A user behavior analysis method based on big data of online marketing promotion

Abstract

The invention discloses a user behavior analysis method based on network marketing promotion big data, which relates to the field of data analysis. The method comprises: collecting browsing records of browsed commodities and purchase records of purchased commodities of users to be analyzed; obtaining the correlation between the purchased commodities of the users and each browsed commodity; determining the target browsed commodity by using the correlation between the purchased commodity and the corresponding browsed commodity; obtaining the decision cycle and average browsing frequency of the users, and obtaining the purchase probability of each category of commodities by the users at the current moment by using the browsing cycle and browsing frequency of each category of browsed commodities and the decision cycle and average browsing frequency of the users; and determining the number of push windows of each category of commodities on the user's browsing page by using the purchase probability and the number of windows of the browsing page. The invention improves the utilization rate of network marketing data and realizes the accurate and reasonable push of commodity windows.

Description

Technical Field

[0001] The present invention relates to the field of data analysis, and in particular to a user behavior analysis method based on network marketing promotion big data. Background Art

[0002] With the popularization of the Internet and mobile terminals and the mature application of cloud computing technology, the arrival of the big data era is an inevitable trend. Using big data to do network marketing and network promotion is one of the main ways for traditional enterprises to do marketing, and it is also the most effective means at present. However, with the development of the Internet and the surge in the number of netizens, traditional network promotion methods are facing huge impacts and changes.

[0003] Generally, enterprises will analyze the consumption needs of users and adjust their marketing plans and strategic directions in a timely manner. There is huge potential value hidden in the user behavior data on the online marketing platform. The marketing platform will screen a large number of products by analyzing the users' browsed products and push related products to users. However, in the same period, users browse too many types of products, and some browsed products do not meet consumer needs, but big data will continue to push and occupy the push window, ignoring users' needs for other products, and failing to truly meet users' actual consumption needs. In the long run, the marketing plans promoted by enterprises will gradually lose users' credibility. Therefore, the present invention proposes a user behavior analysis method based on online marketing promotion big data. Summary of the invention

[0004] The present invention provides a user behavior analysis method based on network marketing promotion big data to solve the existing problems.

[0005] The present invention provides a user behavior analysis method based on network marketing promotion big data, which adopts the following technical scheme:

[0006] Collect the browsing history of the products browsed by the users to be analyzed on the platform and the purchase history of the products purchased;

[0007] Acquire the same commodity features between the user's purchased commodity and each corresponding browsed commodity, and use the same commodity features to acquire the relevance between the purchased commodity and each corresponding browsed commodity;

[0008] Sort the relevance between the purchased product and each corresponding browsed product by size to obtain a relevance sequence, obtain the first mutation relevance in the relevance sequence, and use all browsed products with relevance greater than the mutation relevance as target browsed products for the purchased product;

[0009] The maximum time difference between each purchased product of the user and the corresponding target browsed product is used to obtain the user's decision cycle; the time difference between each purchased product of the user and the corresponding target browsed product and the number of target browsed products are used to obtain the user's average browsing frequency;

[0010] Cluster all browsed products from the last time the user purchased a product to the current moment according to product characteristics to obtain multiple categories of products, and use the browsing time and browsing quantity of each category of browsed products to obtain the browsing cycle and browsing frequency of each category of products;

[0011] The user's browsing cycle, browsing frequency, decision cycle and average browsing frequency of each category of browsing products are used to obtain the purchase probability of each category of products purchased by the user at the current moment;

[0012] The number of push windows for each type of product on the user's browsing page is determined by using the purchase probability of the user purchasing each type of browsing product at the current moment and the number of windows on the browsing page.

[0013] Furthermore, before sorting the relevance of each purchased product to the corresponding browsed product by size,

[0014] The correlation between the product and each corresponding browsed product is corrected, and the correction method includes:

[0015] Construct a neural network and input the same product features of the purchased products and the corresponding browsed products into the neural network.

