Marketing method and device based on user behavior analysis, equipment and medium

Abstract

The application discloses a marketing method and device based on user behavior analysis, equipment and medium, relates to the technical field of network marketing, and the method comprises the following steps: acquiring first user behavior data of a user in a first preset historical time, preprocessing the first user behavior data to acquire second user behavior data; acquiring a first target category of the user by category identification and interest analysis on the second user behavior data; acquiring a target delivery time of the user according to the second user behavior data; and delivering a product of the first target category to the user at the target delivery time. The application can adapt to the interest changes of the user in different periods and improve the accuracy of product delivery.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of network marketing, and in particular to a marketing method and device based on user behavior analysis, equipment and medium. BACKGROUND

[0002] Due to the rise of online shopping, e-commerce platforms can track and analyze users' browsing, searching, clicking, collecting, adding to shopping carts, and purchasing behaviors, etc., to deeply understand consumers' preferences and needs. User behavior data has become an important basis for enterprises to optimize their operation strategies.

[0003] In the current booming Internet marketing, consumers' behavior trajectories, interest preferences, and decision-making paths have been fully digitized. Users have increasingly strong demands for personalized and timely marketing content. With the continuous rise in traffic costs and the increasing homogeneity of competition, unprecedented high standards are required for the precision and personalization of marketing.

[0004] In the prior art, users are analyzed based on a single dimension, such as relying too much on static user portraits and relying too much on label association rules, which cannot fully explore the characteristics and demand differences of different user groups, making it difficult to achieve targeted marketing. SUMMARY

[0005] The purpose of the present application is to provide a marketing method and device based on user behavior analysis, equipment and medium, which can adapt to the changes in users' interests at different times and improve the precision of product placement.

[0006] To achieve the above-mentioned purpose, the present application provides the following solutions:

[0007] In a first aspect, the present application provides a marketing method based on user behavior analysis, comprising:

[0008] Obtaining first user behavior data of a user within a first preset historical time, preprocessing the first user behavior data to obtain second user behavior data;

[0009] Obtaining a first target category of the user by category recognition and interest analysis on the second user behavior data;

[0010] Obtaining a target placement time of the user according to the second user behavior data;

[0011] Placing products of the first target category to the user at the target placement time.

[0012] Optionally, the obtaining of the first target category of the user by category recognition and interest analysis on the second user behavior data comprises:

[0013] According to the second user behavior data, a purchase behavior of a user is recognized to obtain a second target category, and a preset lower category corresponding to the second target category is taken as a first candidate category; the second target category is one of preset upper categories; the preset upper categories include one or more preset lower categories;

[0014] A user interest analysis is performed on the first candidate category to obtain a first target category, and the first target category is one of the first candidate categories.

[0015] Optionally, the user interest analysis on the first candidate category to obtain the first target category comprises:

[0016] If the number of the first candidate categories is greater than 1 and less than a first preset number: according to the second user behavior data, a delivery content score of the first candidate category is obtained, and the first target category is taken as one with the highest delivery content score in the first candidate category; the first preset number is greater than 1,

[0017] If the number of the first candidate categories is greater than the first preset number:

[0018] According to an exponential decay model, an interest value of the first candidate category is obtained, and the first preset number of the first candidate categories with the highest interest value is taken as second candidate categories;

[0019] According to the second user behavior data, a delivery content score of the second candidate category is obtained, and the first target category is taken as one with the highest delivery content score in the second candidate category.

[0020] Optionally, the purchase behavior of the user is recognized according to the second user behavior data to obtain the second target category, and the method comprises:

[0021] According to the second user behavior data, a first feature is extracted; the first feature includes explicit behavior data and implicit behavior data corresponding to each preset upper category;

[0022] A pattern recognition of the first feature is performed through a category recognition model to obtain the first target category;

[0023] The category recognition model is obtained in the following manner:

[0024] Second features of a plurality of users in a historical time and corresponding first preset labels are obtained; the data category of the second feature is the same as the type of the first feature, and the first preset label is used to indicate the preset upper category;

[0025] The second feature and the corresponding first preset label are taken as a first training set to train a first preset classification model to obtain the category recognition model.

[0026] Optionally, the target delivery time of the user is obtained according to the second user behavior data, including:

[0027] The first timestamp behavior feature of the user in a second preset historical time is obtained according to the second user behavior data;

[0028] The first timestamp behavior feature is processed by an active time recognition model to obtain the target delivery time;

[0029] The obtaining method of the active time recognition model includes:

[0030] The second timestamp behavior feature corresponding to a plurality of users in the third preset historical time and the corresponding second preset label are obtained, wherein the second preset label is used to indicate a plurality of preset time regions in a day;

[0031] The second timestamp behavior feature and the second preset label are taken as a second training set to train a second preset classification model to obtain the active time recognition model.

