of the corresponding transaction time and the hot product time, select the one with the smallest time difference value, and use the corresponding time difference as the hot product time difference; According to the popularity and time difference of each user's historical product purchases, the conformity of each user is obtained. The popularity is positively correlated with the conformity, while the time difference of the popularity is negatively correlated with the conformity.

[0006] Furthermore, the method for obtaining product similarity includes: Obtain multi-dimensional information about emerging products and historically purchased products of target users, and obtain word vectors corresponding to each dimensional information; The cosine similarity of the word vectors of each identical dimensional information between the emerging product and each historically purchased product is obtained as the dimensional information similarity; the cumulative value of the similarities of all dimensional information between the emerging product and each historically purchased product is obtained as the product similarity between the emerging product and each historically purchased product.

[0007] Furthermore, the method for obtaining the emerging recommended products includes: For the target user, the product similarities between the emerging product and all the historically purchased products are accumulated and normalized to obtain the overall product similarity between the emerging product and all the historically purchased products; Obtain the average of the overall similarities of all emerging products to the product as the average overall product similarity; If the overall similarity of an emerging product to the aforementioned product is greater than the average overall similarity of the products, the corresponding emerging product will be regarded as an emerging recommended product.

[0008] Furthermore, the method for obtaining the transaction performance includes: The ratio of the number of transactions of each historically purchased product to the sum of the number of transactions of all historically purchased products is obtained as the first performance coefficient; the mean difference of transaction times between the same historically purchased products is obtained and negatively correlated mapping is performed as the second performance coefficient; The product of the first performance coefficient and the second performance coefficient corresponding to each historically purchased product is calculated as the transaction performance of the user corresponding to each historically purchased product.

[0009] Furthermore, the method for obtaining the highly similar product cluster includes: According to the transaction performance of target users corresponding to different historically purchased products, ISODATA clustering is performed on all historically purchased products to obtain product clustering clusters; Obtain the mean of the transaction performance of historically purchased products in each product cluster as the local transaction performance; obtain the mean of the local transaction performance of all product clusters as the overall transaction performance; If the local transaction performance of a product cluster is greater than the overall transaction performance, the corresponding product cluster will be used as a high-similar product cluster for the target user.

[0010] Furthermore, the method for obtaining the interest tendency includes: According to the transaction performance, transaction number and product sales characteristics of each historically purchased product between different users, the purchase tendency of the target user for each historically purchased product is obtained; For highly similar product clusters, obtain the difference between the transaction time of each user for each historically purchased product and the transaction time of each other historically purchased product, and perform negative correlation mapping as the transaction activity; accumulate the transaction activities corresponding to all other purchased products as the transaction frequency; The product of the first performance coefficient and the transaction frequency of each user for each historically purchased product is obtained as the transaction interest. According to the deviation of the transaction interest between the target user and each other user for each historically purchased product, and the purchase tendency of the target user, the interest tendency of the target user for each historically purchased product is obtained, and both the deviation of the transaction interest and the purchase tendency are positively correlated with the interest tendency.

[0011] Furthermore, the method for acquiring the purchase tendency includes: For any historically purchased product, obtain the ratio of the number of transactions of the user corresponding to the historically purchased product to the sum of the product sales of the historically purchased product at all transaction times as the user's purchase ratio; The product of the transaction performance of the user's corresponding historical purchase product and the user's purchase proportion is obtained as the purchase influence index of the user's corresponding historical purchase product; the average of the purchase influence indexes of all other users' corresponding historical purchase products is obtained as the average purchase influence index; According to the deviation between the purchase influence index of the target user's corresponding historical purchase products and the average purchase influence index, the purchase tendency of the target user for the historical purchase products is obtained.

[0012] Furthermore, the method for obtaining the recommendation index includes: Select the product with the largest similarity value between the emerging recommended product and each historically purchased product in the high similarity product cluster of the target user, and use the corresponding product similarity as the first recommendation coefficient, and use the interest tendency of the corresponding historically purchased product as the second recommendation coefficient; According to the popularity of the emerging recommended products, the corresponding conformity of the target users, the first recommendation coefficient and the second recommendation coefficient, the recommendation index of the emerging recommended products corresponding to the target users is obtained. The popularity, conformity, the first recommendation coefficient and the second recommendation coefficient are all positively correlated with the recommendation index.

