Big data-based live broadcast room crowd portrait generation method and system
Through big data processing and deep learning technology, people portraits in the live broadcast room are generated, which solves the accuracy of the live broadcast platform in data analysis, achieves an in-depth understanding of user interests and needs, and improves user experience and retention rates.
Patent Information
- Application Number
- CN202311202178.X
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2023-09-18
- Publication Date
- 2024-11-08
- Estimated Expiration
- 2043-09-18
AI Technical Summary
When live broadcast platforms process and analyze complex user data, it is difficult for them to accurately generate crowd portraits, resulting in the inability to effectively understand user needs and interests.
Using a big data-based method, data cleaning and structured representation are carried out by obtaining user basic information, interactive data and consumption data, and semantic analysis is carried out in combination with deep learning technology to generate multi-scale fusion correlation feature vectors of user information, and finally generate a population portrait.
It has achieved a more comprehensive understanding of users' interests, needs and behavior patterns, improved user experience and retention, and provided personalized content recommendations and precise advertising delivery.
Smart Images

Figure CN118916731A_ABST
Abstract
Description
Technical Field
[0001] The present disclosure relates to the field of crowd portrait generation, and more specifically, to a method and system for generating crowd portraits in a live broadcast room based on big data. Background Art
[0002] The live broadcast industry is booming, and the number of users on live broadcast platforms is huge. Generating crowd portraits in live broadcast rooms can better understand live broadcast users, thereby providing personalized content recommendations, accurate advertising, refined user management and other services. Through crowd portraits, live broadcast platforms can better grasp user needs and interests, provide live broadcast content that meets user preferences, and improve user experience and retention rate.
[0003] However, the large amount of data left behind during the operation of the live broadcast platform is usually complex and disorganized. How to process and analyze this data and accurately generate a portrait of the crowd in the live broadcast room is an important technical issue. Summary of the invention
[0004] In view of this, the present disclosure proposes a method and system for generating crowd portraits in a live broadcast room based on big data, which can more comprehensively understand users' interests, needs and behavior patterns.
[0005] According to one aspect of the present disclosure, a method for generating a crowd portrait in a live broadcast room based on big data is provided, which includes:
[0006] Obtain the basic information, interaction data and consumption data of the live broadcast users being analyzed;
[0007] Performing data cleaning and structured representation on the basic information, interaction data and consumption data of the analyzed live broadcast user to obtain a user basic information coding vector, a user interaction coding vector and a user consumption coding vector;
[0008] Performing semantic analysis on the user basic information coding vector, the user interaction coding vector, and the user consumption coding vector to obtain a multi-scale fusion correlation feature vector of user information; and
[0009] Based on the multi-scale fusion associated feature vector of the user information, a crowd portrait is generated.
[0010] According to another aspect of the present disclosure, a system for generating a crowd portrait in a live broadcast room based on big data is provided, which includes:
[0011] The data acquisition module is used to obtain the basic information, interaction data and consumption data of the live broadcast user being analyzed;
[0012] A data cleaning and structured representation module, used for performing data cleaning and structured representation on the basic information, interaction data and consumption data of the analyzed live broadcast user to obtain a user basic information coding vector, a user interaction coding vector and a user consumption coding vector;
[0013] A semantic analysis module, configured to perform semantic analysis on the user basic information coding vector, the user interaction coding vector, and the user consumption coding vector to obtain a multi-scale fusion correlation feature vector of user information; and
[0014] The crowd portrait generation module is used to generate a crowd portrait based on the multi-scale fusion associated feature vector of the user information.
[0015] According to the embodiment of the present disclosure, it first obtains the basic information, interaction data and consumption data of the analyzed live broadcast user, then performs data cleaning and structured representation on the basic information, interaction data and consumption data of the analyzed live broadcast user to obtain the user basic information coding vector, user interaction coding vector and user consumption coding vector, then performs semantic analysis on the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain the user information multi-scale fusion associated feature vector, and finally, generates a crowd portrait based on the user information multi-scale fusion associated feature vector. In this way, the user's interests, needs and behavior patterns can be more comprehensively understood. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] The accompanying drawings, which are incorporated in and constitute a part of the specification, illustrate exemplary embodiments, features, and aspects of the disclosure and, together with the description, serve to explain the principles of the disclosure.
