Network marketing method and system based on big data

Abstract

The invention relates to the technical field of network marketing, and discloses a big data-based network marketing method and system, and the method comprises the steps: introducing the marketing acceptance of a user, carrying out the dynamic evaluation, and carrying out the marketing preheating of the user through the non-marketing content, thereby achieving the marketing preheating of the user. Under the condition that the marketing acceptability of the user is relatively high, works containing marketing contents are recommended or commodities are pushed, so that the use experience of the user of the self-media platform and the conversion efficiency of network marketing works of authors / merchants of the self-media platform are effectively improved; the technical problems that the acceptance of a self-media platform user on marketing content fluctuates dynamically, a traditional recommendation method only depends on interest portraits, the user is easy to dislike, and the conversion efficiency of network marketing works is low are solved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of network marketing, in particular to a network marketing method and system based on big data. BACKGROUND

[0002] In the current Internet environment, self-media is popular, and in the self-media platform, self-media bloggers will attract interested users through the ways of making short videos and editing scripts, etc., to realize the marketing of their own IPs, and merchants, enterprises, etc. will also realize brand building, information transmission, product promotion, and marketing strategies such as guiding consumers by operating self-media accounts.

[0003] This also generates a large amount of data, including user-side data and self-media-side data. Through data mining, user preferences or user portraits are found to push works that may be of interest to users, which has been widely applied in various Internet platforms. These works may also include works with marketing content such as advertisements and promotions.

[0004] However, for users using the self-media platform, they cannot accept the works with marketing content at any time when using the self-media platform. Meanwhile, in the prior art, when generating user portraits and pushing works to users, the user interest points are mainly used, and the user's acceptance of works with marketing content in the current state is ignored. Sometimes, multiple works with marketing content are recommended to users when they do not want to accept marketing. SUMMARY

[0005] The present application aims to provide a network marketing method and system based on big data, which introduces the marketing acceptance of users and dynamically evaluates it, and uses non-marketing content to preheat marketing for users. Only when the marketing acceptance of users is high, works containing marketing content or goods are recommended, which effectively improves the user experience of the self-media platform and the conversion efficiency of network marketing works of the self-media platform authors / merchants, and solves the technical problems of dynamic fluctuations in the acceptance of marketing content by users of the self-media platform and the low conversion efficiency of network marketing works caused by the traditional recommendation method relying only on interest portraits.

[0006] The present application is achieved by the following technical solutions: In a first aspect, a network marketing method based on big data includes: extracting a marketing acceptance feature of the user according to historical browsing data of the user in the database; calculating a current marketing acceptance of the user according to the marketing acceptance feature and recent browsing data in a target time period; pushing works to the user according to a push scheme matched with the current marketing acceptance of the user, when the current marketing acceptance of the user is less than a first threshold value, only pushing a first target work to the user, when the current marketing acceptance of the user reaches the first threshold value, pushing the first target work or a second target work to the user, and when the current marketing acceptance of the user reaches a second threshold value, matching a target commodity for the user; The first target work is configured as a work for improving the current marketing acceptance of the user and does not contain marketing content, and the second target work is configured as a work for improving the current marketing acceptance of the user and contains marketing content.

[0007] In order to better realize the present application, further, before pushing works to the user according to a push scheme matched with the current marketing acceptance of the user, the method further comprises: establishing a coarse ranking model based on interest recall and social heat recall using works in a first candidate content pool; obtaining a fine ranking model according to the coarse ranking model and performing diversity control on the fine ranking model to generate a first target work list; marketing intensity quantization is performed on works in a second candidate content pool to generate a second target work list.

[0008] In order to better realize the present application, further, the method for extracting a marketing acceptance feature of the user according to historical browsing data of the user in the database comprises: extracting an explicit marketing acceptance feature of the user according to marketing content interaction data of the user in the historical browsing data of the user; extracting an implicit marketing acceptance feature of the user by data mining on the historical browsing data of the user; extracting a deep marketing acceptance feature of the user by using a double-tower DNN model to process the historical browsing data of the user.