[0013] The present invention has the following beneficial effects: The present invention obtains the popularity of each product according to the product sales volume change trend of each product at different transaction moments. Through the sales volume change trend, the sales volume of different products in different time periods can be accurately captured to evaluate the popularity of the products. The conformity of each user can be obtained according to the popularity of different historically purchased products of each user and the transaction time characteristics of the corresponding historically purchased products. For any user as the target user, the product similarity between each emerging product and each historically purchased product can be obtained according to the similarity between each emerging product and the corresponding different historically purchased products, and the emerging recommended products can be screened out. The emerging products can be accurately matched with the historically purchased products of the user, so as to screen out emerging products that meet the user's interests. Recommend products; based on the transaction times and transaction time distribution of different historically purchased products by users, obtain the transaction performance of each historically purchased product by the user, and understand the user's preference for product purchasing behavior; based on the transaction performance of target users corresponding to different users' purchases of products, obtain multiple highly similar product clusters, and products with similar transaction performances can be classified into one category, which is helpful to discover the correlation between products; based on the transaction feature distribution between different users corresponding to each historically purchased product in the highly similar product cluster, obtain the interest tendency of the target user corresponding to each historically purchased product, which is helpful to more deeply understand the user's needs and preferences, thereby providing more personalized recommendations; obtain the recommendation index of the target user corresponding to the emerging recommended product. The present invention obtains the accurate recommendation index of each emerging product for the user, and improves the accuracy of product recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0015] Figure 1 A flowchart of a product intelligent recommendation method based on a collaborative filtering algorithm provided by an embodiment of the present invention; Figure 2 A flow chart of a method for obtaining a hot item provided by an embodiment of the present invention; Figure 3 A flow chart of a method for obtaining product similarity provided by an embodiment of the present invention; Figure 4 A flow chart of a method for acquiring purchase tendency provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation method, structure, features and effects of a product intelligent recommendation method based on a collaborative filtering algorithm proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.

[0017] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0018] The specific scheme of the product intelligent recommendation method based on collaborative filtering algorithm provided by the present invention is described in detail below with reference to the accompanying drawings.

[0019] See also Figure 1 , which shows a flowchart of a product intelligent recommendation method based on a collaborative filtering algorithm provided by an embodiment of the present invention. The specific method includes: Step S1: Based on the product trading platform, obtain the product sales of emerging products and the user's historically purchased products at each transaction time, as well as the number of transactions and transaction time of each historically purchased product by the user.

[0020] In an embodiment of the present invention, since emerging products have no historical information and often appear in the market for the first time, they cannot participate in the recommendation, which is called the cold start of the product. In order to improve users' awareness of and willingness to buy emerging products, it is necessary to analyze the relationship between emerging products and users' historically purchased products, establish social relationships between products, and make personalized product recommendations to users based on users' behavioral preferences. Therefore, it is necessary to analyze the attributes of the products and the behavioral habits of users. First, the product sales of emerging products and users' historically purchased products at each transaction time are obtained through the product trading platform, as well as the number of transactions and transaction time of each historically purchased product by the user.

[0021] It should be noted that in the embodiment of the present invention, the transaction time interval is one day, that is, the product sales volume of the product on each day is obtained as the product sales volume at each transaction time; in other embodiments of the present invention, the size of the transaction time interval can be set according to the specific circumstances, which is not limited or elaborated here.

[0022] It should be noted that, in one embodiment of the present invention, the historical purchases of users on the product trading platform within the current year are obtained for analysis, and emerging products are products that appeared on the product trading platform within the first half of the month at the current time.

[0023] It should be noted that, in the embodiments of the present invention, in order to facilitate subsequent processing of the data and avoid differences between the units and magnitudes of the values ​​of the data, the data is standardized to eliminate the influence of the dimension in the data calculation.

[0024] Step S2: Obtain the popularity of each product based on the sales volume change trend of each product at different transaction times; obtain the conformity of each user based on the popularity of different historically purchased products of each user and the transaction time characteristics of the corresponding historically purchased products.

[0025] Considering that some users have a herd shopping behavior, even if the emerging product does not have a high similarity with the user's historical purchase products, because the product is a hot-selling product in the short term and is purchased by most people, the corresponding emerging product should also be recommended. It is necessary to analyze the sales trend of products at different transaction times to reflect the market dynamics of products in different periods of time. The more the sales of a product surges in the short term, the more likely it is to be a hot-selling product. According to the sales trend of each product at different transaction times, the hot-selling nature of each product can be obtained.

[0026] Preferably, in one embodiment of the present invention, the method for obtaining the hot selling property can be found in Figure 2 , which shows a flow chart of a method for obtaining a hot item, including: Step S201: For any product, select the transaction moment with the largest product sales value as the hot-selling moment.

[0027] Since hot-selling products have the characteristics of quickly becoming popular and surging sales in a short period of time, the transaction moment with the largest product sales value is selected as the hot-selling moment, which represents the peak sales volume of the product, is the most popular moment of the product, and is the key reference point for evaluating the hot-selling nature of the product.