[0016] Output the influence weight of each identical product feature;

[0017] Utilize the time difference between the user's purchase of goods and the corresponding browsing of goods, as well as the time difference between the purchase of goods and the corresponding browsing of goods.

[0018] The influence weight of the same product features of the browsed products is used to correct the correlation between the purchased products and each browsed product.

[0019] The corrected correlation between the purchased item and each browsed item is obtained.

[0020] Further, the step of obtaining the corrected association between the purchased product and each browsed product includes:

[0021] Normalizing the time difference between the purchase of a product and the corresponding browsed product in inverse proportion to obtain a first correction coefficient of the correlation between the purchase of the product and the browsed product;

[0022] The influence weights of all the same product features of the purchased product and the corresponding browsed product are averaged as a second correction coefficient of the relevance between the purchased product and the browsed product;

[0023] The first correction coefficient and the second correction coefficient of the association between the purchased commodity and each browsed commodity are added together and multiplied by the association between the purchased commodity and the browsed commodity to obtain the corrected association between the purchased commodity and each browsed commodity.

[0024] Furthermore, the step of obtaining the association between the purchased product and each corresponding browsed product includes:

[0025] The relevance between the purchased product and the browsed product is obtained by using the ratio of the number of the same product features of the purchased product and each browsed product to the total number of the product features of the purchased product.

[0026] Further, the steps of obtaining the average browsing frequency of users include:

[0027] Get the maximum time difference between the user's purchase of goods and the corresponding target browsing goods;

[0028] The average browsing frequency of users is obtained by taking the ratio of the number of target browsed products to the maximum time difference.

[0029] Furthermore, the formula for obtaining the purchase probability of the user purchasing this type of product at the current moment is:

[0030]

[0031] Among them, ρ v T represents the purchase probability of the user purchasing the vth category of goods at the current moment; v represents the browsing cycle of the vth category product from the last purchase to the current moment; E v represents the browsing frequency of the vth category of products from the last purchase to the current moment; T represents the user's decision cycle; E represents the user's average browsing frequency; exp() represents an exponential function with the natural constant e as the base.

[0032] Furthermore, the purchase probability of each type of browsed product purchased by the user at the current moment and the number of windows browsing the page are used to determine

[0033] The steps for determining the number of push windows for each category of products on the user's browsing page include:

[0034] The number of push windows for each type of product on the user's browsing page is obtained by multiplying the number of browsed pages by the user's purchase probability of each type of product at the current moment.

[0035] The beneficial effects of the present invention are as follows: the user behavior analysis method based on the big data of online marketing promotion of the present invention corrects the correlation between the purchased goods and each browsed goods through the time difference between the user's purchased goods and the corresponding browsed goods in the historical data, and the influence weight of the same goods feature, and obtains the corrected correlation. Through the corrected correlation, some browsed goods with low correlation are cleaned to obtain the target browsed goods. Compared with the conventional user behavior analysis method, the present invention deeply analyzes the user's browsing behavior and cleans the redundant noise data, so that the analysis results are more in line with the user's actual consumption intention. The user behavior data is used to analyze the user's decision cycle and average browsing frequency, and compared with the real-time browsed goods, and the probability value of the user's purchase behavior corresponding to each type of browsed goods is calculated, and the push window allocation of the user's browsing page is adjusted accordingly, and then the purchase behavior of the real-time browsing data is predicted, and the product push plan is adjusted according to the prediction result, which greatly improves the availability of consumption data. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be 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.

[0037] Figure 1 The present invention is a flowchart of the overall steps of an embodiment of a user behavior analysis method based on online marketing promotion big data. DETAILED DESCRIPTION

[0038] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0039] An embodiment of a user behavior analysis method based on network marketing promotion big data of the present invention is as follows: Figure 1 As shown, the method includes:

[0040] S1. Collect the browsing records of the products browsed by the user to be analyzed on the platform and the purchase records of the products purchased.