[0032] Optionally, the target delivery time is obtained by processing the first timestamp behavior feature by an active time recognition model, including:

[0033] The first timestamp behavior feature is input into the active time recognition model to obtain an active probability of the preset time region, and the preset time region with an active probability greater than a first preset threshold is selected as a first candidate preset time region;

[0034] The first target category in the first candidate preset time region corresponds to the competitor reach density not greater than a second preset threshold and the highest active probability, and one item is obtained as a second candidate time region;

[0035] The second candidate time region is adjusted according to a preset weather rule to obtain the target delivery time.

[0036] Optionally, the marketing method based on user behavior analysis includes:

[0037] The active time recognition model is monitored and updated, specifically including:

[0038] The early warning feature of the user is obtained in real time, and the early warning feature includes the click rate, conversion rate, complaint rate and ROI index of the product corresponding to the first target category;

[0039] If the early warning feature meets the preset early warning rule, the acquisition method of the active time recognition model is repeated to update the active time recognition model.

[0040] In a second aspect, the present application provides a marketing device based on user behavior analysis, comprising:

[0041] An acquisition module is configured to acquire first user behavior data of a user in a first preset historical time, and preprocess the first user behavior data to obtain second user behavior data.

[0042] An analysis module is configured to:

[0043] The analysis module is further configured to:

[0044] The analysis module is further configured to:

[0045] A delivery module is configured to deliver a product of the first target category to the user at the target delivery time.

[0046] In a third aspect, the present application provides a computer device, comprising a memory, a processor, a computer program stored in the memory and executable on the processor, and the processor executes the computer program to implement the steps of the marketing method based on user behavior analysis according to any one of the above.

[0047] In a fourth aspect, the present application provides a computer readable storage medium having a computer program stored thereon, and the computer program is executed by a processor to implement the steps of the marketing method based on user behavior analysis according to any one of the above.

[0048] In a fifth aspect, the present application provides a computer program product comprising a computer program, and the computer program is executed by a processor to implement the steps of the marketing method based on user behavior analysis according to any one of the above.

[0049] According to the embodiments of the present application, the following technical effects are achieved:

[0050] The present application provides a marketing method, device, equipment and medium based on user behavior analysis, which can accurately determine the category of interest of the user by analyzing the recent behavior of the user, identifying the first target category of the user through category recognition and interest analysis, and combining the target delivery time obtained from the user behavior data, so as to dynamically adapt to the interest changes and active time changes of the user in different periods, and improve the pertinence and accuracy of the marketing strategy. BRIEF DESCRIPTION OF DRAWINGS

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A flowchart illustrating a marketing method based on user behavior analysis, provided as an embodiment of this application;

[0053] Figure 2 for Figure 1 A detailed flowchart of step 102;

[0054] Figure 3 for Figure 2 A detailed flowchart illustrating step 202 in the middle section;

[0055] Figure 4 for Figure 2 A detailed flowchart of step 201;

[0056] Figure 5 for Figure 1 A detailed flowchart of step 103;

[0057] Figure 6 A schematic flowchart illustrating a model monitoring and updating method provided in an embodiment of this application;

[0058] Figure 7 for Figure 5 A detailed flowchart of step 502;

[0059] Figure 8 A schematic diagram of the functional modules of a marketing device based on user behavior analysis provided in an embodiment of this application;

[0060] Figure 9 This is a schematic diagram of the structure of a computer device provided in an embodiment of this application. Detailed Implementation

[0061] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0062] It should be noted that the terms "first", "second", "third", "fourth" and the like in the description, claims, and drawings of the application, if any, are used only to distinguish between similar objects and do not necessarily have to describe a particular sequential or chronological order. It should be understood that the use of such terms is interchangeable under appropriate circumstances such that the embodiments of the application described herein are capable of operation in other sequences than those illustrated or otherwise described herein.

[0063] The above objects, features and advantages of the present application will become more apparent from the following detailed description of the application when taken in conjunction with the accompanying drawings and specific embodiments.

[0064] In one exemplary embodiment, as shown in Figure 1 a user behavior analysis-based marketing method is provided, including the following steps 101 to 104. Among them:

[0065] Step 101, obtaining first user behavior data of a user within a first preset historical time, and preprocessing the first user behavior data to obtain second user behavior data;

[0066] It should be noted that the user information (including but not limited to user equipment information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in the present application are all information and data authorized by the user or authorized by all parties, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards, and provide corresponding operation portal for user to choose authorization or refusal.

[0067] Specifically, each piece of data in the first user behavior data includes the user's behavior (such as clicking, adding to cart, collecting, browsing), the time point of the behavior, the product affected by the behavior, the upper and lower categories corresponding to the product, etc.

[0068] Explicit purchase behavior, implicit purchase behavior, and user environment data available on the platform;

[0069] Specifically, the first user behavior data can be obtained through API interface, SDK point embedding, and heterogeneous data access of third-party data sources.