[0009] In order to better realize the present application, further, the explicit marketing acceptance feature comprises an advertisement closing rate, a marketing content skipping rate and a negative feedback rate; The implicit marketing acceptance feature comprises a marketing sensitive time period, a marketing content tolerance and a decay weight of a historical marketing interaction rate; The deep marketing acceptance feature is a probability of interaction of the user with marketing content.

[0010] In order to better realize the present application, further, the method for calculating a current marketing acceptance of the user according to the marketing acceptance feature and recent browsing data in a target time period comprises: Input the marketing acceptance characteristics and recent browsing data in the target time period into the GBDT model to obtain the user's current marketing acceptance; When the number of interactions generated by a user exceeds the preset value, the user's current marketing acceptance is recalculated.

[0011] In order to better implement the present invention, further, it also includes: The first threshold and the second threshold are dynamically adjusted using a Bandit algorithm.

[0012] In order to better implement the present invention, further, it also includes: The target products are matched to users through the GraphSAGE matching algorithm and a graph neural network whose node types include user nodes, product nodes, content nodes, and author nodes.

[0013] In order to better implement the present invention, further, after pushing the work to the user according to the push plan that matches the user's current marketing acceptance, the method further includes: Establish a user group portrait set and a work performance data set. After performing cluster analysis on the user group portrait set and attribution analysis on the work performance data set, generate a user in-depth analysis report, a work diagnosis and improvement suggestion report, and a work optimization template and push them to the target author.

[0014] Second, a big data-based online marketing system includes: A marketing acceptance feature extraction module, which is used to extract the user's marketing acceptance features based on the user's historical browsing data in the database; A marketing acceptance dynamic evaluation module, which is used to calculate the user's current marketing acceptance based on marketing acceptance characteristics and recent browsing data within a target time period; An intelligent push module, configured to push works to a user according to a push plan that matches the user's current marketing acceptance. When the user's current marketing acceptance is less than a first threshold, only the first target work is pushed to the user. When the user's current marketing acceptance reaches the first threshold, either the first target work or the second target work is pushed to the user. When the user's current marketing acceptance reaches the second threshold, a target product is matched with the user. The first target work is configured as a work used to improve the user's current marketing acceptance and does not contain marketing content, and the second target work is configured as a work used to improve the user's current marketing acceptance and contains marketing content.

[0015] In order to better implement the present invention, further, it also includes: The creation enabling module is used to establish a user group portrait set and a work performance data set. After performing group cluster analysis on the user group portrait set and attribution analysis on the work performance data set, it generates a user in-depth analysis report, a work diagnosis and improvement suggestion report, and a work optimization template and pushes them to the target author.

[0016] Compared with the prior art, the present invention has the following advantages and beneficial effects: By dynamically calculating the user's current marketing acceptance as the basis for online marketing to users of self-media platforms, and replacing explicit feedback with implicit acceptance detection, the user experience can be improved and silent users who do not actively provide feedback can be covered; When users' marketing acceptance is low, rebuild user trust by using works that can improve users' current marketing acceptance and do not contain marketing content; By gradually increasing user acceptance of marketing, we can capture conversion time and conduct precise online marketing without causing user disgust as much as possible. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] The present invention is further described in conjunction with the following drawings and embodiments, and all concepts and innovations of the present invention should be regarded as disclosed contents and the protection scope of the present invention.

[0018] Figure 1 This is a flow chart of Example 1 of an online marketing method based on big data in this application.

[0019] Figure 2 This is a flow chart of Example 2 of an Internet marketing method based on big data in this application.

[0020] Figure 3 This is a flow chart of Example 4 of a big data-based online marketing method in this application.

[0021] Figure 4 This is a functional module diagram of an embodiment of a big data-based online marketing system in this application.