[0041] Companies will use marketing platforms to collect user behavior data, analyze user behavior data to mine deep information in the user behavior data, and use it to study implicit feedback recommendation problems. They can provide more accurate implicit feedback recommendations for different customers. For example, the Taobao user behavior dataset collected by Alibaba, each row of the dataset represents a user behavior, which consists of user ID, product ID, product category ID, behavior type and timestamp.

[0042] Specifically, in order to analyze the deep needs of users, this marketing platform is used to collect user behavior data sets of the users to be analyzed on this platform. The user behavior data sets include browsing records of browsed products, purchase records of purchased products, product features of browsed products and purchased products, and browsing time of browsed products and purchased products. Product features refer to the smallest inventory unit of a product, which is composed of multiple sales attributes of a product, such as color, style, and size of clothing, and model, memory, processor, size, and color of a mobile phone. Each sales attribute is a product feature.

[0043] S2. Acquire the common commodity features between the purchased commodity of the user and each corresponding browsed commodity, and use the common commodity features to acquire the association between the purchased commodity and each corresponding browsed commodity.

[0044] Specifically, the number of identical product features between the user's purchased products and each corresponding browsed product, as well as the number of all product features of the purchased products, is obtained, and the relevance between the browsed product and the purchased product is obtained by using the ratio of the number of identical product features to the number of product features of the purchased product. The browsed product refers to all browsed products corresponding to the purchased product between the last purchase of the same type of product and this purchase. The formula for calculating the relevance between the purchased product and each browsed product is:

[0045]

[0046] Among them, Q i represents the correlation between the purchased product and the corresponding i-th browsed product; P i represents the number of common product features between the purchased product and the corresponding i-th browsed product; G o Represents the number of all product features of the oth purchased product corresponding to the i-th browsed product.

[0047] Conventional evaluation of the correlation between two products is to calculate the proportion of the overlapping parts between the characteristics of the two products. However, two browsed products with the same correlation may also have different time sequences. That is, for two browsed products with the same correlation with the purchased product, the one that is closer to the purchased product in time sequence has a more direct impact on the final purchase decision. In addition, different influence weights of the same product characteristics have different effects on purchasing behavior.

[0048] Specifically, the time difference between the purchase of a product and the corresponding browsed product is normalized in inverse proportion to obtain a first correction coefficient of the association between the purchase of the product and the browsed product. The formula for calculating the first correction coefficient is:

[0049]

[0050] in, The first correction coefficient representing the correlation between the purchased product and the corresponding i-th browsed product; t i represents the time point of browsing the i-th product; t o represents the time point of the oth purchased product corresponding to the i-th browsed product, (t o -t i ) represents the time difference between the ith browsed product and the corresponding oth purchased product, and the unit of the time difference is days; exp() represents an exponential function with the natural constant e as the base, and the time difference is inversely normalized using the exponential function, which represents the time correlation between the browsed products and the corresponding purchased products in the purchase decision-making process, that is, the closer the browsed products are to the purchased products in time sequence, the greater the probability of influencing the purchase decision, that is, the smaller the time difference, the greater the influence.

[0051] There will be multiple correlations between browsed items and purchased items in the user's behavior data set, that is, the browsed items and purchased items have the same product features, but the features of the same parts are divided into universal and specific. For example, the size and type features of the products overlap a lot but are relatively common and not unique, while the color and style of the products have the user's own unique preference attributes. Therefore, even the same features may have universal features and unique features, so it is necessary to assign influence weights to different features.

[0052] The specific method of obtaining the influence weight of product features is as follows: collect product features of each category of products on this marketing platform, i.e., mobile phones, clothing, supplies, etc., divide the collected data into training set and validation set at a ratio of 7:3, and manually assign different influence weights to all product feature information in the training set, with the weight value between 0-1. Then, train the neural network using the product feature information that has been assigned influence weights. The neural network structure is Encoder-FC, and the loss function is the cross entropy function. Until the loss function converges, input the validation set data for verification, and complete the classification network training. Using the trained classification network, the influence weights can be automatically output for the product features of all categories of products for which purchases have occurred.