[0017] Figure 1 A flowchart of a method for generating a crowd portrait in a live broadcast room based on big data according to an embodiment of the present disclosure is shown.
[0018] Figure 2 A schematic diagram of the architecture of a method for generating a crowd portrait in a live broadcast room based on big data according to an embodiment of the present disclosure is shown.
[0019] Figure 3 A flowchart of sub-step S130 of the method for generating a crowd portrait in a live broadcast room based on big data according to an embodiment of the present disclosure is shown.
[0020] Figure 4 A block diagram of a live broadcast room crowd portrait generation system based on big data according to an embodiment of the present disclosure is shown.
[0021] Figure 5 An application scenario diagram of a method for generating crowd portraits in a live broadcast room based on big data according to an embodiment of the present disclosure is shown. DETAILED DESCRIPTION
[0022] The technical solutions in the embodiments of the present disclosure will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present disclosure, rather than all embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without creative work also fall within the scope of protection of the present disclosure.
[0023] As shown in the present disclosure and claims, unless the context clearly indicates an exception, the words "a", "an", "an" and / or "the" do not refer to the singular and may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list. The method or device may also include other steps or elements.
[0024] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.
[0025] In addition, in order to better illustrate the present disclosure, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present disclosure can also be implemented without certain specific details. In some examples, methods, means, components and circuits well known to those skilled in the art are not described in detail in order to highlight the subject matter of the present disclosure.
[0026] In response to the above technical problems, the technical concept of the present invention is to use deep learning technology, combined with users' basic information, interaction data and consumption data, to realize the intelligent generation of crowd portraits in the live broadcast room.
[0027] Based on this, Figure 1 A flowchart of a method for generating a crowd portrait in a live broadcast room based on big data according to an embodiment of the present disclosure is shown. Figure 2 The following is a schematic diagram showing the architecture of a method for generating a crowd portrait of a live broadcast room based on big data according to an embodiment of the present disclosure. Figure 1 and Figure 2As shown, according to the big data-based method for generating a crowd portrait in a live broadcast room according to an embodiment of the present disclosure, the method includes the following steps: S110, obtaining the basic information, interaction data and consumption data of the analyzed live broadcast user; S120, performing data cleaning and structured representation on the basic information, interaction data and consumption data of the analyzed live broadcast user to obtain a user basic information coding vector, a user interaction coding vector and a user consumption coding vector; S130, performing semantic analysis on the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain a multi-scale fusion associated feature vector of user information; and, S140, generating a crowd portrait based on the multi-scale fusion associated feature vector of user information.
[0028] Specifically, in the technical solution disclosed in the present invention, first, the basic information, interaction data and consumption data of the analyzed live broadcast user are obtained; and the basic information, interaction data and consumption data of the analyzed live broadcast user are cleaned and structured to obtain the user basic information coding vector, user interaction coding vector and user consumption coding vector. It should be understood that the basic information, interaction data and consumption data of the live broadcast user provide information of multiple dimensions, including the user's personal characteristics, behavioral preferences and consumption habits. By comprehensively analyzing this information, the user's interests, needs and behavior patterns can be more comprehensively understood.
[0029] Then, semantic analysis is performed on the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain a multi-scale fusion correlation feature vector of user information. That is, the feature distribution of the user's interests and behavior patterns is mined from the basic information, interaction data and consumption data of the live broadcast user being analyzed.
[0030] In a specific example of the present disclosure, a semantic analysis is performed on the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain an encoding process of a user information multi-scale fusion association feature vector, including: first, the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector are passed through a user information global association feature extractor based on a fully connected layer to obtain a user information global association feature vector; at the same time, the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector are passed through a user information context semantic association feature extractor based on a converter to obtain a user information context semantic association feature vector; then, the mutual gain between the user information global association feature vector and the user information context semantic association feature vector is calculated to obtain a user information multi-scale fusion association feature vector.