[0070] Specifically, the preprocessing step includes data cleaning, data standardization, data desensitization, and the like.

[0071] Further, the preprocessing steps are as follows: set invalid data filtering rules to filter out browsing records with a dwell time of less than 1 second and repeated clicks; at the same time, through normalization mapping technology, such as Min-Max standardization, heterogeneous data (such as text search terms and numerical dwell time) are converted into a unified format; data anonymization technology is used to process sensitive user information (such as ID card number and mobile phone number) to protect user privacy; and a multi-dimensional behavior dataset is constructed as the second user behavior data.

[0072] Step 102: Obtain the user's first target category from the second user behavior data through category identification and interest analysis;

[0073] Specifically, the first target category is one of the preset categories, obtained through processing and interest analysis of the second set of user behavior data. The first target category is the category of the product to be advertised.

[0074] Step 103: Obtain the user's target delivery time based on the second user behavior data;

[0075] Specifically, based on time-related data in the second user behavior data, the time periods in which users are active are obtained for product deployment.

[0076] Step 104: Deliver products of the first target category to users during the target delivery time;

[0077] By implementing steps 101 to 104 above, the user's first target category can be obtained through category identification and interest analysis, which can accurately determine the categories that the user is interested in. Combined with the target delivery time obtained from the second user behavior data, the targeting and accuracy of the marketing strategy are improved.

[0078] In another exemplary embodiment of this application, in order to accurately address the personalized needs and differences of different users and improve the accuracy of determining the first target category, a multi-level optimization strategy combining category identification and interest analysis is provided, such as... Figure 2 As shown, step 102 above is replaced by steps 201 to 202:

[0079] Step 201: Based on the second user behavior data, identify the user's purchase behavior to obtain the second target category, and take the preset sub-category corresponding to the second target category as the first candidate category; wherein, the second target category is one of the preset parent categories; the preset parent category includes one or more preset sub-categories;

[0080] Specifically, the preset upper category and the corresponding preset lower category are product categories obtained according to an e-commerce platform. Some e-commerce platforms are three-level categories or more. The upper and lower categories can be selected according to the number and category to be preset upper and lower categories, or the personalized preset upper and lower categories can be reasonably combined and set.

[0081] Specifically, the second target category can be obtained by identifying the category according to the identification model. The second target category is one of the preset upper categories, and the preset lower category corresponding to the second target category is taken as the first candidate category. The following steps further filter the first candidate category to obtain the first target category. The first target category obtained by the embodiment of the application is the preset lower category.

[0082] Step 202, performing user interest analysis on the first candidate category to obtain the first target category, the first target category being one of the first candidate categories.

[0083] In another exemplary embodiment of the application, in order to dynamically analyze the interest of the user according to the personalized needs of the user, as shown in Figure 3 The above step 202 is replaced by steps 301-305.

[0084] Step 301, performing range judgment on the number of the first candidate category. If the number of the first candidate category is greater than the first preset number, steps 302-303 are executed. If the number of the first candidate category is within the range of 2 to the first preset number, step 304 is executed. If the number of the first candidate category is unique, step 305 is executed.

[0085] Optionally, the first preset number can be set to 3-5.

[0086] Step 302, obtaining the interest value of the first candidate category according to an exponential decay model, and taking the first preset number of the first candidate category with the highest interest value as the second candidate category.

[0087] Specifically, the exponential decay model is:

[0088]

[0089] wherein, is the interest value of the i-th behavior type of a certain first candidate category, is the behavior weight of the i-th behavior type, is the decay coefficient, is the time of the behavior of the i-th behavior type of the first candidate category. i

[0090] ​Further, the behavior type includes a purchase behavior, an add-to-cart behavior, and a deep browsing behavior (a behavior of browsing for more than a first preset time, which is generally set to 1-2 minutes, and preferably set to 1 minute.

[0091] Further, the weight of the purchase behavior can be preset to 4.0-6.0, and the decay coefficient can be preset to 0.005-0.05, the weight of the add-to-cart behavior can be preset to 2.5-4.0, and the decay coefficient can be preset to 0.003-0.03, and the weight of the deep browsing behavior can be preset to 1.0-2.5, and the decay coefficient can be preset to 0.005-0.015.

[0092] As a preferred embodiment, Table 1 is a table of a behavior type and preset parameters.

[0093]

[0094] As another preferred embodiment, Table 2 is another table of a behavior type and preset parameters.

[0095] Optionally, the behavior type further includes a dynamic adjustment type, and the dynamic adjustment type includes a user behavior of a time-limited discount product, etc.

[0096] For example, during a large promotion (such as 618), in addition to the purchase behavior, the add-to-cart behavior, and the deep browsing behavior, the user behavior of the time-limited discount product and the corresponding weight and decay coefficient can be artificially added in the background to update the exponential decay model.