[0022] Figure 5 This is a functional module diagram of a preferred implementation of an embodiment of a big data-based online marketing method in this application. DETAILED DESCRIPTION

[0023] Example 1: like Figure 1 As shown, an embodiment of an Internet marketing method based on big data includes: Extract users’ marketing acceptance characteristics based on their historical browsing data in the database; Calculate the user's current marketing acceptance based on marketing acceptance characteristics and recent browsing data within the target time period; Push works to users according to the push plan that matches the user's current marketing acceptance. When the user's current marketing acceptance is less than a first threshold, only the first target work is pushed to the user. When the user's current marketing acceptance reaches the first threshold, the first target work or the second target work is pushed to the user. When the user's current marketing acceptance reaches the second threshold, the user is matched with the target product. The first target work is configured as a work used to improve the user's current marketing acceptance and does not contain marketing content, and the second target work is configured as a work used to improve the user's current marketing acceptance and contains marketing content.

[0024] Specifically, the user's historical browsing data includes all browsing-related data of all users of the platform obtained by the self-media platform through legal channels and with full respect and protection of user privacy, including but not limited to the user's viewing time of the work, sliding playback data, number of skips of marketing content, number of active feedback, search history, number of clicks on marketing links, comments and likes and other interactive data, data related to browsing works, etc., as well as other user-related data obtained through legal channels and with full respect and protection of user privacy, including but not limited to various local data obtained by the user after granting access rights.

[0025] A user's marketing acceptance is a quantitative value. The higher the user's marketing acceptance, the higher their tolerance for marketing content. For example, when marketing acceptance is high, users accept marketing content and even actively search for it. For example, even if advertising content appears in a work, users will not skip the ad and will let the work play to completion. Or, after seeing an ad, users will actively search for products related to the ad on e-commerce platforms. When marketing acceptance is low, once users determine that a work contains or may contain marketing content, they will immediately abandon the current content. For example, if they see an ad, they will immediately swipe away to move on to the next work, or even actively report that it is an advertisement and wish to be pushed less. When marketing acceptance is at a medium value, if advertising content appears in a work, users may skip the ad and continue browsing subsequent content. The first and second thresholds are used to divide user status. When the user's marketing acceptance is below the first threshold, the user is judged to be completely unacceptable to marketing content. When the user's marketing acceptance reaches the first threshold but is lower than the second threshold, the user may skip the marketing content or accept the marketing content. When the user's marketing acceptance reaches the second threshold, the user is very happy to be pushed marketing content and may even actively search for related products. Marketing acceptance characteristics are all characteristics that affect the change of marketing acceptance values, including user-side characteristics, work-side characteristics and other characteristics (such as time, season, weather, etc.).

[0026] The target time period is calculated from the time the user starts browsing. It can be generated based on the user's browsing habits, such as 5 minutes. For the same user, the target time period will be dynamically adjusted at different times. For example, when predicting the user's commute time, the target time period during working hours will be shorter than the target time period after working hours. Currently, the prerequisite is to be able to accumulate enough recent browsing data to calculate the user's current marketing acceptance.

[0027] The push plan includes the proportion of push work types and the push order of different work types. In this embodiment, the design of the push plan should follow two principles: minimize the continuous push of works with marketing content in a short period of time, and try to improve the current marketing acceptance without the user's active feedback, so as to ensure the user's usage experience and usage stickiness, and reduce user churn. The target products are products that users are likely to click on or even purchase.

[0028] In this embodiment, calculation and push are performed separately for each user. Of course, for new users, there may not be enough historical data recorded. At this time, users with high similarity to the current new user can be retrieved from the database and the corresponding historical browsing data can be used for analysis first.

[0029] Optionally, to better handle large amounts of data, a streaming computing framework such as Flink can be introduced when calculating users' current marketing acceptance to ensure real-time results.

[0030] Optionally, when extracting the user's marketing acceptance features, a time series clustering algorithm is used to extract features related to the time period to ensure the accuracy of feature extraction.

[0031] Example 2 like Figure 2 As shown, this embodiment is further optimized based on the above embodiment 1. In this embodiment, before pushing the work to the user according to the push plan matching the user's current marketing acceptance, it also includes: Using the works in the first candidate content pool, a coarse ranking model is established based on interest recall and social popularity recall; Obtaining a refined ranking model based on the rough ranking model and performing diversity control on the refined ranking model to generate a first target works list; The marketing intensity of the works in the second candidate content pool is quantified to generate a second target work list.