[0031] Accordingly, if Figure 3As shown, semantic analysis is performed on the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain a multi-scale fusion association feature vector of user information, including: S131, extracting the global association features of user information between the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain a global association feature vector of user information; S132, extracting the contextual semantic association features between the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain a contextual semantic association feature vector of user information; and, S133, fusing the global association feature vector of user information and the contextual semantic association feature vector of user information to obtain a multi-scale fusion association feature vector of user information. It should be understood that the generation of the multi-scale fusion association feature vector of user information is performed through the three steps of S131, S132 and S133. In step S131, the correlation features between the user basic information coding vector, the user interaction coding vector and the user consumption coding vector are extracted. These features may include the user's basic attributes, the user's interactive behavior on social media, and the user's consumption behavior. The extracted correlation features can help understand the user's overall behavior pattern and preference. In step S132, the contextual semantic correlation features between the user basic information coding vector, the user interaction coding vector and the user consumption coding vector are extracted. The contextual semantic correlation features may include the user's behavior characteristics in different environments, such as the user's behavior changes in different times, places or social circles. These features can help understand the contextual relevance of user behavior. In step S133, the user information global correlation feature vector and the user information contextual semantic correlation feature vector are fused together to generate a multi-scale fused correlation feature vector of user information. This vector integrates the user's overall behavior pattern and preference and the contextual relevance of user behavior, and provides a more comprehensive user information representation. This multi-scale fused correlation feature vector can be used in subsequent user analysis, personalized recommendation and other tasks.
[0032] More specifically, in step S131, extracting the user information global correlation features between the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain the user information global correlation feature vector includes: passing the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector through a user information global correlation feature extractor based on a fully connected layer to obtain the user information global correlation feature vector. It is worth mentioning that the fully connected layer is one of the most commonly used layer types in a deep neural network. Its function is to connect each input neuron with each output neuron to form a fully connected network structure. The main function of the fully connected layer is to learn the complex nonlinear relationships between input data and encode these relationships as feature representations. In the user information global correlation feature extractor, the fully connected layer is used to combine and fuse the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to generate a global correlation feature vector of user information. Through the fully connected layer, the network can learn the weights and relationships between different input features, thereby capturing higher-level abstract features. These abstract features can better represent the user's overall behavior patterns and preferences, helping the model to better understand the user's characteristics and behaviors. The fully connected layer usually includes weight parameters and bias parameters, which are learned and updated through the back propagation algorithm during the training process to minimize the prediction error of the model. The output of the fully connected layer can be used as the input of the next layer or as the final output result for different tasks and application scenarios.
[0033] More specifically, in step S132, extracting the contextual semantic association features between the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain the user information contextual semantic association feature vector includes: passing the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector through a converter-based user information contextual semantic association feature extractor to obtain the user information contextual semantic association feature vector. It is worth mentioning that in the user information contextual semantic association feature extractor, a converter-based method is used to extract the contextual semantic association feature vector of user information. The converter refers to the Transformer model, which is a neural network structure based on the self-attention mechanism. The main function of the converter is to encode and model the input sequence to capture the contextual relationship and semantic information between the elements in the sequence. In the user information contextual semantic association feature extractor, the converter is used to process the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to extract the contextual semantic association features between them. Through the self-attention mechanism of the converter, the model can globally pay attention to and interact with the input vector, thereby capturing the correlation and importance between different elements. The converter gradually models and extracts features from the input through multiple stacked self-attention layers and feedforward neural network layers to form contextual semantic association feature vectors. The advantage of the converter is that it can handle long sequences and model global dependencies, while having good parallel computing performance. It can learn complex patterns and relationships in the input sequence and extract richer contextual semantic association features, which helps to better understand the contextual relevance of user behavior. These feature vectors can be used for subsequent tasks.