[0028] Step S202: Obtain the difference between the product sales of the corresponding product and the product sales of other products at each transaction moment within the neighborhood of the hot-selling moment as the relative sales difference.

[0029] Within the neighborhood of a hit moment, the sales volume of different products will vary, and the relative sales volume difference can more accurately reflect the relative performance of the hit product.

[0030] It should be noted that, in one embodiment of the present invention, the method for obtaining the neighborhood range includes: taking the hot-selling moment as a benchmark, obtaining the difference in product sales between each transaction moment and the next transaction moment on the left and right sides of the hot-selling moment, and calculating the ratio of the difference result and the product sales at the next transaction moment as the sales change rate of each transaction moment; if the sales change rate of the transaction moments on the left and right sides of the hot-selling moment is smaller than the sales change rate of the previous adjacent transaction moment, the corresponding transaction moment is taken as the end time; the time range between the end times on both sides constitutes the neighborhood range of the hot-selling moment.

[0031] Step S203: Obtain the cumulative sum of the relative sales volume differences of all transaction times of each product within the neighborhood of the hot-selling time as the hot-selling nature of each product.

[0032] In one embodiment of the present invention, the formula for the popularity of a product is expressed as: ; in, Indicates The popularity of a product; Indicates Products in Product sales at the time of transaction; Indicates Other products in Product sales at the time of transaction; Indicates the quantity of other products at the corresponding trading moment; Indicates the number of transaction moments within the scope of the hot moment.

[0033] In the explosive formula, Indicates Products and Other products in Product sales volume at the time of transaction, i.e., relative sales volume difference; It means calculating the cumulative sum of the relative sales differences of each product at all times within the neighborhood of the hot-selling moment, that is, the hot-selling nature. The larger the relative sales difference, the higher the Products in The greater the product sales volume at the time of transaction relative to other products, the more likely it is that the product will be a hit and its potential for popularity will be higher.

[0034] By comprehensively analyzing the popularity of different historical products purchased by users and the transaction time characteristics of the corresponding historical products purchased, we can quantify the herd tendency of users to purchase products. If a user frequently purchases a product with a high popularity, it is considered that the user has a strong herd tendency. The transaction time characteristics of historical products purchased help to understand consumers' preferences for different types of products in different time periods. According to the popularity of different historical products purchased by each user and the transaction time characteristics of the corresponding historical products purchased, the herd tendency of each user can be obtained.

[0035] Preferably, in one embodiment of the present invention, the method for obtaining conformity includes: For each historically purchased product of each user, obtain the time difference between each transaction time in the neighborhood of the corresponding transaction time and the hot product time, select the one with the smallest time difference value, and use the corresponding time difference as the hot product time difference; According to the popularity and time difference of each user's historical product purchases, the conformity of each user is obtained. The popularity is positively correlated with the conformity, while the time difference of the popularity is negatively correlated with the conformity.

[0036] In one embodiment of the present invention, the formula of conformity is expressed as: ; in, Indicates The conformity of individual users; Indicates The popularity of the historically purchased products; Indicates The transaction time of historical purchase of products; Indicates The transaction time in the neighborhood of the hot moment of the historical purchase of the product; Indicates The number of historically purchased products by a user; Indicates taking the minimum value; represents the logistic function.

[0037] In the formula of conformity, The greater the popularity of a product purchased in history, the greater the probability that the user tends to buy the popular product, and the greater the herd tendency; Indicates The time difference between the transaction time of the historical purchase product and each transaction time in the neighborhood range of the hot-selling time is taken, and the one with the smallest time difference value is taken, that is, the hot-selling time difference. The larger the hot-selling time difference, the farther the transaction time of the historical purchase product is from the neighborhood range of the corresponding hot-selling time, and the smaller the herd mentality.

[0038] Step S3: For any user as the target user, the product similarity between each emerging product and each historically purchased product is obtained according to the similarity between each emerging product and the corresponding different historically purchased products, and the emerging recommended products are screened out.

[0039] Each user's interests and needs are unique, and different products have different performance characteristics at different levels. By analyzing the similarity between emerging products and users' historically purchased products, we can deeply explore the internal connection between products and more accurately capture the correlation between emerging products and user needs, thereby providing recommendations that meet user preferences. For any user as the target user, based on the similarity between each emerging product and the corresponding different historically purchased products, we can obtain the product similarity between each emerging product and each historically purchased product.

[0040] Preferably, in one embodiment of the present invention, the method for obtaining product similarity is as follows: Figure 3 , which shows a flow chart of a method for obtaining product similarity, including: Step S301: Obtain multi-dimensional information of emerging products and historically purchased products of target users, and obtain word vectors corresponding to each dimensional information.