[0032] Optionally, based on the social heat signal, the content knowledge graph and the user behavior log in the recent browsing data, a first candidate content pool and a second candidate content pool are initially constructed. Specifically, by processing the Kafka log stream through Spark Streaming, the browsing, liking, collecting and the like in the recent browsing data are collected in real time, a similar content candidate set is generated based on collaborative filtering (Item-CF), a content knowledge graph is constructed by extracting structured work features and unstructured features such as work tags, author influence, content quality score, a social heat signal is obtained by performing trend mining based on time decay TF-IDF and by constructing a propagation force model through computing sharing rate / comment density for propagation analysis, and the first candidate content pool and the second candidate content pool are initially constructed by comprehensively considering the three.

[0033] Optionally, a double-tower DNN model is used for interest-based recall to realize personalized deep matching. Specifically, a user tower is constructed according to the user ID and the historical behavior sequence, a content tower is constructed according to the content features, positive samples are constructed by using user clicks / completed content, and negative samples are constructed by using exposed non-clicked content and global random sampling.

[0034] Optionally, a heat formula is constructed based on the number of shares, the number of comments and the author. Specifically, the expression of the heat formula is wherein n is the number of shares, m is the number of comments, M is an author fan quantity coefficient, which is the square root of the author fan quantity in the embodiment, is the number of hours from the content publishing time to the current time, and λ is an adjustable decay intensity coefficient, which constitutes an adjustable time decay factor; in order to suppress the monopoly of head authors, an upper limit of single-author content proportion can also be set, and in order to protect the freshness, new content within 24 hours can also be weighted.

[0035] Optionally, the top-ranked features in the coarse ranking model are screened through SHAP values, feature distillation is performed, and a fine ranking model is constructed; in the embodiment, the top 50 features in the coarse ranking model are screened.

[0036] Optionally, the fine ranking model is controlled in diversity, including rearranging the output list of the fine ranking model. Specifically, the category scattering, similarity penalty and potential content protection can be performed in the rearrangement stage, for example, the category scattering is performed on the interval of the author / same theme content in the front row ≥ 3, the potential content protection is performed on the low exposure high interaction content by setting a scoring item; the diversity detection based on the ecological balance algorithm can be performed on the output list of the fine arrangement model, the original order is kept for the output list with reasonable distribution, the exploration strategy is started for the output list with high homogeneity, the high potential new content is inserted, of course, the exploration strategy is prohibited in sensitive fields such as medical / financial, and the exploration flow is not more than 5% of the total exposure; at the same time, business rules are formulated, the content is automatically promoted in the first week to start the new product, and the same category content is weighted in 24 hours to control fatigue.

[0037] By using the embodiment, the content breadth is ensured by the interest / hotness double recall, the priority of the high acceptance content is ensured by the cooperative optimization of coarse arrangement and fine arrangement, the information cocoon is broken by the diversity rearrangement, and the balance between the marketing acceptance and the marketing target is realized.

[0038] Embodiment 3 The embodiment is further optimized on the basis of the above-mentioned embodiments 1 or 2, in the embodiment, the method for extracting the marketing acceptance feature of the user according to the historical browsing data of the user in the database, comprising: extracting the explicit marketing acceptance feature of the user according to the marketing content interaction data of the user in the historical browsing data of the user; extracting the implicit marketing acceptance feature of the user by data mining on the historical browsing data of the user; extracting the deep marketing acceptance feature of the user by using the double-tower DNN model to process the historical browsing data of the user.

[0039] Further optionally, the explicit marketing acceptance feature comprises the advertisement closing rate, the marketing content skipping rate and the negative feedback rate; The implicit marketing acceptance feature comprises the marketing sensitive time period, the marketing content tolerance and the decay weight of the historical marketing interaction rate; The deep marketing acceptance feature is the probability of the interaction of the user with the marketing content.

