Live broadcast preference content recommendation method and system based on big data

Abstract

The application discloses a live broadcast preference content recommendation method and system based on big data, relates to the technical field of big data, and is used for solving the problem of low benefits caused by invalid users in live broadcast recommendation, and comprises the following steps: collecting user data and live broadcast history data, obtaining user favorite tags according to the user data and screening out valid users, screening out live broadcasts corresponding to the tags according to the live broadcast history data and classifying the live broadcasts into benefit live broadcasts and public welfare live broadcasts, collecting valid user behavior data, classifying the users into promotion type users and transaction type users, matching and recommending the promotion type users with the public welfare live broadcasts, collecting live broadcast data, analyzing live broadcast promotion requirements by using fuzzy reasoning according to the live broadcast data, matching and recommending the transaction type users with the benefit live broadcasts when the live broadcast promotion requirements are low, collecting matching and recommendation feedback data, constructing a support vector machine to evaluate user driving force by comprehensively considering the user data and the matching and recommendation feedback data, reevaluating valid users according to the user driving force, and improving the problem of low live broadcast recommendation benefits.

Description

Technical Field

[0001] This invention relates to the field of big data technology, and more specifically, to a method and system for recommending live streaming content based on big data preferences. Background Technology

[0002] Big data technology is a digital technology that offers advantages such as enhanced business insights, optimized decision-making, and innovative services. When applied to live streaming recommendations, big data technology can push suitable live streaming content to users with high interest, thereby increasing the effectiveness of live streaming while meeting user needs.

[0003] The existing technology has the following shortcomings:

[0004] In the past, live streaming pushes were based on user preferences, without confirming the user's driving force for the live stream. This resulted in a large amount of traffic but low efficiency in pushing live streaming content, thus leading to low efficiency in live streaming pushes. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a method and system for recommending live streaming content based on big data preferences. By analyzing user activity, live streaming type, and demand, the system rationally plans live streaming pushes to solve the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] The live streaming content recommendation method based on big data includes the following steps:

[0008] Step S1: Collect user data and live streaming history data, filter valid users based on user data and obtain user favorite tags, filter live streams with corresponding tags from live streaming history data and classify live stream types.

[0009] Step S2: Collect valid user behavior data, determine user types, and match and recommend public welfare live streams according to user types;

[0010] Step S3: Collect live streaming data, use fuzzy reasoning to analyze live streaming promotion requirements, and match and recommend profitable live streams according to the live streaming promotion requirements;

[0011] Step S4: Collect matching recommendation feedback data, combine user data and matching recommendation feedback data to construct a support vector machine to evaluate user driving force, and re-evaluate effective users based on user driving force.

[0012] In a preferred embodiment, in step S1, the user data includes user login frequency and percentage of user favorite tags, and the live streaming history data includes the status of the tipping function, the number of product links, and the number of online users in the live streaming room.

[0013] In a preferred embodiment, in step S1, an activity threshold is set based on the user login frequency using the percentile method, and users whose login frequency exceeds the activity threshold are identified as valid users and marked accordingly; the live stream is classified into public welfare live stream and revenue-generating live stream using a binary classification method.

[0014] In a preferred embodiment, in step S2, the effective user behavior data includes user transaction volume and user interaction data, and the user interaction data includes the number of user comments on live streams, the number of user forwards of live streams, and the number of user likes on live streams.

[0015] In a preferred embodiment, in step S2, the number of user live stream comments, the number of user live stream reposts, and the number of user live stream likes are weighted and summed to calculate the user interaction coefficient. When calculating the user interaction coefficient, the control variable method is used to conduct simulation tests to set different interaction weights. The user transaction volume and the user interaction coefficient are compared with the preset transaction threshold and interaction threshold, respectively. Based on the comparison, the user type is determined and classified.

[0016] In a preferred embodiment, in step S2, when determining the user type, users whose transaction volume exceeds the transaction threshold are marked as transactional users; users whose interaction coefficient exceeds the interaction threshold are marked as promotional users; if a user's transaction volume exceeds the transaction threshold and their interaction coefficient also exceeds the interaction threshold, then the user is marked with both classifications; otherwise, no user is classified.