[0041] It should be noted that obtaining multi-dimensional information of products based on product trading platforms, including product uses, parameters, prices, brands and other attributes reflecting the products, is helpful for understanding the products; in order to facilitate data processing, the natural language NLP algorithm is first used to digitize the multi-dimensional information, and the jieba algorithm is used to remove meaningless function words and prepositions from the multi-dimensional information of emerging products and users' historical purchases, and decompose them into phrases. The context-embedded BERT algorithm model is then used to generate word vectors for the information, which can be converted into a numerical form that the model can process, which helps capture the correlation between products; the specific means are technical means well known to those skilled in the art and will not be elaborated here.

[0042] Step S302: Obtain the cosine similarity of the word vectors of each identical dimensional information between the emerging product and each historically purchased product as the dimensional information similarity; obtain the cumulative value of all dimensional information similarities between the emerging product and each historically purchased product as the product similarity between the emerging product and each historically purchased product.

[0043] Word vectors not only contain directional values. Similar information often has similar directions, while word vectors with different directions represent different meanings. Cosine similarity can capture the directional similarity between word vectors and analyze the correlation between dimensional information. The greater the cosine similarity of the word vectors of dimensional information, the greater the similarity between products.

[0044] In one embodiment of the present invention, for a target user, the formula for product similarity is expressed as: ; in, Indicates emerging products With Product similarity between historically purchased products; Indicates emerging products No. word vector with dimensional information; Indicates The first product purchased in history word vector with dimensional information; Indicates the number of dimensional information in the product; Indicates the number of products that the user has purchased in the past; Represents the cosine similarity function.

[0045] In the formula for product similarity, It means calculating the cosine similarity of the word vectors of each identical dimensional information between the emerging product and each historically purchased product, that is, the dimensional information similarity, and calculating the cumulative value of the similarity of all dimensional information, that is, the product similarity between the emerging product and each historically purchased product. The greater the similarity of the word vectors of all dimensional information, the greater the product similarity.

[0046] It should be noted that in other embodiments of the present invention, the Pearson correlation coefficient algorithm can also be used to obtain the similarity of the word vectors of the dimensional information between products. The specific means are technical means well known to those skilled in the art and will not be elaborated here.

[0047] The similarity between emerging products and historically purchased products can more accurately grasp users' purchasing preferences and needs, and can efficiently screen out emerging products that are similar to users' historical purchase records, screen emerging recommended products for each user, and improve the accuracy and efficiency of recommendations.

[0048] Preferably, in one embodiment of the present invention, the method for obtaining emerging recommended products includes: For the target user, the product similarities between the emerging product and all the historically purchased products are accumulated and normalized to be the overall product similarity between the emerging product and all the historically purchased products; the mean of the overall product similarities corresponding to all emerging products is obtained as the average overall product similarity; If the overall similarity of the corresponding products of an emerging product is greater than the overall similarity of the average products, the corresponding emerging product will be regarded as an emerging recommended product.

[0049] Step S4: Based on the transaction times and transaction time distributions of different historically purchased products by the user, obtain the transaction performance of each historically purchased product by the user; based on the transaction performance of the target user corresponding to the purchase of products by different users, obtain multiple highly similar product clusters; based on the transaction feature distributions between different users for each historically purchased product within the highly similar product clusters, obtain the interest tendency of the target user for each historically purchased product.

[0050] Analyzing the number of user transactions and transaction times helps us understand changes in user purchasing behavior over different time periods. This can reveal users’ purchasing habits, preferences, and activity levels, as well as whether they show a sustained interest in purchasing certain products. Therefore, based on the number of transactions and transaction time distributions for different historically purchased products, we can obtain the user’s transaction performance for each historically purchased product.

[0051] Preferably, in one embodiment of the present invention, the method for obtaining transaction performance includes: The ratio of the number of transactions of each historically purchased product to the sum of the number of transactions of all historically purchased products is obtained as the first performance coefficient; the mean difference of transaction times between the same historically purchased products is obtained and negatively correlated mapping is performed as the second performance coefficient; The product of the first performance coefficient and the second performance coefficient corresponding to each historically purchased product is calculated as the transaction performance of the user corresponding to each historically purchased product. In one embodiment of the present invention, the formula for the transaction performance is expressed as: ; in, Indicates User's Transaction performance of historical purchased products; Indicates User's The number of historical purchase transactions of the product; Indicates The sum of all historical purchase transaction times of a user; Indicates User's The mean difference in transaction time between historically purchased products and the same historically purchased products.