[0053] Specifically, each identical product feature of the purchased product and the corresponding browsed product is input into the neural network, and the influence weight of each identical product feature is output. The influence weights of all identical product features of the purchased product and the corresponding browsed product are averaged as the second correction coefficient of the association between the purchased product and the browsed product, and the second correction coefficient is calculated using the following formula:

[0054]

[0055] in, The second correction coefficient representing the correlation between the purchased product and the corresponding i-th browsed product; β s represents the influence weight of the sth common product feature between the purchased product and the corresponding i-th browsed product; P i It represents the number of identical product features between the purchased product and the corresponding i-th browsed product; that is, the influence weights of the identical product features of the browsed product and the purchased product are accumulated and averaged as the second correction coefficient.

[0056] The first correction coefficient and the second correction coefficient of each browsed commodity are used to correct the correlation between the browsed commodity and the corresponding purchased commodity to obtain the corrected correlation between the purchased commodity and the browsed commodity. The formula for calculating the corrected correlation is:

[0057]

[0058] Among them, ω i represents the corrected correlation between the purchased product and the corresponding i-th browsed product; Q i represents the association between the purchased product and the corresponding i-th browsed product; a first correction coefficient representing the association between a purchased product and the corresponding i-th browsed product; The second correction coefficient representing the correlation between the purchased product and the corresponding i-th browsed product, Representatives A norm calculation is performed, that is, the association between each purchased commodity and its corresponding browsed commodity is corrected respectively by using the first correction coefficient and the second correction coefficient.

[0059] S3. Sort the relevance between the purchased product and each corresponding browsed product by size to obtain a relevance sequence, obtain the first sudden relevance in the relevance sequence, and use all browsed products with a relevance greater than the sudden relevance as target browsed products for the purchased product.

[0060] For the collected data set, the browsed items between each purchase behavior and the previous purchase behavior are considered to be the decision-making process that leads to the purchase and are counted as a decision cycle. However, the browsed items that affect the purchase behavior are divided into targeted and random occurrences. After determining the relevant browsing behaviors of a certain type of goods to be purchased and the random browsing of various types of goods, the browsing behaviors that make purchase decisions are determined. That is, it is necessary to clean the browsed item data of users in the decision-making process and remove the randomly browsed items to obtain browsing data with decision-making influence, so as to obtain accurate user purchasing habits.

[0061] For browsing data, we do not consider cleaning duplicate data, but only cleaning noise data. Noise data refers to data that has low similarity with purchasing behavior and has no influence on decision-making.

[0062] Specifically, the corrected correlations between purchased items and corresponding browsed items are arranged from large to small to obtain a correlation sequence for each purchased item. The Mann-Kendall mutation detection algorithm is used to detect mutation points in the correlation sequence, and the first mutation point is marked. The correlation corresponding to the first mutation point is used as the mutation correlation, and the browsed items corresponding to the correlation greater than the mutation correlation are used as target browsed items with decision-making influence, and the target browsed items are restored to the time sequence.

[0063] S4. Utilize the maximum time difference between each purchased product of the user and the corresponding target browsed product to obtain the user's decision cycle; utilize the time difference between each purchased product of the user and the corresponding target browsed product and the number of target browsed products to obtain the user's average browsing frequency.

[0064] Specifically, the user's decision cycle and average browsing frequency are calculated according to the following formula:

[0065]

[0066] Among them, t o represents the time point of the oth purchase of the product, It indicates the time point of the target browsing the product with the longest time from the oth product purchase in this decision cycle. represents the actual decision cycle of the oth purchased commodity, M represents the number of all purchased commodities in the historical data, and the actual decision cycles of all purchased commodities in the historical data are accumulated and averaged to obtain the decision cycle of the user's purchasing habits; G oRepresents the number of target browsed products in the actual decision cycle of the oth purchase behavior; the browsing frequency is obtained by dividing the number of target browsed products in the decision cycle by the decision cycle. Similarly, the historical data is accumulated and averaged to obtain the average browsing frequency of the user's purchasing habits, which also represents the user's purchasing decision-making power. That is, the smaller the average browsing frequency, the greater the purchasing decision-making power.