[0097] As an exemplary embodiment, the method is used to provide a marketing strategy for a user on a shopping platform, for example, the first candidate category obtained after step 301 includes “outdoor sports”, and the interest value corresponding to “outdoor sports” is calculated, and the behavior data corresponding to “outdoor sports” obtained includes:

[0098]

[0099] 2025-06-01 10:15, a pair of mountaineering shoes is purchased (behavior type: purchase, );

[0100] 2025-06-03 16:30, an outdoor backpack is added to the cart (behavior type: add-to-cart, );

[0101] 2025-06-05 09:45, a piece of outdoor jacket is browsed for 3 minutes (behavior type: deep browsing, );

[0102] The current calculation time ;

[0103] Interest value calculation:

[0104] Calculate the interest contribution of each behavior to the "outdoor sports" category respectively:

[0105] Purchase behavior: (time difference 99.75 hours);

[0106] Add to cart behavior: (time difference 45.5 hours);

[0107] Browsing behavior: (time difference 4.25 hours);

[0108] Then the interest value .

[0109] Step 303, according to the second user behavior data, obtain the delivery content score of the second candidate category, and take the one with the highest delivery content score in the second candidate category as the first target category.

[0110] Specifically, the formula for obtaining the delivery content score is:

[0111]

[0112] Wherein, is the delivery content score of a certain second candidate category, , , is the weight coefficient, is the number of times sent in the content within the preset number of days, and its reciprocal is the content rarity, CTR is the historical click rate of the behavior corresponding to the second candidate category, and CVR is the conversion rate of the behavior corresponding to the second candidate category, which can be obtained through the second user behavior data.

[0113] Optionally, the preset number of days can be set to 5-10 days, preferably 7 days.

[0114] As an exemplary embodiment, taking a certain user of a certain tourism e-commerce as an example, the second candidate category obtained according to the above steps includes platinum card air mileage doubling class (CTR=12%, CVR=8%, sent 3 times in the past 7 days), high-end hotel stay two-sending one class (CTR=9%, CVR=10%, sent 2 times in the past 7 days); Points exchange luxury goods class (CTR=7%, CVR=5%, sent 5 times in the past 7 days);

[0115] Then according to this step, the delivery content score of the platinum card air mileage doubling class is: 0.35x12%+0.4x8%+0.15x(1 / 3)+0.1x1.2=0.042+0.032+0.05+0.12=0.244;

[0116] The launch content score of the high-end hotel stay two and send one type is: 0.35x9%+0.4x10%+0.15x(1 / 2)+0.1x1.2=0.0315+0.04+0.075+0.12=0.2665;

[0117] The launch content score of the luxury product type is: 0.35x7%+0.4x5%+0.15x(1 / 5)+0.1x1.2=0.0245+0.02+0.03+0.12=0.1945.

[0118] Specifically, the first target category is finally determined as the high-end hotel stay two and send one type. According to the product launch method provided in the application, the user click conversion rate finally reaches 11.2%, which is 37% higher than the group average.

[0119] Step 304, obtaining the launch content score of the first candidate category according to the second user behavior data, and taking the item with the highest launch content score in the first candidate category as the first target category;

[0120] Specifically, this step is similar to step 303, which will not be described here.

[0121] Step 305, taking the first candidate category as the first target category.

[0122] In another exemplary embodiment of the application, in order to accurately obtain the first candidate category, as shown in the following steps 401~402 replace the above step 201: Figure 4

[0123] Step 401, obtaining the first feature according to the second user behavior data; wherein the first feature includes explicit behavior data and implicit behavior data corresponding to each preset superior category.

[0124] Specifically, the explicit behavior includes the frequency and time of browsing behavior, the frequency and time of purchase behavior, the frequency and time of adding to cart behavior, and the number and time of collection behavior.

[0125] Specifically, the implicit behavior data includes search hot words and page stay time and sliding speed when browsing, purchasing, adding to cart and collecting respectively.

[0126] Optionally, the second user behavior data includes multiple dimension labels, such as the frequency and time of browsing, purchasing, adding to cart and collecting, etc., and the first feature can be extracted through label filtering;

[0127] Step 402, obtaining the first target category by pattern recognition of the first feature through a category recognition model;

[0128] ​Specifically, the obtaining manner of the category recognition model comprises:

[0129] obtaining second features of a plurality of users in a historical time and corresponding first preset labels, the data category of the second features being the same as the type of the first features, and the first preset labels being used for indicating preset superior categories;

[0130] training the first preset classification model by taking the second features and the corresponding first preset labels as a first training set to obtain the category recognition model.

