Machine learning-based accurate matching method and system for arriving person and brand

Abstract

The invention relates to the technical field of brand matching, in particular to a machine learning-based accurate matching method and system for a person and a brand. Comprising the following steps: S1, collecting audience data reaching a live broadcasting room, and meanwhile, performing audience list effective screening on the audience data according to watching duration; s2, collecting fan data of the arrival person, collecting a fan list according to the fan data, obtaining like data and comment data of the arrival person works at the same time, and collecting a potential user list according to the like data and the comment data; data are divided into continuous time periods according to collection time to generate user sub-portraits in all the time periods, evolution of user behaviors along with time is captured, portraits in all the time periods are gathered to form a user total portrait, meanwhile, the dynamic trend of a user group is quantified through difference analysis, dynamic map observation is provided for brand matching, and the user experience is improved. And the influence factor of the hot spot on the user portrait change is quantified by introducing network hot spot data, and the burst trend is identified.

Description

Technical Field

[0001] The present invention relates to the field of brand matching technology, and more specifically, to a method and system for accurately matching influencers and brands based on machine learning. Background Art

[0002] In the convergence of social media and e-commerce, accurately matching influencers with brands is a key component in improving marketing efficiency. This approach primarily relies on identifying potential partners by comparing user profiles with the underlying characteristics of a brand's target audience, thereby reducing marketing costs and improving conversion efficiency. However, most existing technologies are only applicable to scenarios with stable user groups and slowly changing market trends, such as static user portrait analysis on traditional e-commerce platforms. In the current short video live broadcast scenario with strong real-time performance and fragmented user behavior, relying solely on static fan portraits or single interaction data is unable to capture the dynamic preference changes of users caused by network hotspots and time periods, resulting in matching results lagging behind market trends and insufficient timeliness in influencer and brand matching marketing. Therefore, a method and system for precise influencer and brand matching based on machine learning is proposed. Summary of the Invention

[0003] The purpose of the present invention is to provide a method and system for accurately matching influencers and brands based on machine learning to solve the problems raised in the above background technology.

[0004] To solve the above technical problems, one of the objectives of the present invention is to provide a method for accurately matching influencers and brands based on machine learning, comprising the following steps: S1. Collect audience data from live broadcasts of influencers and effectively filter audience lists based on viewing time. S2. Collect the influencer's fan data and compile a fan list based on the fan data. At the same time, obtain the likes and comments data of the influencer's works and compile a potential user list based on the likes and comments data. S3. Generate a network profile for each user based on the audience list, fan list, and potential user list at different time periods. At the same time, summarize the network profiles of the corresponding users of the expert at different time periods and perform data analysis on user differences at different time periods. S4. Obtain network hotspot data and record changes in each user's network profile. Then, analyze the hotspot impact data for each user based on the network hotspot data. Build a machine learning model based on the network hotspot data, user difference data, and hotspot impact data. The machine learning model outputs the expert's real-time user profile and predicted user profile. S5. Obtain historical order data and match historical purchase records for the corresponding user based on the historical order data. Then, use the machine learning model to analyze the purchase preferences of each user based on their historical purchase records. S6. Perform correlation analysis based on the influencer's real-time user profile, predicted user profile, and predicted purchase preferences with the brands to be matched, and make brand matching recommendations for the influencer based on the correlation.

[0005] As a further improvement of this technical solution, the steps of S1 are as follows: S1.1. When the influencer enters the live broadcast mode, the audience data of the live broadcast room is collected in real time, and the viewing time of each viewer is calculated in real time based on the audience data; S1.2. Based on the set effective viewing time, the real-time viewing time of each viewer is compared with the effective viewing time. If the real-time viewing time exceeds the effective viewing time, the viewer is considered effective. Otherwise, if the real-time viewing time does not exceed the effective viewing time, the monitoring is continued. S1.3. Collect the audience members determined as valid in S1.2 and obtain the audience list based on the collection results.