[0067] S5. Cluster all browsed products from the last time the user purchased a product to the current moment according to product characteristics to obtain multiple categories of products, and use the browsing time and browsing quantity of each category of browsed products to obtain the browsing cycle and browsing frequency of each category of products; use the browsing cycle and browsing frequency of each category of browsed products by the user as well as the user's decision cycle and average browsing frequency to obtain the purchase probability of each category of products by the user at the current moment.

[0068] Specifically, all browsed products from the last time the user purchased a product to the current moment are clustered according to product characteristics to obtain multiple categories of products, and the time difference between the browsed product with the longest time from the current moment to the current moment among all browsed products of each category of products is taken as the browsing cycle of this category of products; the browsing frequency of this category of products is obtained by using the ratio of the number of all browsed products of each category of products to the browsing cycle.

[0069] The purchase probability of each type of product by the user at the current moment is calculated according to the following formula:

[0070]

[0071] Among them, ρ v T represents the purchase probability of the user purchasing the vth category of goods at the current moment; v represents the browsing cycle of the vth category product from the last purchase to the current moment; E v represents the browsing frequency of the vth category product from the last purchase to the current moment; T represents the user's decision cycle; E represents the user's average browsing frequency; exp() represents an exponential function with the natural constant e as the base; They represent the ratio of the browsing cycle of the vth category of products browsed by the user from the last purchase to the current moment to the user's decision cycle, and the ratio of the browsing frequency of the vth category of products to the average browsing frequency. Represents the calculation of the L2 norm of the two. Theoretically, when When both are 1, it best fits the user's decision cycle and average browsing frequency, and the probability of purchasing the vth category of goods is the highest. The theoretical optimum value is represent Theoretical optimum value The smaller the value, the more closely the current browsing cycle and browsing frequency of the v-th category product fit the user's decision cycle and average browsing frequency. The exponential function exp is used to normalize the logic inversely. The smaller the The larger the value between 0 and 1, the greater the probability of purchasing the vth category of goods.

[0072] S6. Determine the number of push windows for each type of product on the user's browsing page by using the purchase probability of the user purchasing each type of browsed product at the current moment and the number of windows on the browsing page.

[0073] Specifically, the browsing page of the marketing platform can accommodate multiple product push windows, and the multiple product push windows are allocated according to the user's purchase probability of each type of product. The push window for each type of product is calculated according to the following formula:

[0074]

[0075] Among them, ρ v H represents the purchase probability of the user purchasing the vth category of goods at the current moment, v Indicates the number of push windows for the vth category of products on the user's browsing interface. Represents the round-up symbol, which means that the higher the purchase probability, the more push windows there are for this type of product on the current page.

[0076] In summary, the present invention provides a user behavior analysis method based on big data of online marketing promotion. The correlation between the purchased goods and each browsed goods is corrected by the time difference between the user's purchased goods and the corresponding browsed goods in the historical data, and the influence weight of the same goods feature, and the corrected correlation is obtained. The corrected correlation is used to clean up some browsed goods with low correlation to obtain the target browsed goods. Compared with the conventional user behavior analysis method, the present invention deeply analyzes the user's browsing behavior and cleans up the redundant noise data, so that the analysis results are more in line with the user's actual consumption intention. The user's decision cycle and average browsing frequency are analyzed by using the user behavior data, and compared with the real-time browsed goods, and the probability value of the user's purchase behavior corresponding to each type of browsed goods is calculated, and the push window allocation of the user's browsing page is adjusted accordingly, and then the purchase behavior of the real-time browsing data is predicted, and the product push plan is adjusted according to the prediction results, which greatly improves the availability of consumption data.