[0052] In the trading performance formula, Indicates the calculation User's The ratio of the number of transactions of the historical purchase product to the sum of the number of transactions of all historical purchase products is the first performance coefficient. The larger the ratio, the higher the performance coefficient. User's The greater the number of transactions of a historically purchased product, the more historically purchased products a user has traded, and the more obvious the transaction performance is; Indicates User's The second performance coefficient is normalized by the mean difference in transaction time between the historically purchased products and the same historically purchased products. The larger the second performance coefficient is, the greater the difference in transaction time between the same historically purchased products is, and the smaller the purchase performance of the corresponding historically purchased products is.

[0053] It should be noted that, in one embodiment of the present invention, The negative correlation mapping is performed by taking the inverse to form a negative correlation relationship in which the larger the difference mean of the transaction time is, the smaller the second performance coefficient is, and the smaller the purchase performance is. In other embodiments of the present invention, it can also be used The function performs negative correlation mapping, and the specific means are technical means well known to those skilled in the art and will not be described in detail here.

[0054] When purchasing products, users often show certain behavior patterns or preferences. By analyzing the transaction performance of target users in relation to different users purchasing products, we can identify the user's behavior patterns in purchasing products and group products with similar characteristics into one category. This helps to screen out products that meet user preferences for targeted analysis. Based on the target user's transaction performance in relation to different users purchasing products, we can obtain multiple highly similar product clusters.

[0055] Preferably, in one embodiment of the present invention, the method for obtaining a highly similar product cluster includes: According to the transaction performance of target users corresponding to different historically purchased products, ISODATA clustering is performed on all historically purchased products to obtain product clustering clusters; Obtain the mean of the transaction performance of historically purchased products in each product cluster as the local transaction performance; obtain the mean of the local transaction performance of all product clusters as the overall transaction performance; If the local transaction performance of a product cluster is greater than the overall transaction performance, the corresponding product cluster will be used as a high-similar product cluster for the target user.

[0056] It should be noted that ISODATA clustering is a technical means well known to those skilled in the art and will not be described in detail here.

[0057] The transaction characteristics of users can reflect their purchasing habits, preferences and interests. In a highly similar product cluster, there is often a certain correlation between products. By analyzing the transaction characteristics of products between different users, we can infer the target user's interest in different products. Based on the distribution of transaction characteristics between different users for each historically purchased product in the highly similar product cluster, we can obtain the target user's interest tendency for each historically purchased product.

[0058] Preferably, in one embodiment of the present invention, the method for acquiring interest tendency includes: According to the transaction performance, transaction number and product sales characteristics of each historically purchased product between different users, the purchase tendency of the target user for each historically purchased product is obtained; Preferably, in one embodiment of the present invention, the method for obtaining the purchase tendency is as follows Figure 4 , which shows a flow chart of a method for obtaining purchase tendency, including: Step S401: For any historically purchased product, obtain the ratio of the number of transactions of the user corresponding to the historically purchased product to the sum of the product sales of the historically purchased product at all transaction times as the user's purchase ratio.

[0059] By analyzing the ratio, we can reflect the user's purchase preference for historically purchased products, and by comparing the number of user transactions with the total sales volume of the product, we can quantify the user's relative preference for the product.

[0060] Step S402: The transaction performance of the user's corresponding historical purchase products and the user's purchase proportion are integrated to obtain the purchase influence index of the user's corresponding historical purchase products; the average of the purchase influence indexes of all other users' corresponding historical purchase products is obtained as the average purchase influence index.

[0061] It should be noted that, in some embodiments of the present invention, the transaction performance and the user purchase ratio can be integrated by addition or multiplication. The specific means are technical means well known to those skilled in the art and will not be elaborated here.

[0062] Transaction performance reflects the user's purchasing power and activity, and the user purchase percentage reflects the user's relative preference for the product. Combined analysis can provide a more comprehensive understanding of the user's purchase influence index, which is used to evaluate the user's influence on the purchase of specific products.

[0063] The purchase influence indicators of all other users corresponding to historically purchased products are quantified by taking the average value to obtain the average purchase influence index, which is used to compare the purchase influence levels of different user groups on historically purchased products.

[0064] Step S403: obtaining the target user's purchase tendency for the historically purchased products according to the deviation between the purchase influence index of the target user corresponding to the historically purchased products and the average purchase influence index.

[0065] It should be noted that, in one embodiment of the present invention, the formula for purchasing propensity is expressed as: ; in, Indicates the target user No. The purchase tendency of historically purchased products; Indicates the target user No. Transaction performance of historical purchased products; Indicates the target user No. The number of historical purchase transactions of the product; Represents the sum of the sales volume of historically purchased products at all transaction moments; Indicates The number of other users corresponding to the historically purchased products; Indicates of other users The number of historical purchase transactions of the product; Indicates of other users Transaction performance of historical purchased products; represents the logistic function.