[0131] As an optional implementation manner, the first preset classification model selects a LightGBM model (Light Gradient Boosting Machine), and the LightGBM model comprises a self-provided important feature screening module, which can perform feature evaluation on the input features and output the importance of each feature in the training process. The output category recognition model is a classification model based on important features.

[0132] As another optional implementation manner, before step 402, the following step is further included: performing feature screening on the first features to obtain updated first features.

[0133] Optionally, the variance screening method can be used for feature screening, that is, for each first feature, the variance of each feature is calculated, and the new first features are selected by taking the variance greater than a preset variance value.

[0134] Specifically, the variance calculation formula is:

[0135]

[0136] wherein, is the number of samples, is the feature value of the i-th sample, is the average value of the feature value, is the feature variance.

[0137] In another exemplary embodiment of the present application, in order to realize accurate product delivery time, user behavior is analyzed to find out the user's habit of reading time (target delivery time in the following) for product delivery, as shown in FIG. 5, then the above step 103 is replaced by the following steps 501~step 502: Figure 5

[0138] Step 501: obtaining first timestamp behavior features of a user in a second preset historical time according to second user behavior data;

[0139] ​​Specifically, the first timestamp behavior feature includes a behavior type feature in a second preset historical time, a timestamp corresponding to the behavior type, and an active duration corresponding to the behavior type.

[0140] Specifically, the second preset historical time is one day.

[0141] In step 502, the first timestamp behavior feature is processed by the active time recognition model to obtain a target delivery time.

[0142] Specifically, the target delivery time is a delivery time region in a first preset future time. Generally, the first preset future time is set to one day.

[0143] Further, the obtaining method of the active time recognition model includes:

[0144] The second timestamp behavior feature and the corresponding second preset label of a plurality of users in a third preset historical time are obtained, and the second preset label is used to indicate a plurality of preset time regions in one day.

[0145] The second timestamp behavior feature and the second preset label are used as a second training set to train the second preset classification model to obtain the active time recognition model.

[0146] Specifically, the third preset historical time can be set to one month, the second timestamp behavior feature includes a behavior type feature in the third preset historical time, a timestamp corresponding to the behavior type, and an active duration corresponding to the behavior type; the first timestamp behavior feature and the second timestamp behavior feature are of the same format and type.

[0147] As an implementation manner, each hour is set as a time region, and 24 preset time regions are set.

[0148] As an implementation manner, an LSTM (Long Short-Term Memory, long short-term memory network) model is set as the second preset model.

[0149] Further, the loss function used by the LSTM model in training is set as:

[0150]

[0151] wherein, is the number of samples, is the number of time steps, is the input feature, is the active probability predicted by the model.

[0152] Further, the LSTM model uses the SGD algorithm (Stochastic Gradient Descent) to update the model parameters during training, and the formula is as follows:

[0153]

[0154] wherein, is the model parameter, is the learning rate, is the gradient of the loss function with respect to the parameter.

[0155] In another exemplary embodiment of the present application, in order to ensure the accuracy of the model, as shown in Figure 6 The present application also provides a model monitoring and updating method, comprising the following steps:

[0156] Step 601, real-time acquisition of user early warning features, the early warning features including one or more of the click rate, conversion rate, complaint rate and ROI index of the product corresponding to the first target category;

[0157] Step 602, if the early warning features meet the preset early warning rules, repeat the acquisition method of the active time recognition model to update the active time recognition model.

[0158] Optionally, the preset early warning rule can be set to be greater than or less than the corresponding threshold value, or the early warning features are within a certain preset range, and the time when the early warning features meet the threshold condition can also be set, for example: the click rate decreases by >20% for 3 consecutive days.

[0159] In another exemplary embodiment of the present application, in order to ensure the accuracy of the model, the active time recognition model can be incrementally learned within a certain period.

[0160] Specifically, the random gradient descent SGD algorithm is used to update the interest model parameters daily, and the new strategy effect is verified by A / B testing (allocating 5% traffic) (shuffling according to user ID hash), and after t-test (p<0.05), the full deployment is carried out.

[0161] In another exemplary embodiment of the present application, in order to realize the dynamic optimization of the target delivery time, as shown in Figure 7 The above step 502 can be replaced by the following steps 701~703:

[0162] Step 701, inputting the first timestamp behavior feature into the active time recognition model to obtain the active probability of the preset time region, and screening the preset time region with an active probability greater than a first preset threshold as a first candidate preset time region;

[0163] Step 702, obtaining a first candidate time region in which the first target category corresponding to the competitor reach density is not greater than the second preset threshold and the active probability is the highest as the second candidate time region;

[0164] Specifically, the reach density of the first target category corresponding to the competitor platform marketing in the first candidate period is obtained through public opinion monitoring, and the first candidate preset time region with the highest active probability is judged. If the competitor reach density in the same period exceeds the threshold (such as > 5 times / user), the suboptimal active period is selected for further judgment until the condition is met (such as the second highest probability in the candidate window 17:00-18:00).