[0017] In a preferred embodiment, in step S3, the live streaming data includes the proportion of transaction volume in the live streaming room and the click-through rate of goods in the live streaming room. The specific steps for analyzing the live streaming promotion requirements using fuzzy inference are as follows:

[0018] The proportion of transaction volume in the live broadcast room and the click-through rate of goods in the live broadcast room are defined as input variables, and the requirements for live broadcast promotion are defined as output variables, which are then divided into different fuzzy sets.

[0019] Formulate fuzzy rules and obtain output results based on the input variables according to the fuzzy rules;

[0020] Based on the output results, determine the live streaming promotion requirements, and push the live stream to users with different category tags according to the live streaming promotion requirements.

[0021] In a preferred embodiment, in step S4, the matching recommendation feedback data consists of changes in the number of online users in the live stream and changes in user login frequency. The specific steps for constructing a support vector machine to assess user driving force are as follows:

[0022] Data preparation: Changes in online users and changes in user login frequency are used as input features; user drive is used as the output result;

[0023] Feature transformation: Selecting a kernel function to calculate the output of the support vector machine;

[0024] Model training: The input features are divided into training and test sets. The loss function is used to optimize the kernel function by using the output results of the support vector machine on the training and test sets and to make the output results converge.

[0025] Evaluation threshold setting: Set the final converged output of the support vector machine as the evaluation threshold;

[0026] Assessing user engagement: The changes in online users and user login frequency in the live stream are compared with an assessment threshold using a support vector machine to assess user engagement.

[0027] The big data-based live streaming preference content recommendation system is used to implement the above-mentioned big data-based live streaming preference content recommendation method, including a data acquisition module, a time update module, a threshold adjustment module, and a feedback adjustment module.

[0028] The data acquisition module is used to collect all the analytical data involved in the method and send them to different subsequent modules for processing.

[0029] The type identification module is used to receive user data and live streaming history data, and to filter and classify users and live streams based on the received data;

[0030] The matching and recommendation module is used to receive live streaming data, analyze live streaming promotion requirements, and match and recommend users and live streams after classification.

[0031] The feedback update module is used to receive matching recommendation feedback data analysis and user-driven updates to user tags.

[0032] The technical effects and advantages of this invention, which is a live streaming preference content recommendation method and system based on big data, are as follows:

[0033] This invention collects user data and live streaming history data. Based on user data, it obtains user-favorite tags and filters out effective users. Based on live streaming history data, it filters live streams with corresponding tags and categorizes them into revenue-generating and charity-related live streams. It collects effective user behavior data to classify users into promotional users and transactional users. Promotional users are matched and recommended with charity live streams. It collects live streaming data and analyzes the live stream promotion requirements. If the promotion requirements are low, only transactional users are matched and recommended with revenue-generating live streams. It collects matching and recommendation feedback data, and comprehensively evaluates user engagement based on user data and matching and recommendation feedback data. Based on user engagement, it re-evaluates effective users. Attached Figure Description

[0034] Figure 1 This is a schematic main image illustrating the live streaming preference content recommendation method based on big data according to the present invention.

[0035] Figure 2 This is a schematic sub-figure illustrating the live streaming preference content recommendation method based on big data according to the present invention.

[0036] Figure 3 This is a flowchart of the live streaming preference content recommendation system based on big data according to the present invention. Detailed Implementation

[0037] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0038] This invention collects user data and live streaming history data. Based on user data, it obtains user-favorite tags and filters out effective users. Based on live streaming history data, it filters live streams with corresponding tags and categorizes them into revenue-generating and charity-related live streams. It collects effective user behavior data to classify users into promotional users and transactional users, matching promotional users with charity-related live streams for recommendations. It collects live streaming data and analyzes the live stream promotion requirements; if the promotion requirements are low, it only matches transactional users with revenue-generating live streams for recommendations. It collects matching and recommendation feedback data, comprehensively assesses user engagement based on user data and matching and recommendation feedback data, and re-evaluates effective users based on user engagement, thereby planning and improving problems such as low live streaming efficiency.

[0039] Example 1: A live streaming content recommendation method based on big data, such as... Figure 1 and Figure 2 As shown, it includes the following steps:

[0040] Step S1: Collect user data and live streaming history data, filter valid users based on user data and obtain user favorite tags, filter live streams with corresponding tags from live streaming history data and classify the live stream types.

[0041] Step S2: Collect valid user behavior data, determine user types, and match and recommend public welfare live streams based on user types.