[0006] As a further improvement of this technical solution, the steps of S2 are as follows: S2.1. Log in to the influencer's online account and collect fan data in the online account to obtain the influencer's fan data. Collect a fan list based on the fan data to obtain the influencer's corresponding fan list. S2.2. Using the influencer's online account to collect likes and comments on the influencer's online works, and then setting likes and comments thresholds based on the number of influencer's online works; S2.3. Perform user attribution analysis based on the likes and comments data to obtain the number of likes and comments for each user, and then compare the number of likes and comments for each user with the likes and comments thresholds and comment thresholds; When one of the number of likes and comments of a user exceeds the threshold of the number of likes and the threshold of the number of comments, the user is determined to be a potential user and entered into the potential user list; When the number of likes and comments of a user does not exceed the like threshold and comment threshold, the user will continue to be monitored.

[0007] As a further improvement to this technical solution, when S3 generates a network profile, if the same user appears in the audience list, fan list, and potential user list, only one list is assigned to the user, and the priority of the assignment is as follows: Fans list > audience list; Audience list > potential users.

[0008] As a further improvement of this technical solution, the steps of S3 are as follows: S3.1. Divide the audience list, fan list, and potential user list into time periods based on the collection time, generate network profiles for each user in the audience list, fan list, and potential user list for different time periods, and obtain the corresponding network profile for each user based on the generated results; S3.2. Summarize the user's online profile and divide it into time periods to obtain the corresponding profile of each user in different time periods and the total user profile of the influencer in different time periods; S3.3. Perform user difference analysis on the total network portraits of all users corresponding to the expert at different time periods to obtain user difference data corresponding to the expert.

[0009] As a further improvement of this technical solution, the steps of S4 are as follows: S4.1. Obtain network hotspot data for different time periods, and simultaneously analyze the changes in each user's network profile over different time periods to obtain the corresponding change data for the user's network profile over different time periods. Combine this change data with the network hotspot data to perform hotspot impact data analysis to obtain the hotspot impact data corresponding to the user; S4.2. Establish a machine learning model based on real-time network hotspot data, change data, user difference data, and hotspot impact data. Then, after learning the data, the machine learning model outputs the real-time user total portrait and predicted user total portrait corresponding to the expert.

[0010] As a further improvement of this technical solution, the user total portrait of S3.2 and the real-time user total portrait of S4.2 are different. The user total portrait of S3.2 is defined as the basic user total portrait, and the real-time user total portrait of S4.2 is defined as the basic user total portrait after correction.

[0011] As a further improvement of this technical solution, the steps of S5 are as follows: S5.1. Obtain order history data from the influencer's online account and the brand's sales portal. Match the user's purchase history with the order history data to obtain each user's corresponding purchase history. S5.2. Calculate the linkage value between each product purchase, then input each user's historical purchase record and linkage data into the machine learning model, and the machine learning model outputs each user's predicted purchase preference.

[0012] A second object of the present invention is to provide a precise matching system for influencers and brands based on machine learning, including any of the aforementioned precise matching methods for influencers and brands based on machine learning, including a list collection module, a list analysis module, and a brand matching module; The list collection module is used to collect audience data for effective screening of the audience list, collect fan data for fan list collection, obtain likes data and comment data for potential user list collection, generate network portraits for each user based on the audience list, fan list and potential user list in different time periods, and perform user difference data analysis in different time periods; The list analysis module is used to obtain network hotspot data and record the change data of each user's network profile. Then, the hotspot impact data analysis is performed on each user in combination with the network hotspot data. Then, a machine learning model is established based on the network hotspot data, user difference data, and hotspot impact data. The machine learning model outputs the real-time user profile and predicted user profile of the expert. The brand matching module is used to obtain historical order data, and match historical purchase records for corresponding users based on the historical order data. Then, the machine learning model is combined with each user's historical purchase records to perform a predicted purchase preference analysis. The correlation analysis is performed based on the influencer's real-time user profile, predicted user profile, and predicted purchase preferences with the brand to be matched, and brand matching recommendations are made for the influencer based on the correlation.

[0013] Compared with the prior art, the present invention has the following beneficial effects: 1. A machine learning-based precise matching method and system for influencers and brands divides data into continuous time periods according to the collection time to generate user sub-portraits for each time period, captures the evolution of user behavior over time, and aggregates the portraits of each time period to form a user overall portrait. At the same time, dynamic trends of user groups are quantified through difference analysis, providing dynamic map observation for brand matching. Then, by introducing network hot spot data, the influencing factors of hot spots on changes in user portraits are quantified, sudden trends are identified, and a dynamic model is constructed using an LSTM network. Real-time features are input to generate a real-time overall portrait, and future trends are predicted based on historical sequences, so that brand matching can respond to current needs and predict long-term trends.