[0066] In the purchase propensity formula, Indicates the target user No. The ratio of the number of transactions of a historically purchased product to the sum of the sales volume of the historically purchased product at all transaction times, that is, the user purchase ratio. The larger the user purchase ratio, the more important the historically purchased product is to the user. Indicates the target user No. The product of the transaction performance of the historically purchased product and the user's purchase proportion is the purchase influence index. The larger the purchase influence index, the greater the transaction performance, the greater the user's purchase proportion, and the greater the purchase tendency of the target user for the corresponding historically purchased product; It means calculating the average purchase influence index of all other users corresponding to the historical purchase products, that is, the average purchase influence index. Indicates the target user No. The deviation between the purchase influence index of a historically purchased product and the average purchase influence index. The larger the deviation, the greater the purchase tendency of the target user for the historically purchased product compared with other users.

[0067] For highly similar product clusters, obtain the difference between the transaction time of each user for each historically purchased product and the transaction time of each other historically purchased product, and perform negative correlation mapping as the transaction activity; accumulate the transaction activities corresponding to all other purchased products as the transaction frequency; The product of the first performance coefficient and the transaction frequency of each user for each historically purchased product is obtained as the transaction interest. According to the deviation of the transaction interest between the target user and each other user for each historically purchased product, and the purchase tendency of the target user, the interest tendency of the target user for each historically purchased product is obtained. Both the deviation of the transaction interest and the purchase tendency are positively correlated with the interest tendency.

[0068] In one embodiment of the present invention, for any historically purchased product in a highly similar product cluster, the formula for interest tendency is expressed as: ; in, Indicates the target user No. The interest tendency of historically purchased products; Indicates The number of other users corresponding to the historically purchased products; Indicates the target user No. The purchase tendency of historically purchased products; Indicates The number of other historically purchased products in the high-similar product cluster corresponding to the historically purchased product; Indicates Target users with historically purchased products The transaction time and The difference between the transaction moments of other historical purchases of the product; Indicates The historical purchase product corresponds to The transaction time of other users is The difference between the transaction moments of other historical purchases of the product; Indicates the target user No. The number of historical purchase transactions of the product; Indicates the target user The sum of all historical purchase transaction times of products; Indicates of other users The number of historical purchase transactions of the product; Indicates The sum of all historical purchase transaction times of other users; represents the logistic function.

[0069] In the formula of interest tendency, Indicates the target user No. The ratio of the transaction number of a historically purchased product to the sum of the transaction numbers of all historically purchased products is the first performance coefficient of the target user. Indicates Target users with historically purchased products The transaction time and The difference between the transaction times of other historical purchase products is negatively correlated, that is, the transaction activity. The smaller the difference, the closer the transaction time, and the greater the transaction activity; It means calculating and accumulating the transaction activity corresponding to all other purchased products as the transaction frequency. It means calculating the product of the first performance coefficient and the transaction frequency as the transaction interest. That is, the larger the first performance coefficient is, the greater the transaction activity is, the greater the transaction frequency is, the more likely it is that there is a greater demand, and the corresponding transaction interest is greater; It represents the deviation of the transaction interest between the target user and every other user for each historically purchased product. The larger the deviation, the greater the transaction interest of the target user for the historically purchased product relative to other users, and the greater the interest tendency in the historically purchased product. The greater the purchase tendency and the greater the interest tendency, the more likely it is a fast-moving consumer product.

[0070] Step S5: According to the popularity of the emerging recommended products, the conformity, interest tendency and product similarity of the target users, the recommendation index of the emerging recommended products corresponding to the target users is obtained.

[0071] Products with strong popularity can attract more users' attention and increase product exposure and popularity; the target users' conformity makes users more susceptible to popular products or hot recommendations, thereby increasing their willingness to buy; interest tendencies help understand users' preference for specific types of products; product similarity reflects the similarity between emerging products and historically purchased products. The higher the similarity, the easier it is for users to accept and recognize it. Therefore, based on the popularity of emerging recommended products, the corresponding conformity, interest tendencies and product similarities of target users, the recommendation index of emerging recommended products corresponding to target users is obtained.

[0072] Preferably, in one embodiment of the present invention, the method for obtaining the recommendation index includes: Select the product with the largest similarity value between the emerging recommended product and each historically purchased product in the target user's highly similar product cluster, and use the corresponding product similarity as the first recommendation coefficient, and use the interest tendency of the corresponding historically purchased product as the second recommendation coefficient; According to the popularity of the emerging recommended products, the corresponding conformity of the target users, the first recommendation coefficient and the second recommendation coefficient, the recommendation index of the emerging recommended products corresponding to the target users is obtained. The popularity, conformity, the first recommendation coefficient and the second recommendation coefficient are all positively correlated with the recommendation index.