[0042] Step S3: Collect live streaming data, use fuzzy reasoning to analyze live streaming promotion requirements, and match and recommend profitable live streams according to these requirements.

[0043] Step S4: Collect matching recommendation feedback data, combine user data and matching recommendation feedback data to construct a support vector machine to evaluate user driving force, and re-evaluate effective users based on user driving force.

[0044] The specific implementation is as follows:

[0045] In step S1, user data includes user login frequency and the percentage of user-favorited tags, while live streaming history data includes the status of the tipping function, the number of product links, and the number of online users in the live stream. User data can be obtained from the software management system, and live streaming history data can be obtained from the live streaming history database. When filtering users, a sufficient number of users are selected as sample users, and an activity threshold is set based on user frequency using the percentile method.

[0046] Before setting the activity threshold using the percentile method, user login frequencies are arranged from smallest to largest. A threshold percentage is set, and the user login frequencies in the threshold percentage sequence are used as the activity threshold. For example, if the threshold percentage is set to 70%, then the user login frequencies of more than 70% of the sample users are used as the activity threshold.

[0047] Users whose login frequency exceeds the activity threshold are identified as valid users and marked. The percentage of liked tags of marked users is collected. Selection parameters are set, and the number of liked tags within the selection parameters is used as valid tags. The corresponding live streaming history data is collected. For example, if the selection parameter is set to 3, the top three liked tags with the highest percentage of liked tags are used as valid tags.

[0048] Collect sufficient historical live stream data with valid tag types. Combine the reward function status and the number of product links in the historical live stream data to set the live stream classification threshold. The formula can be: FL = d × h, where FL is the live stream revenue coefficient, d is the reward status variable, and h is the product link coefficient. It should be noted that the reward status variable can be set using binary classification. For example, if the current live stream has the reward function enabled, the reward status variable is set to 1; if the current live stream has the reward function disabled, the reward status variable is set to 0.

[0049] When setting the live streaming category threshold, the average number of live streaming product links with sufficient valid tag types can be used as the product link coefficient, and the reward status variable can be set to 1. The calculated live streaming revenue coefficient can be used as the live streaming category threshold. When calculating the live streaming revenue coefficient for the current live stream, the product link coefficient is the number of product links in the current live stream.

[0050] If the calculated revenue coefficient of the current live stream exceeds the live stream classification threshold, the current live stream will be marked as a revenue-generating live stream; if the calculated revenue coefficient of the current live stream is lower than the live stream classification threshold, the current live stream will be marked as a charity live stream.

[0051] It should be noted that the number of product links refers to the number of product links listed within the live stream. Charity live streams often involve links to charitable goods, which are fewer in number. Therefore, the larger the number of product links listed within a live stream, the greater the corresponding revenue. The software management system is a system used to manage and control the development process. It includes user data that needs to be collected. The live stream history database stores various historical data from the software, including the status of the reward function in historical live streams, the number of product links, and the number of users online. The method for setting the activity threshold and the formula for calculating the live stream revenue coefficient are not unique and can be adjusted according to the actual situation.

[0052] In step S2, user behavior data includes user transaction volume and user interaction data. User interaction data includes the number of user comments on live streams, the number of user reposts on live streams, and the number of user likes on live streams. After filtering and tagging users in step S1, sufficient labeled user behavior data is collected. When determining user types based on labeled user behavior data, the user interaction data needs to be preprocessed. The number of user comments on live streams, the number of user reposts on live streams, and the number of user likes on live streams are assigned different interaction weights, and a weighted sum is calculated to obtain the user interaction coefficient. When setting different interaction weights, simulation tests can be conducted using the controlled variable method.

[0053] Select a period of time as the sample time, use multiple users to conduct single-item tests on the same live stream, compare the maximum number of online viewers during the sample time, and then use the quotient of the maximum number of online viewers for the corresponding single-item test and the total maximum number of online viewers for the test as the corresponding single-item interaction weight. For example, use multiple users to like the same live stream only once, without forwarding or commenting, and collect the maximum number of online viewers during the sample time. Similarly, obtain the maximum number of online viewers for the three single-item test and calculate the corresponding single-item interaction weight.