[0014] 2. A precise matching method and system for influencers and brands based on machine learning. By collecting audience data, fan data, and work interaction data from influencers' live broadcast rooms, a three-dimensional data set covering user travel, social attributes, and content preferences is constructed to avoid the one-sidedness of single-dimensional evaluation. At the same time, core audiences are screened based on effective viewing time to prevent transient audiences from interfering with analysis. High-engagement potential users are identified through thresholds for the number of likes and comments. Secondly, priority rules are used to ensure that each user is assigned only a unique high-value label, providing a pure data foundation for subsequent portrait generation. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 It is the overall flow chart of the present invention; Figure 2 This is a flowchart of the present invention for obtaining an audience list based on the collection results; Figure 3 A flowchart of the present invention for obtaining a fan list corresponding to the expert; Figure 4 This is a flowchart of the present invention for obtaining user difference data corresponding to experts; Figure 5 This is a flowchart of the present invention outputting a real-time user profile and a predicted user profile corresponding to an expert after the machine learning model learns data; Figure 6 A flowchart of the present invention in which a machine learning model outputs predicted purchase preferences for each user. DETAILED DESCRIPTION

[0016] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0017] like Figures 1-6 As shown, one of the purposes of the present invention is to provide a method for accurately matching influencers and brands based on machine learning, comprising the following steps: S1. Collect audience data from live broadcasts of influencers and effectively filter audience lists based on viewing time. The steps of S1 are as follows: S1.1. When the influencer enters the live broadcast mode, the audience data of the live broadcast room is collected in real time, and the viewing time of each viewer is calculated in real time based on the audience data; When the influencer starts the live broadcast mode, the monitoring system of the live broadcast platform is used to obtain data on all viewers who enter the live broadcast room during the current live broadcast period, including the audience's entry and exit timestamps and login IDs; For each viewer, the real-time viewing time from the viewer's entry to the current moment is calculated using the collected entry time and the current time monitored in real time; The counting stops when the user exits, or restarts when the user exits and re-enters, until the user's one-time viewing time in the live broadcast room exceeds the effective viewing time.

[0018] S1.2. Based on the set effective viewing time (the effective viewing time threshold is set in the system, which can be determined based on industry experience and past live broadcast data of experts, for example, set to 10 minutes), each viewer's real-time viewing time is then compared with the effective viewing time. If the real-time viewing time exceeds the effective viewing time, the viewer is considered effective. Conversely, if the real-time viewing time does not exceed the effective viewing time, the monitoring continues; Determine that the viewer is a valid viewer and add his / her identification (such as user ID) to the valid viewer list; S1.3. Collect the audience members identified as valid in S1.2 and obtain an audience list based on the collection results; The above judgment is continuously executed during the live broadcast. After the live broadcast ends, the temporary list of valid viewers is compiled and the final audience list is obtained after deduplication.

[0019] S2. Collect the influencer's fan data and compile a fan list based on the fan data. At the same time, obtain the likes and comments data of the influencer's works and compile a potential user list based on the likes and comments data. The steps of S2 are as follows: S2.1. Log in to the influencer's online account and collect fan data in the online account to obtain the influencer's fan data. Collect a fan list based on the fan data to obtain the influencer's corresponding fan list. Utilize the platform's open API interface to obtain structured data related to fans from logged-in influencer online accounts, including fan unique identifiers (such as user IDs), follow-up time, interactive relationship tags, and other information; The collected fan data is cleaned, duplicate and invalid (such as fan records that have been unfollowed but not updated in time) fan records are eliminated, the fan unique identification information is extracted, and it is organized and summarized in a unified format (such as a list, database table, etc.) to form a fan list corresponding to the influencer, which is used for subsequent user profiling, matching analysis and other links.

[0020] S2.2. Use the influencer's online account to collect likes and comments on the influencer's online works. Then, set the likes threshold and comment threshold based on the number of the influencer's online works (calculate the total number of the influencer's online works, and set the likes threshold and comment threshold based on the number of works and industry experience. For example, if the influencer has ten online works, set three likes or three comments). The formula is as follows: ; Among them, T p is the threshold of the number of likes, n is the total number of the talent's online works, p i is the number of likes for the i-th work, α p Adjust the coefficient for likes; ; Among them, T c is the comment count threshold, c i is the number of comments on the i-th work, α c Adjustment factor for comments.