[0040] Specifically, the advertisement closing rate is the ratio of the number of closed advertisements and the number of exposed advertisements, the marketing content skipping rate is the ratio of the number of fast-forward or swiping marketing videos and the total number of exposures, and the negative feedback rate is the ratio of the number of clicks on “not interested” and the total number of exposures; The marketing sensitive time period is obtained by using K-Shape clustering. First, a user's 24-hour segmented marketing avoidance rate matrix is established according to historical browsing data, wherein the marketing avoidance rate refers to the ratio of the number of times that the user skips (or closes) the marketing content in a specific hour to the total number of times that the user is exposed to the marketing content in the specific hour. For example, user A has 10 exposures from 0 to 1 o'clock, of which 2 are skipped, so the avoidance rate of 0-1 o'clock is 0.2. From 1 to 2 o'clock, there are 5 exposures, of which 1 is skipped, so the avoidance rate is 0.2. The clustering output is a classification cluster number, 0 to n, indicating which cluster the user is divided into, and n+1 is the number of preset clusters. Each cluster represents a typical avoidance mode, and the avoidance peak period is the hour with the highest avoidance rate in 24 hours. The number of clusters n+1 is a preset hyperparameter that can be adjusted by methods such as silhouette coefficients. In this embodiment, n is 5, and there are 6 clusters. Cluster 0 represents that the avoidance rate is high all day, but especially peaks from 14 to 17 o'clock in the afternoon, which is called the working time cluster of office workers. Cluster 1 represents that the avoidance rate is high from 21 o'clock at night to 2 o'clock in the morning, which is called the cluster of users who stay up late. Cluster 2 represents that the avoidance rate is high during the morning and evening peaks from 8 to 10 o'clock and from 18 to 20 o'clock, which is called the commuting time cluster. Cluster 3 represents that the avoidance rate is low all day, and there is no obvious peak, which is called the cluster of users who are not sensitive to marketing. The avoidance rate is high from 12 to 13 o'clock in the afternoon and from 19 to 21 o'clock in the evening, which is called the student cluster. Cluster 5 represents random avoidance, which is called the cluster with no obvious pattern. In this way, not only can we know which cluster the user is in, but also we can know when the user is most resistant to marketing.

[0041] Optionally, the marketing content tolerance is obtained by establishing an LDA topic model. Specifically, first, a set of non-marketing content text browsed by the user is obtained, which can be the content explicitly marked as "non-marketing" in the user's historical browsing records. The text information contained in each content item is processed in the format of "title + body + comments". Then, all non-marketing content of the user is concatenated into a single document to complete document construction and form a text corpus representing the user's safe interest area. A dictionary is created through word segmentation processing, filtering of stop words, establishment of a word library, and the like to obtain a word-ID mapping table. Next, the document vectors of all users are input, and the LDA model is trained according to the preset number of topics and the number of iterations. Finally, the target user's document is input into the LDA model to obtain the target user's topic distribution vector and safe area radius. The target user's topic distribution vector represents the user's interest intensity in each preset topic. When the topic distribution of new content and the user's distribution have a JS divergence of >(1-R), it is determined that the safe area is exceeded.

[0042] Optionally, the Holt-Winters model is used to calculate the decay weight of the historical marketing interaction rate. Specifically, first, a sliding window is used in weeks, and the single-week interaction rate is the ratio of the number of interactions with the marketing content in that week to the number of exposures. A list of interaction rates sorted by time is established. If the historical data of the target user does not meet the requirements in terms of time length, data with high similarity is retrieved from the database for processing; then, the three elements of the horizontal component, trend component and seasonal component are initialized, among which the initial value of the horizontal component is the first week interaction rate, the initial value of the trend component is the difference between the second week interaction rate and the first week interaction rate, and the initial value of the seasonal component is the difference between the average interaction rate of the i-th week and the global interaction rate average; then, the three elements are iteratively updated, and finally, the next period is predicted, and the attenuation weight sequence of the historical marketing interaction rate is output.