[0073] Among them, the greater the popularity, the more likely the emerging recommended product will be purchased by most users in the short term. The greater the user's tendency to follow the crowd, the more frequently they will buy the explosive product. The larger the first recommendation coefficient and the second recommendation coefficient, the greater the similarity between the emerging recommended product and the historically purchased product, and the historically purchased product is in line with the user's shopping preferences. Therefore, the emerging recommended product needs to be recommended more.

[0074] In one embodiment of the present invention, the formula of the recommendation index is expressed as: ; in, Indicates new recommended products Target users Recommendation index: Indicates the target user The herd mentality; Indicates new recommended products The explosiveness of Indicates new recommended products With target users The first in the highly similar product cluster Product similarity between historically purchased products; It represents the maximum value function; Indicates selection of emerging recommended products With target users Different historical purchase products within a highly similar product cluster The product with the largest similarity value is the first recommendation coefficient; Indicates the target user when the product similarity value is the largest The historical interest tendency of purchasing products, i.e. the second recommendation coefficient; represents the logistic function.

[0075] In the formula of the recommendation index, when the emerging product is a hit product, the stronger the user's herd mentality, the higher the recommendation rate of the new product. The higher the explosiveness of The larger it is, the greater the possibility of being recommended; the greater the product similarity, the larger the first recommendation coefficient, the higher the interest tendency in historical product purchases, the larger the second recommendation coefficient, the more likely it is that the products are of the same category, and the larger the recommendation index of emerging recommended products.

[0076] Based on this, for each user as a target user, the recommendation index of each user's emerging recommended product is obtained, and the overall similarity between the emerging product and the historically purchased product can be weighted and adjusted, so that when the collaborative filtering algorithm is subsequently performed, the emerging product with a higher recommendation index occupies a more important position in the recommendation list, thereby increasing the exposure of the emerging product and helping to accurately recommend products to users. The specific collaborative filtering algorithm is a technical means well known to those skilled in the art and will not be described in detail here.

[0077] In summary, the present invention obtains the popularity of each product based on the sales volume change trend of each product at different transaction moments; obtains the conformity of each user in combination with the transaction time characteristics of the corresponding historically purchased products; for any user as the target user, obtains the product similarity between each emerging product and each historically purchased product based on the similarity between each emerging product and different historically purchased products, and screens out emerging recommended products; obtains the transaction performance of the user corresponding to each historically purchased product based on the transaction times and transaction time distribution of different historically purchased products; and then obtains the interest tendency of the target user corresponding to the historically purchased products; obtains the recommendation index of the emerging recommended product corresponding to the target user. The present invention obtains the accurate recommendation index of each emerging product for the user, and improves the accuracy of product recommendations.

[0078] It should be noted that the sequence of the above embodiments of the present invention is only for description and does not represent the advantages and disadvantages of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0079] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

Claims

1. A product intelligent recommendation method based on collaborative filtering algorithm, characterized in that: The method comprises: Based on the product trading platform, obtain the product sales of emerging products and users' historically purchased products at each transaction time, as well as the number of transactions and transaction time of each historically purchased product by the user; According to the sales trend of each product at different transaction times, the popularity of each product is obtained; according to the popularity of each user's different historical purchase products and the transaction time characteristics of the corresponding historical purchase products, the conformity of each user is obtained; For any user as the target user, the product similarity between each emerging product and each corresponding historically purchased product is obtained based on the similarity between each emerging product and each corresponding historically purchased product, and the emerging recommended products are screened out; According to the transaction times and transaction time distribution of different historically purchased products by the user, the transaction performance of each historically purchased product by the user is obtained; according to the transaction performance of the target user corresponding to the products purchased by different users, multiple highly similar product clusters are obtained; according to the transaction feature distribution between different users for each historically purchased product in the highly similar product cluster, the interest tendency of the target user corresponding to each historically purchased product is obtained; According to the popularity of the emerging recommended products, the corresponding conformity, interest tendencies and product similarities of the target users, the recommendation index of the emerging recommended products corresponding to the target users is obtained.

2. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 1, characterized in that: The method for obtaining the explosive product includes: For any product, select the transaction moment with the largest sales value as the hot selling moment; Obtain the difference between the product sales of the corresponding product and the product sales of other products at each transaction moment in the neighborhood of the hot-selling moment as the relative sales difference; The cumulative sum of the relative sales volume differences of each product at all transaction times within the neighborhood of the hot-selling time is obtained as the hot-selling nature of each product.

3. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 2 is characterized in that: The method for obtaining the conformity includes: For each historically purchased product of each user, obtain the time difference between each transaction time in the neighborhood of the corresponding transaction time and the hot product time, select the one with the smallest time difference value, and use the corresponding time difference as the hot product time difference; According to the popularity and time difference of each user's historical product purchases, the conformity of each user is obtained. The popularity is positively correlated with the conformity, while the time difference of the popularity is negatively correlated with the conformity.

4. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 1 is characterized in that: The method for obtaining the product similarity includes: Obtain multi-dimensional information about emerging products and historically purchased products of target users, and obtain word vectors corresponding to each dimensional information; The cosine similarity of the word vectors of each identical dimensional information between the emerging product and each historically purchased product is obtained as the dimensional information similarity; the cumulative value of the similarities of all dimensional information between the emerging product and each historically purchased product is obtained as the product similarity between the emerging product and each historically purchased product.

5. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 1 is characterized in that: The method for obtaining the emerging recommended products includes: For the target user, the product similarities between the emerging product and all the historically purchased products are accumulated and normalized to obtain the overall product similarity between the emerging product and all the historically purchased products; Obtain the average of the overall similarities of all emerging products to the product as the average overall product similarity; If the overall similarity of an emerging product to the aforementioned product is greater than the average overall similarity of the products, the corresponding emerging product will be regarded as an emerging recommended product.

6. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 1 is characterized in that: The method for obtaining the transaction performance includes: The ratio of the number of transactions of each historically purchased product to the sum of the number of transactions of all historically purchased products is obtained as the first performance coefficient; the mean difference of transaction times between the same historically purchased products is obtained and negatively correlated mapping is performed as the second performance coefficient; The product of the first performance coefficient and the second performance coefficient corresponding to each historically purchased product is calculated as the transaction performance of the user corresponding to each historically purchased product.

7. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 6 is characterized in that: The method for obtaining the highly similar product cluster includes: According to the transaction performance of target users corresponding to different historically purchased products, ISODATA clustering is performed on all historically purchased products to obtain product clustering clusters; Obtain the mean of the transaction performance of historically purchased products in each product cluster as the local transaction performance; obtain the mean of the local transaction performance of all product clusters as the overall transaction performance; If the local transaction performance of a product cluster is greater than the overall transaction performance, the corresponding product cluster will be used as a high-similar product cluster for the target user.

8. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 7 is characterized in that: The method for obtaining the interest tendency includes: According to the transaction performance, transaction number and product sales characteristics of each historically purchased product between different users, the purchase tendency of the target user for each historically purchased product is obtained; For highly similar product clusters, obtain the difference between the transaction time of each user for each historically purchased product and the transaction time of each other historically purchased product, and perform negative correlation mapping as the transaction activity; accumulate the transaction activities corresponding to all other purchased products as the transaction frequency; The product of the first performance coefficient and the transaction frequency of each user for each historically purchased product is obtained as the transaction interest. According to the deviation of the transaction interest between the target user and each other user for each historically purchased product, and the purchase tendency of the target user, the interest tendency of the target user for each historically purchased product is obtained, and both the deviation of the transaction interest and the purchase tendency are positively correlated with the interest tendency.

9. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 8, characterized in that: The method for obtaining the purchase tendency includes: For any historically purchased product, obtain the ratio of the number of transactions of the user corresponding to the historically purchased product to the sum of the product sales of the historically purchased product at all transaction times as the user's purchase ratio; The product of the transaction performance of the user's corresponding historical purchase product and the user's purchase proportion is obtained as the purchase influence index of the user's corresponding historical purchase product; the average of the purchase influence indexes of all other users' corresponding historical purchase products is obtained as the average purchase influence index; According to the deviation between the purchase influence index of the target user's corresponding historical purchase products and the average purchase influence index, the purchase tendency of the target user for the historical purchase products is obtained.

10. The product intelligent recommendation method based on collaborative filtering algorithm according to claim 1, characterized in that: The method for obtaining the recommendation index includes: Select the product with the largest similarity value between the emerging recommended product and each historically purchased product in the high similarity product cluster of the target user, and use the corresponding product similarity as the first recommendation coefficient, and use the interest tendency of the corresponding historically purchased product as the second recommendation coefficient; According to the popularity of the emerging recommended products, the corresponding conformity of the target users, the first recommendation coefficient and the second recommendation coefficient, the recommendation index of the emerging recommended products corresponding to the target users is obtained. The popularity, conformity, the first recommendation coefficient and the second recommendation coefficient are all positively correlated with the recommendation index.

Images