[0021] Through the authorized interface of the influencer's online account, batch obtain all the online works (such as videos, pictures, etc.) published by the influencer. For each work, extract the likes data (including the list of likers' IDs) and comments data (including the comment user ID, content, and timestamp). S2.3. Perform user attribution analysis based on the likes and comments data to obtain the number of likes and comments for each user, and then compare the number of likes and comments for each user with the likes and comments thresholds and comment thresholds; When one of the number of likes and comments of a user exceeds the threshold of the number of likes and the threshold of the number of comments, the user is determined to be a potential user and entered into the potential user list; When the number of likes and comments of a user does not exceed the like threshold and comment threshold, the user will continue to be monitored.

[0022] S3. Generate a network profile for each user based on the audience list, fan list, and potential user list at different time periods. At the same time, summarize the network profiles of the corresponding users of the expert at different time periods and perform data analysis on user differences at different time periods. When S3 generates a network profile, if the same user appears in the audience list, fan list, and potential user list, only one list is assigned to the user, and the assignment priority is as follows: Fans list > audience list; Audience list > potential users.

[0023] If a user appears in both the fan list and other lists, priority will be given to the fan list; If a user does not appear in the fan list but appears in both the audience list and the potential user list, he / she will be assigned to the audience list first; For each user, only the list affiliation with the highest priority is retained, and other list associations with lower priority are discarded; Based on the priority allocation results, reorganize the fan list, audience list, and potential user list to ensure that each user only appears in one list.

[0024] The steps of S3 are as follows: S3.1. Divide the audience list, fan list, and potential user list into time periods based on the collection time. Generate a network profile for each user in the audience list, fan list, and potential user list for each time period. Based on the generated results, obtain the corresponding network profile for each user. The specific steps are as follows: Time period division: Based on the data collection time range (one month), the timeline is divided into several consecutive and non-overlapping time periods. For each user record in the audience list, fan list, and potential user list, it is assigned to the sub-list of the corresponding time period according to the time of their first appearance or interaction (such as the time of entering the live broadcast room, the time of following, and the time of first comment); Profile feature extraction: For each user sublist within each time period, extract user features from multiple dimensions: Behavioral characteristics (watching time, interaction frequency, content preferences, etc.); Social characteristics (following relationships, interaction objects, community attributes, etc.); Network profile generation: Integrate the extracted features of each user in each time period to generate a network profile corresponding to that time period.

[0025] S3.2. Summarize the user's online profile and divide it into time periods to obtain the corresponding profile of each user in different time periods and the total user profile of the influencer in different time periods; The network portraits generated by each user in different time periods are associated according to the user ID to form a time series of each user's portrait. For each time period, the portraits of all users in the time period are summarized to construct the total user portrait (statistical mean) of the expert in the time period.

[0026] S3.3. Perform user difference analysis on the total network profile of all users corresponding to the expert at different time periods to obtain user difference data corresponding to the expert. The user difference data includes the following: Changes in feature values ​​(increases or decreases in average viewing time and interaction frequency); Distribution shift (shift in age distribution and interest preferences); User churn or new additions (analyzing the intersection and complement of user sets at different time periods); The analysis results are organized into user difference data corresponding to the experts (change trend chart).