[0043] Optionally, deep marketing acceptance features are also obtained using the dual-tower DNN model; Specifically, the user tower here uses the historical behavior sequence encoded by Transformer as input, and the content tower uses work labels and marketing intensity features as input, and outputs the probability of users interacting with marketing content as the deep marketing acceptance feature.

[0044] Finally, in this embodiment, the user marketing acceptance feature vector is expressed as [explicit feature, implicit feature, DNN output value].

[0045] Optionally, a method for calculating a user's current marketing acceptance based on marketing acceptance characteristics and recent browsing data within a target time period includes: Input the marketing acceptance characteristics and recent browsing data in the target time period into the GBDT model to obtain the user's current marketing acceptance; When the number of interactions generated by a user exceeds the preset value, the user's current marketing acceptance is recalculated.

[0046] Specifically, the GBDT model can be an XGBoost model or a LightGBM model. The input of the model is the user's marketing acceptance feature vector and user behavior statistics in recent browsing data, and the output is the current marketing acceptance probability value; when the number of interactions generated by the user is greater than the preset value, such as when the user generates more than 5 interactions, the user's current marketing acceptance is recalculated through Flink stream processing.

[0047] Optionally, the method further includes: dynamically adjusting the first threshold and the second threshold using a Bandit algorithm.

[0048] Specifically, first, in the exploration phase, a preliminary understanding of the threshold effect is established, which lasts for the first 24 hours, and random allocation and uniform exploration are carried out. During the uniform exploration, it is ensured that each threshold group obtains sufficient exposure opportunities, such as 200 times, and user response data is recorded. For example, users accepting marketing and converting are recorded as success, and users skipping / closing / marking dislike are recorded as failure. At the same time, each threshold group is initialized with a uniform distribution; then, in the utilization phase, the long-term conversion rate is maximized, probability distribution sampling is performed, and the threshold group with the largest sampling value is selected for feedback update.

[0049] Further optionally, when the first threshold and the second threshold are dynamically adjusted using the Bandit algorithm, user stratification can also be performed to classify users into new users, high-value users, and ordinary users, and Bandit can be run independently for each type of user.

[0050] Optionally, it also includes: matching target products for users through the GraphSAGE matching algorithm and a graph neural network whose node types include user nodes, product nodes, content nodes, and author nodes.

[0051] Specifically, the edge relationships formed by user nodes, product nodes, content nodes, and author nodes include user-content, user-product, content-product, and author-product. The user-content edge relationship represents browsing / liking / sharing, with the weight being the intensity of the behavior. The user-product edge relationship represents adding to cart / purchasing, with the weight being the transaction amount. The content-product and author-product edge relationships represent content mentioning products, with the weight being the association strength. The author-product edge relationship represents the relationship of bringing products to consumers, with the weight being the depth of cooperation. In this way, a heterogeneous information network can be obtained. When using the GraphSAGE matching algorithm, first, the node corresponding to the target user is found in the heterogeneous graph. Then, GraphSAGE is used to perform two to three layers of multi-hop neighbor aggregation to obtain the matching scores of each candidate product, including cosine similarity and meta-path similarity. Finally, using cosine similarity, meta-path similarity and adjustable weight parameters, the total matching score of each candidate product is obtained and ranked, with the one or more products ranked highest as the target product.

[0052] Example 4 like Figure 3 As shown, this embodiment is further optimized based on the above embodiment 2 or 3. In this embodiment, after pushing the work to the user according to the push plan matching the user's current marketing acceptance, it also includes: Establish a user group portrait set and a work performance data set. After performing cluster analysis on the user group portrait set and attribution analysis on the work performance data set, generate a user in-depth analysis report, a work diagnosis and improvement suggestion report, and a work optimization template and push them to the target author.