[0165] Step 703, adjusting the second candidate time region according to the preset weather rule to obtain the target delivery time.

[0166] Specifically, the weather rule can be set to optimize the second candidate time region.

[0167] Optionally, the preset weather rule includes: according to the weather change, the second candidate time region is advanced or delayed by one or more time regions as the target delivery region;

[0168] For example, if there is thunderstorm, the time region is advanced by one time region.

[0169] The embodiment of the application provides a marketing method based on user behavior analysis, which has the following technical effects:

[0170] The scheme adopts a three-stage multi-level screening processing method of "category identification → category interest value screening → content score value screening". First, the upper category is accurately determined through category identification, then the user behavior (such as browsing, collecting, purchasing, etc.) weight analysis is used to screen out multiple lower categories that the user is interested in as candidate categories, and finally the conversion rate, content rarity and other business indicators are combined for secondary sorting, layer-by-layer filtering, effectively narrowing the target range and improving the relevance and value of the final delivery category.

[0171] The LightGBM model is used for multi-category identification, which fully utilizes its advantages in processing high-dimensional sparse features, has strong non-linear fitting ability and classification accuracy, and can quickly and accurately identify the categories that the user may be interested in from massive user behavior data. The LightGBM feature importance module is used to automatically quantify the contribution of each input feature to the category identification result. This not only helps to understand the model decision logic, but also provides data support for subsequent optimization of feature engineering.

[0172] In the category interest value screening stage, different weights are assigned to different categories according to the behavior types, and the behavior types cover high value and long-term impact, medium value and medium-term impact, and low value but short-term effective impact, and appropriate weights are assigned to obtain a comprehensive "interest value". The assignment of weights overcomes the contingency of single behavior data, more comprehensively reflects user preferences, and enhances the stability and credibility of the screening results.

[0173] The method guarantees the accuracy of user personalized recommendation, and further screens through the interest content score, realizes the upgrade from "general recommendation" to "precise selection", and significantly improves the user matching degree and business conversion efficiency.

[0174] The application introduces the LSTM model for predicting the future optimal target delivery time, realizes the fine decision upgrade from "what to deliver" to "when to deliver". The LSTM model is good at processing time series data, can effectively model the behavior rules (such as browsing, clicking, purchasing, etc.) of users in different time periods (such as day, week, month), and identify the dynamic characteristics of individuals or groups such as active period, consumption rhythm, response preference, etc., providing strong support for time dimension prediction and improving the utilization efficiency of operation resources.

[0175] In summary, the method of the application can dynamically adapt to the changes of user interests and active time in different periods, and improve the pertinence and accuracy of marketing strategies.

[0176] The application also provides an application scenario of the marketing method based on user behavior analysis. Specifically, the marketing method based on user behavior analysis provided by the embodiment can be applied in a network shopping platform, a financial e-commerce platform, or a tourism e-commerce platform. According to the method, a first target category and a target delivery time are provided for users according to user behavior, and products of the first target category are delivered to the users at the target delivery time.

[0177] Based on the same inventive concept, the embodiment of the application also provides a marketing device based on user behavior analysis for implementing the marketing method based on user behavior analysis. The implementation scheme of the problem solving provided by the device is similar to the implementation scheme described in the above method, so the specific limitations in one or more marketing device embodiments based on user behavior analysis provided below can refer to the limitations of the marketing method based on user behavior analysis in the above text, and will not be repeated here.

[0178] In an exemplary embodiment, as shown in Figure 8 a marketing device based on user behavior analysis is provided, including:

[0179] The collection module is configured to acquire first user behavior data of a user within a first preset historical time, and preprocess the first user behavior data to acquire second user behavior data.

[0180] The analysis module is configured to:

[0181] acquire a first target category of the user by category recognition and interest analysis on the second user behavior data;

[0182] acquire a target delivery time of the user according to the second user behavior data;

[0183] The delivery module is configured to deliver a product of the first target category to the user at the target delivery time.

[0184] As an optional implementation, the analysis module is further configured to:

[0185] acquire a second target category of the user by purchase behavior recognition according to the second user behavior data, and take a preset subcategory corresponding to the second target category as a first candidate category; the second target category is one of preset supercategories; the preset supercategories include one or more preset subcategories;

[0186] acquire the first target category by user interest analysis on the first candidate category; the first target category is one of the first candidate categories.

[0187] As an optional implementation, the analysis module is further configured to:

[0188] if the number of the first candidate categories is greater than 1 and less than a first preset number: acquire a delivery content score of the first candidate categories according to the second user behavior data, and take one of the first candidate categories with the highest delivery content score as the first target category; the first preset number is greater than 1,

[0189] if the number of the first candidate categories is greater than the first preset number:

[0190] acquire an interest value of the first candidate categories according to an exponential decay model, and take a second candidate category of the first preset number with the highest interest value as the first target category;

[0191] acquire a delivery content score of the second candidate categories according to the second user behavior data, and take one of the second candidate categories with the highest delivery content score as the first target category.