[0054] Collect sufficient labeled user behavior data, calculate the user interaction coefficient for different users, set interaction thresholds and transaction thresholds using the percentile method, and determine the user type by comparing the current user behavior data with the set thresholds. The user type determination rules in this example are as follows:

[0055] If a user's transaction volume exceeds the transaction threshold, the user will be categorized as a transactional user; if a user's interaction coefficient exceeds the interaction threshold, the user will be categorized as a promotional user; if a user's transaction volume exceeds the transaction threshold and their interaction coefficient exceeds the interaction threshold at the same time, the user will be categorized in both ways; otherwise, the user will not be categorized.

[0056] Charity live streams are matched and recommended based on different user tags. If the current user has promotional user tags, the charity live streams are pushed to the user according to the percentage of the user's favorite tags. If the current user only has transactional user tags, the data of revenue-generating live streams to be recommended are collected and the revenue-generating live streams are filtered and recommended.

[0057] It should be noted that user interaction data exists, but is not limited to, the three types given in this example. The methods for setting interaction weights, relevant thresholds, calculating user interaction coefficients, and determining user types are not unique; the above are merely illustrative examples.

[0058] In step S3, the live streaming data for the revenue-generating live stream includes the transaction amount ratio of the live stream room and the click-through rate of the goods in the live stream room. The transaction amount ratio of the live stream room can be calculated according to the type of goods.

[0059] Collect sufficient total transaction amount and transaction volume of goods with similar tags. Calculate the transaction amount ratio of the goods based on the total transaction amount and transaction volume as the transaction amount ratio of the corresponding tag goods. Sum the transaction amount ratios of all types of tagged goods displayed in the live broadcast room and average them to obtain the transaction amount ratio of the live broadcast room.

[0060] It should be noted that the click-through rate of goods in the live broadcast room, the total transaction amount of goods with similar tags, and the transaction volume of goods can be obtained by accessing the live broadcast backend database. The live broadcast backend database is used to store and manage various information and data in the live broadcast system, including information such as the click-through rate of goods, various product tags, the total transaction amount of goods, and the transaction volume of goods.

[0061] The following steps are used to analyze the requirements for livestream promotion by combining the proportion of transaction volume in the livestream room and the click-through rate of goods in the livestream room:

[0062] The proportion of transaction volume in the live broadcast room and the click-through rate of goods in the live broadcast room are defined as input variables, and they are divided into different fuzzy sets to analyze the revenue of the live broadcast.

[0063] For example, the transaction volume ratio in a live stream can be defined as "High" or "Low", and the click-through rate of goods in a live stream can be defined as "more" or "less".

[0064] Define the requirements for live streaming promotion as output variables and classify them into fuzzy sets.

[0065] For example, the requirements for live streaming promotion can be set as "High" or "Low".

[0066] Formulate fuzzy rules, obtain output results by passing input variables through fuzzy rules, and determine the live streaming promotion requirements for the current revenue stream based on the output results. The fuzzy rules can be set according to the actual situation.

[0067] For example, we can define the percentage of transaction volume in a live stream as J, the click-through rate of goods in a live stream as D, and the requirements for live stream promotion as T.

[0068] Rule 1:IF(J is High)AND(D is more)THEN(T is Low)

[0069] Rule 2:IF(J is Low)AND(D is less)THEN(T is High) ...

[0071] Based on the established fuzzy rules, fuzzy reasoning is used to analyze the promotion requirements of the current revenue-generating live stream. When the output result is "Low", it is determined that the promotion requirements of the current revenue-generating live stream are low, and the current revenue-generating live stream is only pushed to users whose user tags include transactional users. When the output result is "High", it is determined that the promotion requirements of the current revenue-generating live stream are high, and the current revenue-generating live stream is pushed to users whose tags have been classified.

[0072] It should be noted that the fuzzy set classification can be adjusted according to the actual situation. For example, if the transaction volume ratio of the current revenue-generating live stream exceeds 0.6, it is judged as having a high transaction volume ratio and is labeled "High"; if the product click-through rate of the current revenue-generating live stream exceeds 0.7, it is judged as having a high product click-through rate and is labeled "More". When the output result is "Low", the current revenue-generating live stream has a high transaction volume ratio and a high product click-through rate, indicating a lower demand for live stream promotion and more significant benefits when pushed to transactional users. Furthermore, this example only classifies fuzzy sets into two types; the types of fuzzy sets can be adjusted or further classified according to the actual situation to improve the accuracy of the output results.