[0053] Specifically, we established user group portraits that included user marketing acceptance distribution, content preference heat maps, behavioral pattern clustering labels, and conversion path analysis. We also established a content performance database that included content metadata, interaction metrics, conversion funnel data, and user feedback. We used an improved K-Means++ algorithm to process high-dimensional mixed data, using marketing acceptance level, content consumption depth, and conversion sensitivity points as clustering dimensions to output 5 to 8 core user groups and their characteristic descriptions. We then used dual machine learning to quantify the contribution of each content element to conversion and conduct causal inference. The output includes a target user analysis report on group profiles, content consumption behaviors, and conversion drivers. For example, the group profile is "25-35-year-old working mothers: prefer practical tips content," the content consumption behavior is "golden viewing hours: 20:00-22:00," and the conversion driver is "price sensitivity: medium, but high sensitivity to gifts." The output diagnostic dimensions include structural analysis, content element evaluation, and emotional matching work diagnostic report and improvement suggestions. For example, the work diagnostic report is: Structural analysis: "The dropout rate in the first 5 seconds is 35% higher than that of similar works"; Content element evaluation: "The product display time is insufficient (currently 12%, recommended 20%-25%)" Emotional matching: "The target group prefers rational explanations, and the current level of entertainment is too high."

[0054] For example, suggestions for improving the work are: Video ID: V20230815003 Problem: 00:45-00:52 The information density of the oral broadcast is insufficient Recommended solution: 1. Compress the empty shots by 2 seconds 2. Add data annotation barrage 3. Insert user testimonials Expected improvement: completion rate +15%, conversion rate +8%.

[0055] Specifically, the construction process of the work optimization template includes: building a template selector based on the target group characteristics and popular content patterns, instantiating and dynamically filling the template according to the template selector, and realizing multi-version output.

[0056] Specifically, we will build a creator dashboard that includes content health scores, competitor benchmarking analysis, real-time optimization suggestion push, and user group migration warnings, as well as an automated template factory that outputs editing project files, structured scripts, and cross-platform adaptation packages.

[0057] Specifically, regularly obtain the target author's suggestion adoption rate, effect improvement, template usage frequency and creator growth curve to establish a personal ability profile of the target author.

[0058] By adopting this implementation method, user behavior data is converted into executable creation guidelines, breaking through the limitations of traditional "guessing what you like" and realizing data-driven creation; by establishing personal ability portraits of creators and identifying the characteristics of their works, dynamic capacity building of creators is achieved; brands can accurately guide co-creators and realize platform ecological empowerment.

[0059] Example 5 like Figure 4 As shown, an embodiment of an online marketing system based on big data includes: A marketing acceptance feature extraction module, which is used to extract the user's marketing acceptance features based on the user's historical browsing data in the database; A marketing acceptance dynamic evaluation module, which is used to calculate the user's current marketing acceptance based on marketing acceptance characteristics and recent browsing data within a target time period; An intelligent push module, configured to push works to a user according to a push plan that matches the user's current marketing acceptance. When the user's current marketing acceptance is less than a first threshold, only the first target work is pushed to the user. When the user's current marketing acceptance reaches the first threshold, either the first target work or the second target work is pushed to the user. When the user's current marketing acceptance reaches the second threshold, a target product is matched with the user. The first target work is configured as a work used to improve the user's current marketing acceptance and does not contain marketing content, and the second target work is configured as a work used to improve the user's current marketing acceptance and contains marketing content.

[0060] like Figure 5 As shown, a preferred embodiment of a network marketing system based on big data also includes: The creation enabling module is used to establish a user group portrait set and a work performance data set. After performing group cluster analysis on the user group portrait set and attribution analysis on the work performance data set, it generates a user in-depth analysis report, a work diagnosis and improvement suggestion report, and a work optimization template and pushes them to the target author.

[0061] The above description is only a preferred embodiment of the present invention and does not limit the present invention in any form. Any simple modification or equivalent change made to the above embodiment based on the technical essence of the present invention falls within the scope of protection of the present invention.

Claims

1. A network marketing method based on big data, characterized in that: include: Extract users’ marketing acceptance characteristics based on their historical browsing data in the database; Calculate the user's current marketing acceptance based on marketing acceptance characteristics and recent browsing data within the target time period; Push works to users according to the push plan that matches the user's current marketing acceptance. When the user's current marketing acceptance is less than a first threshold, only the first target work is pushed to the user. When the user's current marketing acceptance reaches the first threshold, the first target work or the second target work is pushed to the user. When the user's current marketing acceptance reaches the second threshold, the user is matched with the target product. The first target work is configured as a work used to improve the user's current marketing acceptance and does not contain marketing content, and the second target work is configured as a work used to improve the user's current marketing acceptance and contains marketing content.