[0027] S4. Obtain network hotspot data and record changes in each user's network profile. Then, analyze the hotspot impact data for each user based on the network hotspot data. Build a machine learning model based on the network hotspot data, user difference data, and hotspot impact data. The machine learning model outputs the expert's real-time user profile and predicted user profile. The steps of S4 are as follows: S4.1. Obtain network hotspot data for different time periods, and analyze the changes in each user's network profile at different time periods to obtain the corresponding change data of the user's network profile at different time periods. Combine the change data of different time periods with the network hotspot data to perform hotspot impact data analysis to obtain the hotspot impact data corresponding to the user. The specific steps are as follows: Acquisition of online hotspot data: We collect online hot topics, keywords, and their popularity values ​​(such as search volume and discussion volume) from authoritative information platforms, social media, and other channels over different time periods. We also clean and classify the hotspot data (e.g., by industry or event type) to construct hotspot feature vectors. User profile change analysis: This compares the network profiles of the same user over adjacent or specific time periods, calculates the change values ​​of each feature dimension (such as changes in interest tag weights and shifts in consumer preferences), and organizes these change values ​​into a user profile change vector, reflecting the dynamic adjustments in user behavior and preferences. Hotspot Impact Data Analysis: This method performs correlation analysis between user profile change vectors and network hotspot feature vectors for the same period, identifies the impact of hot topics on user behavior (calculates the correlation between hot topics and user profile changes, and quantifies the degree of hotspot impact). This method generates hotspot impact data for each user using the following formula: ; Among them, I j，k is the user's influence factor on the hotspot, Corr is the correlation function, which measures the correlation between the hotspot and the change of user profile, h k is the popularity value of the kth hot topic in time period t, Amp (h k ) is the influence amplifier of the hot spot, ΔP j is the user's portrait change vector from time period t to t+1.

[0028] S4.2. Build a machine learning model based on real-time network hotspot data, change data, user difference data, and hotspot impact data. Then, after learning the data, the machine learning model outputs the real-time user profile and predicted user profile corresponding to the expert. The formula is as follows: ; Among them, X m is the feature vector of the mth input sample, Diff j is the user difference data vector; ; Among them, G output is the total user portrait output by the model, Model is the machine learning model function (LSTM network), and θ is the model parameter set; ; ; Among them, G real-time For the real-time user portrait, X current is the real-time feature vector of the current period, G prediction To predict the total user profile, X history It is a historical multi-period characteristic sequence.

[0029] There is a difference between the user total profile of S3.2 and the real-time user total profile of S4.2. The user total profile of S3.2 is defined as the basic user total profile, and the real-time user total profile of S4.2 is defined as a correction of the basic user total profile.

[0030] S5. Obtain historical order data and match historical purchase records for the corresponding user based on the historical order data. Then, use the machine learning model to analyze the purchase preferences of each user based on their historical purchase records. The steps of S5 are as follows: S5.1. Obtain order history data from the influencer's online account and the brand's sales portal. Match the user's purchase history with the order history data to obtain each user's corresponding purchase history. Through the influencer network account backend and brand sales platforms (such as e-commerce platforms and brand official websites), we obtain user order data, including fields such as order ID, user ID, purchased product information (such as category, brand, price), purchase time, etc., and use the user ID as the index to associate and integrate all order records of the same user to form the user's historical purchase sequence.

[0031] S5.2. Calculate the linkage value between each product purchase, then input each user's historical purchase history and linkage data into a machine learning model, which then outputs each user's predicted purchase preference. Analyze the correlation between adjacent products in the user's purchase sequence: Count product combinations (e.g., "shampoo + conditioner") purchased multiple times by the same user, calculate their co-occurrence frequency, and use the association rule algorithm (Apriori) to quantify the linkage strength between products; Convert each user's historical purchase records (product ID sequence) and linked values ​​(such as association rule weights) into a feature vector that the model can recognize. Input the feature vector into the training model, with the goal of predicting the product categories or brands that the user is likely to purchase in the future. Input the user's latest purchase records and chain data into the trained model, and output the user's preference probability or ranking for various products / brands to form a predicted purchase preference list.

[0032] S6. Analyze the correlation between the influencer's real-time user profile, predicted user profile, and predicted purchase preferences and the brands to be matched. Recommend matching brands to the influencer based on the correlation. The specific steps are as follows: Brand portrait construction: Collect basic information about the brand to be matched, including target audience characteristics (such as age, gender, and geographical distribution), product categories, marketing tone (such as high-end, youthful, and environmentally friendly), and recent promotional keywords; Calculate real-time user profiles and brand relevance: Match the influencer's real-time user profile with the brand profile, compare the overlap between user characteristics and the brand's target audience, and analyze the correlation between recent user interaction hotspots and brand marketing themes. Predicting the long-term fit between the user profile and the brand: This involves assessing the long-term compatibility between the brand and the influencer audience based on the predicted user profile. For example, this involves predicting whether changes in user demand align with the brand's product line upgrades (e.g., increasing user demand for smart homes and the brand's planned launch of related products). Furthermore, the predicted user purchase preferences are compared with the brand's core product categories and SKUs to calculate category fit. Comprehensive relevance rating and recommendation: Set the weights of each dimension to calculate the comprehensive relevance score between influencers and brands. Generate a brand recommendation list in descending order of scores, giving priority to highly relevance brands. The formula is as follows: ; Among them, R real is the cosine similarity between the real-time user portrait and the brand, and B is the brand portrait vector; ; Among them, R pred To predict the trend correlation coefficient between the user's overall portrait and the brand's future positioning, B futur Characteristic vector for the brand’s future strategy; ; Among them, R purchase is the purchase preference matching, E is the product category set, w e is the weight of category e, Match (p e , b e ) is the user’s preferred category p e Brand main category b e Matching function, match is 1, no match is 0; ; Among them, A is the comprehensive score, β1, β2, and β3 are the weight coefficients of each dimension, β1+β2+β3=1.