[0073] In step S4, the matching recommendation feedback data includes the change in the number of online users in the live stream and the change in user login frequency. The number of online users in the same live stream and the user login frequency of the current live stream being matched and recommended are randomly selected at time points before and after the matching recommendation. The change in the number of online users in the live stream after the matching recommendation is calculated by subtracting the number of online users in the live stream before the matching recommendation. The change in user login frequency is calculated by subtracting the user login frequency after the current live stream being matched and recommended from the user login frequency before the matching recommendation.

[0074] Multiple sets of matching recommendation time points before and after the recommendation were randomly selected to obtain the changes in online users and user login frequency in the live stream, which were then merged into online user change datasets and login change datasets, respectively. A support vector machine (SVM) model was constructed using the changes in online users and user login frequency as input variables. The user-driven performance was evaluated using the output of the SVM model. The specific steps are as follows:

[0075] Data preparation: Use the online change dataset and the login change dataset as input features; use user drive as the output.

[0076] Feature transformation: The radial basis function is chosen as the kernel function to map the input features to a high-dimensional feature space. The expression for the radial basis function is as follows: in For the output of the support vector machine, x and γ is two random input vectors from the online change dataset and the login change dataset, where e is the natural base and γ is a hyperparameter used to control the width of the kernel function.

[0077] Model Training: Step A1, Data Preprocessing: Remove outliers and normalize input features. The online or logged-in change datasets can be sorted from smallest to largest, and the mean and standard deviation can be calculated. The sum of the mean and the two standard deviations is used as a selection factor and compared with the data in the dataset to remove data exceeding the selection factor. Min-Max normalization can be used to normalize the data, that is, the data in the online or logged-in change datasets are normalized using the Min-Max normalization formula: Where, x nore For the normalized result of the corresponding dataset, x max x represents the data with the largest value in the corresponding dataset. min This refers to the data with the smallest value in the corresponding dataset.

[0078] Step A2: Divide the dataset into training and testing sets. Randomly divide the online and login change datasets into training and testing sets, for example, by dividing the datasets in a 7:3 ratio. Then, randomly select a set of data from each dataset and perform radial basis function calculations to obtain the support vector machine outputs for the training and testing sets.

[0079] Step A3, function optimization: The hinge loss function is selected to optimize the support vector machine (SVM). The hinge loss formula is as follows: L(y, f(x)) = max(0, 1 - yf(x)), where L(y, f(x)) is the kernel loss, y is the SVM output calculated on the test set, and f(x) is the SVM output calculated on the training set. The SVM output is then converged based on the kernel loss. The converged result is used as the user motivation rating coefficient. The convergence formula is as follows: in The user-driven rating coefficient, This is the output of the support vector machine before convergence.

[0080] Evaluation threshold setting: Set different hyperparameters and calculate multiple user drive rating coefficients, and take the average value as the evaluation threshold.

[0081] Assessing user drive: The changes in the number of online users in the current live stream and the changes in the login frequency of the matched recommended users are used as inputs. The average of the hyperparameters set when calculating the user drive assessment coefficient is taken and the result of the support vector machine calculation obtained through the radial basis function is compared with the assessment threshold to determine the user drive.

[0082] If the support vector machine calculation result exceeds the evaluation threshold, it is determined that the current user has high user driving force and the matching recommendation effect is good; if the support vector machine calculation result is lower than the evaluation threshold, it is determined that the current user has low user driving force, all labels of the current user are removed, a label update time is set, and the current user data is collected again after the label update time. Based on the user data analysis, it is determined whether the current user should be regarded as a valid user and labeled.

[0083] It should be noted that the kernel function and loss function in support vector machines are not unique and can be adjusted or switched according to the actual situation. The label update time is set by professionals in this field and will not be elaborated here.

[0084] Example 2: A live streaming preference content recommendation system based on big data, such as... Figure 3 As shown, the method for recommending live streaming content based on big data includes a data acquisition module, a type identification module, a matching recommendation module, and a feedback update module.

[0085] The data acquisition module is used to collect all the analytical data involved in the method and send them to different subsequent modules for processing.

[0086] The type identification module is used to receive user data and live streaming history data, and to filter and classify users and live streams based on the received data;

[0087] The matching and recommendation module is used to receive live streaming data, analyze live streaming promotion requirements, and match and recommend users and live streams after classification.