2. The big data-based online marketing method according to claim 1, characterized in that: Before pushing works to users based on a push plan that matches their current marketing acceptance, it also includes: Using the works in the first candidate content pool, a coarse ranking model is established based on interest recall and social popularity recall; Obtaining a refined ranking model based on the rough ranking model and performing diversity control on the refined ranking model to generate a first target works list; The marketing intensity of the works in the second candidate content pool is quantified to generate a second target work list.

3. The big data-based online marketing method according to claim 1, characterized in that: The method for extracting the user's marketing acceptance characteristics based on the user's historical browsing data in the database includes: Extract the user's explicit marketing acceptance characteristics based on the user's marketing content interaction data in the user's historical browsing data; Conduct data mining on users’ historical browsing data to extract users’ implicit marketing acceptance characteristics; Use the dual-tower DNN model to process users' historical browsing data and extract users' deep marketing acceptance features.

4. The big data-based online marketing method according to claim 3, characterized in that: Explicit marketing acceptance characteristics include ad closing rate, marketing content skipping rate, and negative feedback rate; Implicit marketing acceptance characteristics include marketing sensitive time periods, marketing content tolerance, and the attenuation weight of historical marketing interaction rates; The deep marketing acceptance characteristic is the probability of users interacting with marketing content.

5. The big data-based online marketing method according to claim 1, characterized in that: Methods for calculating a user's current marketing acceptance based on marketing acceptance characteristics and recent browsing data within a target time period include: Input the marketing acceptance characteristics and recent browsing data in the target time period into the GBDT model to obtain the user's current marketing acceptance; When the number of interactions generated by a user exceeds the preset value, the user's current marketing acceptance is recalculated.

6. The big data-based online marketing method according to claim 1, characterized in that: Also includes: The first threshold and the second threshold are dynamically adjusted using a Bandit algorithm.

7. The big data-based online marketing method according to claim 1, characterized in that: Also includes: The target products are matched to users through the GraphSAGE matching algorithm and a graph neural network whose node types include user nodes, product nodes, content nodes, and author nodes.

8. The big data-based online marketing method according to claim 1, characterized in that: After pushing works to users according to the push plan that matches the user's current marketing acceptance, it also includes: Establish a user group portrait set and a work performance data set. After performing cluster analysis on the user group portrait set and attribution analysis on the work performance data set, generate a user in-depth analysis report, a work diagnosis and improvement suggestion report, and a work optimization template and push them to the target author.

9. A big data-based online marketing system, characterized in that: include: A marketing acceptance feature extraction module, which is used to extract the user's marketing acceptance features based on the user's historical browsing data in the database; A marketing acceptance dynamic evaluation module, which is used to calculate the user's current marketing acceptance based on marketing acceptance characteristics and recent browsing data within a target time period; An intelligent push module, configured to push works to a user according to a push plan that matches the user's current marketing acceptance. When the user's current marketing acceptance is less than a first threshold, only the first target work is pushed to the user. When the user's current marketing acceptance reaches the first threshold, either the first target work or the second target work is pushed to the user. When the user's current marketing acceptance reaches the second threshold, a target product is matched with the user. The first target work is configured as a work used to improve the user's current marketing acceptance and does not contain marketing content, and the second target work is configured as a work used to improve the user's current marketing acceptance and contains marketing content.

10. The big data-based online marketing system according to claim 9, characterized in that: Also includes: The creation enabling module is used to establish a user group portrait set and a work performance data set. After performing group cluster analysis on the user group portrait set and attribution analysis on the work performance data set, it generates a user in-depth analysis report, a work diagnosis and improvement suggestion report, and a work optimization template and pushes them to the target author.

Images