[0034] More specifically, in step S133, the user information global correlation feature vector and the user information context semantic correlation feature vector are fused to obtain the user information multi-scale fusion correlation feature vector, including: calculating the mutual gain between the user information global correlation feature vector and the user information context semantic correlation feature vector to obtain the user information multi-scale fusion correlation feature vector. It is worth mentioning that mutual gain refers to the evaluation of the gain degree provided by different feature vectors in the fused feature by comparing the correlation and complementarity between them in the fused feature in the multi-scale feature fusion. In the step, the mutual gain is used to calculate the complementarity and correlation between the user information global correlation feature vector and the user information context semantic correlation feature vector to obtain the multi-scale fusion correlation feature vector of the user information. Specifically, the mutual gain can be calculated in different ways, and common methods include mutual information, correlation coefficient, etc. These methods can measure the correlation and complementarity between two feature vectors. By calculating the mutual gain, the contribution and importance of the user information global correlation feature vector and the user information context semantic correlation feature vector in multi-scale fusion can be understood. This helps determine how to appropriately fuse these feature vectors to obtain a more comprehensive and accurate multi-scale fused correlation feature vector of user information.
[0035] Further, the multi-scale fusion associated feature vector of the user information is passed through a crowd portrait generator based on AIGC to generate a crowd portrait. Accordingly, in step S140, based on the multi-scale fusion associated feature vector of the user information, a crowd portrait is generated, including: passing the multi-scale fusion associated feature vector of the user information through a crowd portrait generator based on AIGC to generate a crowd portrait. It is worth mentioning that AIGC (Adaptive Information Granularity Clustering) is a method based on adaptive information granularity clustering for generating crowd portraits. AIGC can take the multi-scale fusion associated feature vector of user information as input, and divide users into different groups or categories by adaptive information granularity clustering. It clusters users with similar features together to form different crowd groups by analyzing the similarities and differences between feature vectors. In the clustering process, AIGC will adaptively adjust the information granularity of the clustering according to the characteristics and distribution of the data to adapt to different data features, so as to better capture the subtle differences and common features between user groups. The crowd portrait generated by AIGC can provide an in-depth understanding and description of the user group.
[0036] Further, in the technical solution disclosed in the present invention, the method for generating a crowd portrait in a live broadcast room based on big data is characterized in that it also includes a training step: training the user information global correlation feature extractor based on the fully connected layer, the user information context semantic correlation feature extractor based on the converter, and the crowd portrait generator based on AIGC. It should be understood that the training step plays a key role in the method for generating a crowd portrait in a live broadcast room based on big data. By training the user information global correlation feature extractor based on the fully connected layer, the user information context semantic correlation feature extractor based on the converter, and the crowd portrait generator based on AIGC, the following purposes can be achieved: 1. Parameter optimization: The training step can optimize the parameters of the model so that the feature extractor and the crowd portrait generator can more accurately extract and generate the features of the user information. By repeatedly iterating and optimizing the training data, the model can learn better feature representation and generation capabilities. 2. Improve accuracy: Through training, the feature extractor and the crowd portrait generator can learn patterns and rules that are more suitable for actual data, which can improve the accuracy of the generated crowd portrait and better reflect the characteristics and behaviors of users. 3. Enhanced generalization ability: The training step can help the model learn more generalized feature representation and generation methods, so that it can also perform well on unseen data. This improves the generalization ability of the model and enables it to adapt to different live broadcast rooms and different users. Through the training step, the model can be continuously improved and optimized to better adapt to actual application scenarios and generate more accurate and comprehensive crowd portraits.
[0037] More specifically, the training step includes: obtaining training data, the training data including basic training information, training interaction data and training consumption data of the analyzed live broadcast user, as well as the true value of the crowd portrait; performing data cleaning and structured representation on the basic training information, training interaction data and training consumption data of the analyzed live broadcast user to obtain a training user basic information encoding vector, a training user interaction encoding vector and a training user consumption encoding vector; passing the training user basic information encoding vector, the training user interaction encoding vector and the training user consumption encoding vector through the fully connected layer-based training user information global association feature extractor to obtain a training user information global association feature vector; passing the training user basic information encoding vector, the training user interaction encoding vector and the training user consumption encoding vector through the converter-based user information context semantic association feature extractor to obtain Training user information context semantic association feature vector; calculating the mutual gain between the training user information global association feature vector and the training user information context semantic association feature vector to obtain a training user information multi-scale fusion association feature vector; optimizing the feature distribution of the training user information multi-scale fusion association feature vector to obtain an optimized training user information multi-scale fusion association feature vector; passing the optimized training user information multi-scale fusion association feature vector through the AIGC-based crowd portrait generator to obtain a training generated crowd portrait; calculating the cross entropy function value between the training generated crowd portrait and the true value of the crowd portrait; and using the cross entropy function value as a loss function value to train the fully connected layer-based user information global association feature extractor, the converter-based user information context semantic association feature extractor, and the AIGC-based crowd portrait generator.