[0077] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A user behavior analysis method based on online marketing promotion big data, characterized by: Collect the browsing history of the products browsed by the users to be analyzed on the platform and the purchase history of the products purchased; Acquire the same commodity features between the user's purchased commodity and each corresponding browsed commodity, and use the same commodity features to acquire the relevance between the purchased commodity and each corresponding browsed commodity; Sort the relevance between the purchased product and each corresponding browsed product by size to obtain a relevance sequence, obtain the first mutation relevance in the relevance sequence, and use all browsed products with relevance greater than the mutation relevance as target browsed products for the purchased product; The maximum time difference between each purchased product of the user and the corresponding target browsed product is used to obtain the user's decision cycle; the time difference between each purchased product of the user and the corresponding target browsed product and the number of target browsed products are used to obtain the user's average browsing frequency; Cluster all browsed products from the last time the user purchased a product to the current moment according to product characteristics to obtain multiple categories of products, and use the browsing time and browsing quantity of each category of browsed products to obtain the browsing cycle and browsing frequency of each category of products; The user's browsing cycle, browsing frequency, decision cycle and average browsing frequency of each category of browsing products are used to obtain the purchase probability of each category of products purchased by the user at the current moment; The number of push windows for each category of products on the user's browsing page is determined by using the purchase probability of the user purchasing each category of browsed products at the current moment and the number of windows on the browsing page; Before sorting the relevance between each purchased product and the corresponding browsed product by size, the relevance between the purchased product and each corresponding browsed product is corrected, and the correction method includes: Construct a neural network, input each identical product feature of the purchased product and the corresponding browsed product into the neural network, and output the influence weight of each identical product feature; The correlation between the purchased goods and each browsed goods is corrected by using the time difference between the user's purchase of goods and each corresponding browsed goods, as well as the influence weights of the same product features of the purchased goods and the corresponding browsed goods, to obtain the corrected correlation between the purchased goods and each browsed goods.

2. According to claim 1, a user behavior analysis method based on network marketing promotion big data is characterized in that: The steps of obtaining the corrected association between the purchased item and each browsed item include: Normalizing the time difference between the purchase of a product and the corresponding browsed product in inverse proportion to obtain a first correction coefficient of the correlation between the purchase of the product and the browsed product; The influence weights of all the same product features of the purchased product and the corresponding browsed product are averaged as a second correction coefficient of the relevance between the purchased product and the browsed product; The first correction coefficient and the second correction coefficient of the association between the purchased commodity and each browsed commodity are added together and multiplied by the association between the purchased commodity and the browsed commodity to obtain the corrected association between the purchased commodity and each browsed commodity.

3. The user behavior analysis method based on network marketing promotion big data according to claim 1 is characterized in that: The steps of obtaining the association between the purchased product and each corresponding browsed product include: The relevance between the purchased product and the browsed product is obtained by using the ratio of the number of the same product features of the purchased product and each browsed product to the total number of the product features of the purchased product.

4. The user behavior analysis method based on network marketing promotion big data according to claim 1 is characterized in that: The steps to obtain the average browsing frequency of users include: Get the maximum time difference between the user's purchase of goods and the corresponding target browsing goods; The average browsing frequency of users is obtained by taking the ratio of the number of target browsed products to the maximum time difference.

5. The user behavior analysis method based on network marketing promotion big data according to claim 1 is characterized in that: The formula for obtaining the purchase probability of the user purchasing this type of product at the current moment is: Among them, ρ v T represents the purchase probability of the user purchasing the vth category of goods at the current moment; v represents the browsing cycle of the vth category product from the last purchase to the current moment; E v represents the browsing frequency of the vth category of products from the last purchase to the current moment; T represents the user's decision cycle; E represents the user's average browsing frequency; exp() represents an exponential function with the natural constant e as the base.

6. The user behavior analysis method based on network marketing promotion big data according to claim 1 is characterized in that: The step of determining the number of push windows for each category of products on the user's browsing page by using the purchase probability of the user purchasing each category of browsed products at the current moment and the number of windows on the browsing page comprises: The number of push windows for each type of product on the user's browsing page is obtained by multiplying the number of browsed pages by the user's purchase probability of each type of product at the current moment.

Images