[0192] As an optional implementation, the analysis module is further configured to:

[0193] acquire a first feature by feature extraction according to the second user behavior data; the first feature includes explicit behavior data and implicit behavior data corresponding to each preset supercategory;

[0194] perform pattern recognition on the first feature through the category recognition model to obtain a first target category;

[0195] As an optional implementation, the model training and updating module is configured to:

[0196] obtain the category recognition model, and specifically configured to:

[0197] obtain second features of a plurality of users in a historical time and corresponding first preset labels, the data category of the second features being the same as the type of the first features, and the first preset labels being used to indicate preset superior categories;

[0198] train the first preset classification model by taking the second features and the corresponding first preset labels as a first training set to obtain the category recognition model.

[0199] As an optional implementation, the analysis module is further configured to:

[0200] obtain first timestamp behavior features of the user in a second preset historical time according to the second user behavior data;

[0201] process the first timestamp behavior features through an active time recognition model to obtain a target delivery time;

[0202] As an optional implementation, the model training and updating module is further configured to:

[0203] obtain the active time recognition model, and specifically further configured to:

[0204] obtain second timestamp behavior features of a plurality of users corresponding to a third preset historical time and corresponding second preset labels; wherein the second preset labels are used to indicate a plurality of preset time regions in a day;

[0205] train the second preset classification model by taking the second timestamp behavior features and the second preset labels as a second training set to obtain the active time recognition model.

[0206] As an optional implementation, the analysis module is further configured to:

[0207] input the first timestamp behavior features into the active time recognition model to obtain an active probability of a preset time region, and filter a preset time region with an active probability greater than a first preset threshold as a first candidate preset time region;

[0208] obtain a preset time region corresponding to the first target category in the first candidate preset time region with an active probability highest and a competitor reach density not greater than a second preset threshold as a second candidate time region;

[0209] adjust the second candidate time region according to a preset weather rule to obtain the target delivery time.

[0210] As an optional implementation, the model training and updating module is further configured to:

[0211] monitor and update the active time identification model, in particular for:

[0212] obtaining the early warning features of the user in real time, the early warning features including the click rate, conversion rate, complaint rate and ROI index of the product corresponding to the first target category;

[0213] if the early warning features meet the preset early warning rule, repeating the obtaining method of the active time identification model to update the active time identification model.

[0214] In an exemplary embodiment, a computer device is provided, which can be a server or a terminal, and an internal structure diagram thereof can be as shown in Figure 9 The computer device includes a processor, a memory, an input / output interface (I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operating system and the computer program in the non-volatile storage medium to run. The database of the computer device is configured to store data. The input / output interface of the computer device is configured to exchange information between the processor and external devices. The communication interface of the computer device is configured to communicate with external terminals through network connection. The computer program is executed by the processor to implement a marketing method based on user behavior analysis.

[0215] Those skilled in the art can understand that Figure 9 the structure shown in the figure is only a block diagram of part of the structure related to the scheme of the present application, and does not constitute a limitation on the computer device to which the scheme of the present application is applied. The specific computer device can include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0216] In an exemplary embodiment, a computer device is provided, which includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the above method embodiments.

[0217] In an exemplary embodiment, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by the processor to implement the steps in the above method embodiments.

[0218] In an example embodiment, a computer program product is provided, including a computer program which, when executed by a processor, implements the steps of any of the above method embodiments.

[0219] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer readable storage medium, and when executed, can include the processes of the above-mentioned method embodiments. Any reference to memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical storage, high-density embedded non-volatile memory, resistive memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration but not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0220] The database involved in the embodiments provided in the present application can include at least one of a relational database and a non-relational database. The non-relational database can include a distributed database based on a block chain, etc., without being limited thereto. The processor involved in the embodiments provided in the present application can be a general processor, a central processing unit, a graphics processing unit, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., without being limited thereto.

[0221] The technical features of the above embodiments can be combined in any manner. To make the description concise, not all possible combinations of the technical features in the above embodiments are described, but as long as the combinations of the technical features do not exist, they should be considered as the scope of the present disclosure.

[0222] The principles and implementation manners of the present application are described herein by using specific examples, and the above examples are only used to help understand the method of the present application and its core idea; meanwhile, for those skilled in the art, according to the idea of the present application, the specific implementation manners and application ranges will have changes. In conclusion, the content of the specification should not be understood as a limitation of the present application.