[0038] In the technical solution disclosed in the present invention, since the training user information global association feature vector and the training user information context semantic association feature vector are respectively obtained by the training user basic information encoding vector, the training user interaction encoding vector and the training user consumption encoding vector through a user information global association feature extractor based on a fully connected layer and a user information context semantic association feature extractor based on a converter, the training user information global association feature vector and the training user information context semantic association feature vector will respectively have semantic feature representations corresponding to the local feature distribution scales of each initial encoding vector. Therefore, by calculating the mutual gain between the training user information global association feature vector and the training user information context semantic association feature vector, the The multi-scale fusion associated feature vector of the training user information will also have a semantic feature representation corresponding to the local feature distribution scale of each initial coding vector. In this way, when the multi-scale fusion associated feature vector of the training user information passes through the crowd portrait generator based on AIGC, it will also perform a scale-inspired distribution probability density mapping based on the local feature distribution scale to obtain the training generated crowd portrait. However, considering that under the local feature distribution scale of the multi-scale fusion associated feature vector of the training user information, it will simultaneously include the global associated semantic features of the vector value granularity of the initial vector and the contextual associated features of the vector granularity, that is, it will include mixed granularity semantic feature representations under the local feature distribution scale, which will reduce the training efficiency of the crowd portrait generator based on AIGC.
[0039] Based on this, during the training process, the applicant of the present disclosure generates a crowd portrait by passing the multi-scale fused associated feature vector of the training user information through an AIGC-based crowd portrait generator, and performs homogenized activation of the semantic information expressed by the feature rank of the multi-scale fused associated feature vector of the training user information.
[0040] Accordingly, in a specific example, the feature distribution of the training user information multi-scale fusion associated feature vector is optimized to obtain the optimized training user information multi-scale fusion associated feature vector, including: the feature distribution of the training user information multi-scale fusion associated feature vector is optimized according to the following optimization formula to obtain the optimized training user information multi-scale fusion associated feature vector; wherein the optimization formula is:
[0041]
[0042] Where V is the multi-scale fusion associated feature vector of the training user information, v iis the i-th eigenvalue of the multi-scale fusion associated feature vector of the training user information, ||V||2 represents the bi-norm of the multi-scale fusion associated feature vector of the training user information, log is the logarithmic function with base 2, and α is the weight hyperparameter, v' i It is the i-th eigenvalue of the multi-scale fusion associated feature vector of the optimized training user information.
[0043] Here, considering the feature distribution mapping of the feature distribution of the multi-scale fusion associated feature vector V of the training user information from the high-dimensional feature space to the generated probability density mapping space, different mapping modes will be presented at different feature distribution levels based on the mixed granularity semantic features, resulting in the inability of the scale-heuristic mapping strategy to obtain the optimal efficiency. Therefore, the rank expression semantic information based on the feature vector norm is homogenized instead of the scale for feature matching, which can activate similar feature rank expressions in a similar manner and reduce the correlation between feature rank expressions with large differences, thereby solving the problem of low efficiency of probability expression mapping of the feature distribution of the multi-scale fusion associated feature vector i of the training user information under different spatial rank expressions, and improving the training efficiency of the AIGC-based crowd portrait generator.
[0044] In summary, the method for generating crowd portraits in a live broadcast room based on big data according to the embodiment of the present disclosure can more comprehensively understand the interests, needs and behavior patterns of users.
[0045] Figure 4 FIG. 1 is a block diagram of a live broadcast crowd portrait generation system 100 based on big data according to an embodiment of the present disclosure. Figure 4 As shown, according to the big data-based live broadcast room crowd portrait generation system 100 of the embodiment of the present disclosure, it includes: a data acquisition module 110, used to obtain the basic information, interaction data and consumption data of the analyzed live broadcast user; a data cleaning and structured representation module 120, used to perform data cleaning and structured representation on the basic information, interaction data and consumption data of the analyzed live broadcast user to obtain a user basic information coding vector, a user interaction coding vector and a user consumption coding vector; a semantic analysis module 130, used to perform semantic analysis on the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain a multi-scale fusion associated feature vector of user information; and a crowd portrait generation module 140, used to generate a crowd portrait based on the multi-scale fusion associated feature vector of user information.