[0033] A second object of the present invention is to provide a precise matching system for influencers and brands based on machine learning, including any of the above-mentioned precise matching methods for influencers and brands based on machine learning, including a list collection module, a list analysis module, and a brand matching module; The list collection module is used to collect audience data for effective screening of the audience list, collect fan data for fan list collection, obtain likes data and comment data for potential user list collection, generate network portraits for each user based on the audience list, fan list and potential user list in different time periods, and perform user difference data analysis in different time periods; The list analysis module is used to obtain network hotspot data and record the change data of each user's network profile. Then, the hotspot impact data analysis is performed on each user in combination with the network hotspot data. Then, a machine learning model is established based on the network hotspot data, user difference data, and hotspot impact data. The machine learning model outputs the real-time user profile and predicted user profile of the expert. The brand matching module is used to obtain historical order data, and match historical purchase records for corresponding users based on the historical order data. Then, the machine learning model is combined with each user's historical purchase records to perform a predicted purchase preference analysis. The correlation analysis is performed based on the influencer's real-time user profile, predicted user profile, and predicted purchase preferences with the brand to be matched, and brand matching recommendations are made for the influencer based on the correlation.

[0034] The above shows and describes the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The above embodiments and descriptions are merely preferred examples of the present invention and are not intended to limit the present invention. Various changes and improvements may be made to the present invention without departing from the spirit and scope of the present invention. Such changes and improvements fall within the scope of the present invention. The scope of protection claimed in the present invention is defined by the appended claims and their equivalents.

Claims

1. A machine learning-based method for accurately matching influencers and brands, characterized by: The steps include: S1. Collect audience data from live broadcasts of influencers and effectively filter audience lists based on viewing time. S2. Collect the influencer's fan data and compile a fan list based on the fan data. At the same time, obtain the likes and comments data of the influencer's works and compile a potential user list based on the likes and comments data. S3. Generate a network profile for each user based on the audience list, fan list, and potential user list at different time periods. At the same time, summarize the network profiles of the corresponding users of the expert at different time periods and perform data analysis on user differences at different time periods. S4. Obtain network hotspot data and record changes in each user's network profile. Then, analyze the hotspot impact data for each user based on the network hotspot data. Build a machine learning model based on the network hotspot data, user difference data, and hotspot impact data. The machine learning model outputs the expert's real-time user profile and predicted user profile. S5. Obtain historical order data and match historical purchase records for the corresponding user based on the historical order data. Then, use the machine learning model to analyze the purchase preferences of each user based on their historical purchase records. S6. Perform correlation analysis based on the influencer's real-time user profile, predicted user profile, and predicted purchase preferences with the brands to be matched, and make brand matching recommendations for the influencer based on the correlation.

2. The method for accurately matching influencers and brands based on machine learning according to claim 1, characterized in that: The steps of S1 are as follows: S1.

1. When the influencer enters the live broadcast mode, the audience data of the live broadcast room is collected in real time, and the viewing time of each viewer is calculated in real time based on the audience data; S1.

2. Based on the set effective viewing time, the real-time viewing time of each viewer is compared with the effective viewing time. If the real-time viewing time exceeds the effective viewing time, the viewer is considered effective. Otherwise, if the real-time viewing time does not exceed the effective viewing time, the monitoring is continued. S1.

3. Collect the audience members determined as valid in S1.2 and obtain the audience list based on the collection results.

3. The method for accurately matching influencers and brands based on machine learning according to claim 1, characterized in that: The steps of S2 are as follows: S2.