Claims

1. A marketing method based on user behavior analysis, characterized by, The marketing method based on user behavior analysis comprises: obtaining first user behavior data of a user in a first preset historical time, and preprocessing the first user behavior data to obtain second user behavior data; identifying a purchase behavior of the user according to the second user behavior data to obtain a second target category, and taking a preset lower category corresponding to the second target category as a first candidate category; wherein the second target category is one of preset upper categories; the preset upper categories comprise one or more preset lower categories; performing user interest analysis on the first candidate category to obtain a first target category, which is one of the first candidate categories; obtaining a first timestamp behavior feature of the user in a second preset historical time according to the second user behavior data; processing the first timestamp behavior feature through an active time recognition model to obtain a target delivery time; delivering a product of the first target category to the user at the target delivery time; wherein the method for obtaining the active time recognition model comprises: obtaining second timestamp behavior features of a plurality of users in a third preset historical time and corresponding second preset labels, wherein the second preset labels are used to indicate a plurality of preset time zones in a day; training a second preset classification model by taking the second timestamp behavior features and the second preset labels as a second training set to obtain the active time recognition model.

2. The marketing method based on user behavior analysis according to claim 1, characterized in that, The user interest analysis on the first candidate category to obtain the first target category comprises: if the number of the first candidate categories is greater than 1 and less than a first preset number: obtaining a delivery content score of the first candidate categories according to the second user behavior data, and taking one with the highest delivery content score in the first candidate categories as the first target category; wherein the first preset number is greater than 1, if the number of the first candidate categories is greater than the first preset number: obtaining an interest value of the first candidate categories according to an exponential decay model, and taking the first preset number of the first candidate categories with the highest interest values as second candidate categories; obtaining a delivery content score of the second candidate categories according to the second user behavior data, and taking one with the highest delivery content score in the second candidate categories as the first target category.

3. The marketing method based on user behavior analysis according to claim 1, characterized in that, The purchase behavior identification of the user according to the second user behavior data to obtain the second target category comprises: extracting first features according to the second user behavior data; wherein the first features comprise explicit behavior data and implicit behavior data corresponding to each preset upper category; performing pattern recognition on the first features through a category recognition model to obtain the first target category; the method for obtaining the category recognition model comprises: obtaining second features of a plurality of users in a historical time and corresponding first preset labels; the data categories of the second features are the same as the types of the first features, and the first preset labels are used to indicate the preset upper categories. The second feature and the corresponding first preset label are taken as a first training set to train a first preset classification model to obtain the category recognition model.

4. The marketing method based on user behavior analysis according to claim 1, characterized in that, The processing of the first timestamp behavior feature by the active time recognition model to obtain the target delivery time comprises: The first timestamp behavior feature is input into the active time recognition model to obtain an active probability of the preset time region, and the preset time region with an active probability greater than a first preset threshold is selected as a first candidate preset time region; A second candidate time region is obtained by selecting, from the first candidate preset time region, a time region corresponding to the first target category and having a competitor reach density not greater than a second preset threshold and a highest active probability; The target delivery time is obtained by adjusting the second candidate time region according to a preset weather rule.

5. The marketing method based on user behavior analysis according to claim 1, characterized in that, The marketing method based on user behavior analysis further comprises: The active time recognition model is monitored and updated, and specifically comprises: Real-time acquisition of early warning features of the user, wherein the early warning features comprise a click rate, a conversion rate, a complaint rate and an ROI index of a product corresponding to the first target category; If the early warning features meet a preset early warning rule, the active time recognition model is repeatedly obtained to update the active time recognition model.

6. A marketing device based on user behavior analysis, characterized by, The marketing device based on user behavior analysis comprises: A collection module configured to acquire first user behavior data of a user in a first preset historical time, and to preprocess the first user behavior data to obtain second user behavior data; An analysis module configured to: identify a purchase behavior of the user based on the second user behavior data to obtain a second target category, and take a preset subcategory corresponding to the second target category as a first candidate category; wherein the second target category is one of preset superordinate categories; the preset superordinate categories comprise one or more preset subcategories; perform user interest analysis on the first candidate category to obtain a first target category, which is one of the first candidate categories; acquire first timestamp behavior features of the user in a second preset historical time based on the second user behavior data; process the first timestamp behavior features by an active time recognition model to obtain a target delivery time; The analysis module is further configured to: acquire second timestamp behavior features of a plurality of users in a third preset historical time and corresponding second preset labels, wherein the second preset labels are used to indicate a plurality of preset time regions in a day; take the second timestamp behavior features and the second preset labels as a second training set to train a second preset classification model to obtain the active time recognition model; A delivery module configured to deliver a product of the first target category to the user at the target delivery time.

7. A computer device comprising: A memory, a processor and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the marketing method based on user behavior analysis according to any one of claims 1-5.

8. A computer readable storage medium having stored thereon a computer program, characterized in that, The computer program, which is executed by a processor, implements the steps of the marketing method based on user behavior analysis according to any one of claims 1-5.

Images