[0088] The feedback update module is used to receive matching recommendation feedback data analysis and user-driven updates to user tags.

[0089] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented using software, the above embodiments can be implemented, in whole or in part, in the form of a computer program product.

[0090] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and inventive constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0091] In addition, the functional modules in the various embodiments of this application can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module.

[0092] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0093] In conclusion, the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A live streaming content recommendation method based on big data, characterized in that, Includes the following steps, Step S1: Collect user data and live streaming history data, filter valid users based on user data and obtain user favorite tags, filter live streams with corresponding tags from live streaming history data and classify live stream types. Step S2: Collect valid user behavior data, which includes user transaction volume and user interaction data, including the number of user comments on live streams, the number of user reposts on live streams, and the number of user likes on live streams. The user interaction coefficient is calculated by weighting and summing user live stream comments, user live stream reposts, and user live stream likes with different interaction weights. When calculating the user interaction coefficient, the controlled variable method is used to conduct simulation tests to set different interaction weights. The user transaction volume and user interaction coefficient are compared with the preset transaction threshold and interaction threshold, respectively. Based on the comparison, the user type is determined and classified. When determining user types, users whose transaction volume exceeds the transaction threshold are marked as transactional users; users whose interaction coefficient exceeds the interaction threshold are marked as promotional users; if a user's transaction volume exceeds the transaction threshold and their interaction coefficient also exceeds the interaction threshold, they are marked with both categories; otherwise, no user is categorized; and public welfare live streams are matched and recommended based on user type. Step S3: Collect live streaming data, which includes the transaction volume ratio and product click-through rate in the live streaming room. Analyze the live streaming promotion requirements using fuzzy inference. The specific steps are as follows: Define the transaction volume ratio and product click-through rate in the live streaming room as input variables, and define the live streaming promotion requirements as output variables, dividing them into different fuzzy sets; formulate fuzzy rules and obtain the output results through the input variables according to the fuzzy rules; determine the live streaming promotion requirements based on the output results, and push the live stream to users with different category tags according to the live streaming promotion requirements. Match and recommend profitable live streams according to the live stream promotion requirements; Step S4: Collect matching recommendation feedback data, combine user data and matching recommendation feedback data to construct a support vector machine to evaluate user driving force, and re-evaluate effective users based on user driving force.

2. The live streaming preference content recommendation method based on big data according to claim 1, characterized in that: In step S1, user data includes user login frequency and percentage of user favorite tags, and live streaming history data includes the status of the tipping function, the number of product links, and the number of online users in the live streaming room.

3. The live streaming preference content recommendation method based on big data according to claim 2, characterized in that: In step S1, an activity threshold is set based on the user login frequency using the percentile method. Users whose login frequency exceeds the activity threshold are considered valid users and are marked accordingly. The live stream is then classified into public welfare live streams and revenue-generating live streams using a binary classification method.

4. The live streaming preference content recommendation method based on big data according to any one of claims 1-3, characterized in that: In step S4, the matching recommendation feedback data consists of changes in the number of online users in the live stream and changes in user login frequency. The specific steps for constructing a support vector machine to assess user driving force are as follows: Data preparation: Changes in online users and changes in user login frequency were used as input features; User-driven outcomes are taken as the output. Feature transformation: Selecting a kernel function to calculate the output of the support vector machine; Model training: The input features are divided into training and test sets. The loss function is used to optimize the kernel function by using the output results of the support vector machine on the training and test sets and to make the output results converge. Evaluation threshold setting: Set the final converged output of the support vector machine as the evaluation threshold; Assessing user engagement: The changes in online users and user login frequency in the live stream are compared with an assessment threshold using a support vector machine to assess user engagement.

5. A live streaming preference content recommendation system based on big data, based on the live streaming preference content recommendation method based on big data as described in any one of claims 1-4, characterized in that, It includes a data acquisition module, a type identification module, a matching and recommendation module, and a feedback and update module; The data acquisition module is used to collect all the analytical data involved in the method and send them to different subsequent modules for processing. The type identification module is used to receive user data and live streaming history data, and to filter and classify users and live streams based on the received data; The matching and recommendation module is used to receive live streaming data, analyze live streaming promotion requirements, and match and recommend users and live streams after classification. The feedback update module is used to receive matching recommendation feedback data analysis and user-driven updates to user tags.

Images