1. Log in to the influencer's online account and collect fan data in the online account to obtain the influencer's fan data. Collect a fan list based on the fan data to obtain the influencer's corresponding fan list. S2.

2. Using the influencer's online account to collect likes and comments on the influencer's online works, and then setting likes and comments thresholds based on the number of influencer's online works; S2.

3. Perform user attribution analysis based on the likes and comments data to obtain the number of likes and comments for each user, and then compare the number of likes and comments for each user with the likes and comments thresholds and comment thresholds; When one of the number of likes and comments of a user exceeds the threshold of the number of likes and the threshold of the number of comments, the user is determined to be a potential user and entered into the potential user list; When the number of likes and comments of a user does not exceed the like threshold and comment threshold, the user will continue to be monitored.

4. The method for accurately matching influencers and brands based on machine learning according to claim 1, characterized in that: When S3 generates a network profile, if the same user appears in the audience list, fan list, and potential user list, only one list is assigned to the user, and the assignment priority is as follows: Fans list > audience list; Audience list > potential users.

5. The method for accurately matching influencers and brands based on machine learning according to claim 1, characterized in that: The steps of S3 are as follows: S3.

1. Divide the audience list, fan list, and potential user list into time periods based on the collection time, generate network profiles for each user in the audience list, fan list, and potential user list for different time periods, and obtain the corresponding network profile for each user based on the generated results; S3.

2. Summarize the user's online profile and divide it into time periods to obtain the corresponding profile of each user in different time periods and the total user profile of the influencer in different time periods; S3.

3. Perform user difference analysis on the total network portraits of all users corresponding to the expert at different time periods to obtain user difference data corresponding to the expert.

6. The method for accurately matching influencers and brands based on machine learning according to claim 1, characterized in that: The steps of S4 are as follows: S4.

1. Obtain network hotspot data for different time periods, and simultaneously analyze the changes in each user's network profile over different time periods to obtain the corresponding change data for the user's network profile over different time periods. Combine this change data with the network hotspot data to perform hotspot impact data analysis to obtain the hotspot impact data corresponding to the user; S4.

2. Establish a machine learning model based on real-time network hotspot data, change data, user difference data, and hotspot impact data. Then, after learning the data, the machine learning model outputs the real-time user total portrait and predicted user total portrait corresponding to the expert.

7. The method for accurately matching influencers and brands based on machine learning according to claim 5, characterized in that: There is a difference between the user total profile of S3.2 and the real-time user total profile of S4.

2. The user total profile of S3.2 is defined as the basic user total profile, and the real-time user total profile of S4.2 is defined as a correction of the basic user total profile.

8. The method for accurately matching influencers and brands based on machine learning according to claim 1, characterized in that: The steps of S5 are as follows: S5.

1. Obtain order history data from the influencer's online account and the brand's sales portal. Match the user's purchase history with the order history data to obtain each user's corresponding purchase history. S5.

2. Calculate the linkage value between each product purchase, then input each user's historical purchase record and linkage data into the machine learning model, and the machine learning model outputs each user's predicted purchase preference.

9. A system for accurately matching influencers and brands based on machine learning, for implementing the method for accurately matching influencers and brands based on machine learning as described in any one of claims 1 to 8, characterized in that: Includes list collection module, list analysis module and brand matching module; The list collection module is used to collect audience data for effective screening of the audience list, collect fan data for fan list collection, obtain likes data and comment data for potential user list collection, generate network portraits for each user based on the audience list, fan list and potential user list in different time periods, and perform user difference data analysis in different time periods; The list analysis module is used to obtain network hotspot data and record the change data of each user's network profile. Then, the hotspot impact data analysis is performed on each user in combination with the network hotspot data. Then, a machine learning model is established based on the network hotspot data, user difference data, and hotspot impact data. The machine learning model outputs the real-time user profile and predicted user profile of the expert. The brand matching module is used to obtain historical order data, and match historical purchase records for corresponding users based on the historical order data. Then, the machine learning model is combined with each user's historical purchase records to perform a predicted purchase preference analysis. The correlation analysis is performed based on the influencer's real-time user profile, predicted user profile, and predicted purchase preferences with the brand to be matched, and brand matching recommendations are made for the influencer based on the correlation.

Images