[0046] In a possible implementation, the semantic analysis module 120 includes: a global correlation feature extraction unit, used to extract the global correlation features of user information between the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain a global correlation feature vector of user information; a semantic correlation feature extraction unit, used to extract the contextual semantic correlation features between the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain a contextual semantic correlation feature vector of user information; and a fusion unit, used to fuse the user information global correlation feature vector and the user information contextual semantic correlation feature vector to obtain the user information multi-scale fused correlation feature vector.
[0047] Here, those skilled in the art can understand that the specific functions and operations of the various units and modules in the above-mentioned live broadcast room crowd portrait generation system 100 based on big data have been referred to above. Figures 1 to 3 It has been introduced in detail in the description of the live broadcast room crowd portrait generation method based on big data, and therefore, its repeated description will be omitted.
[0048] As described above, the live broadcast room crowd portrait generation system 100 based on big data according to the embodiment of the present disclosure can be implemented in various wireless terminals, such as a server with a live broadcast room crowd portrait generation algorithm based on big data. In one possible implementation, the live broadcast room crowd portrait generation system 100 based on big data according to the embodiment of the present disclosure can be integrated into the wireless terminal as a software module and / or a hardware module. For example, the live broadcast room crowd portrait generation system 100 based on big data can be a software module in the operating system of the wireless terminal, or it can be an application developed for the wireless terminal; of course, the live broadcast room crowd portrait generation system 100 based on big data can also be one of the many hardware modules of the wireless terminal.
[0049] Alternatively, in another example, the live broadcast room crowd portrait generation system 100 based on big data and the wireless terminal may also be separate devices, and the live broadcast room crowd portrait generation system 100 based on big data may be connected to the wireless terminal via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.
[0050] Figure 5 The following is a diagram showing an application scenario of a method for generating a crowd portrait in a live broadcast room based on big data according to an embodiment of the present disclosure. Figure 5 As shown, in this application scenario, first, the basic information of the live broadcast user to be analyzed is obtained (for example, Figure 5 D1 as shown in the figure), interactive data (e.g. Figure 5 D2) and consumption data (e.g. Figure 5 Then, the basic information, interaction data and consumption data of the analyzed live broadcast user are input into a server that is deployed with a live broadcast room crowd portrait generation algorithm based on big data (for example, Figure 5 S) as shown in , wherein the server is able to use the live broadcast room crowd portrait generation algorithm based on big data to process the basic information, interaction data and consumption data of the analyzed live broadcast users to generate a crowd portrait.
[0051] The flow chart and block diagram in the accompanying drawings show the possible architecture, function and operation of the system, method and computer program product according to multiple embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a part of a module, program segment or instruction, and a part of the module, program segment or instruction includes one or more executable instructions for realizing the specified logical function. In some alternative implementations, the function marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two continuous square boxes can actually be executed substantially in parallel, and they can sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs the specified function or action, or can be implemented with a combination of special hardware and computer instructions.
[0052] The embodiments of the present disclosure have been described above, and the above description is exemplary, not exhaustive, and is not limited to the disclosed embodiments. Many modifications and changes will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The selection of terms used herein is intended to best explain the principles of the embodiments, practical applications, or improvements to the technology in the market, or to enable other persons of ordinary skill in the art to understand the embodiments disclosed herein.
Claims
1. A method for generating crowd portraits in a live broadcast room based on big data, characterized in that: include: Obtain the basic information, interaction data and consumption data of the live broadcast users being analyzed; Performing data cleaning and structured representation on the basic information, interaction data and consumption data of the analyzed live broadcast user to obtain a user basic information coding vector, a user interaction coding vector and a user consumption coding vector; Performing semantic analysis on the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain a multi-scale fusion correlation feature vector of user information; as well as Based on the multi-scale fusion associated feature vector of the user information, a crowd portrait is generated.
2. The method for generating crowd portraits in a live broadcast room based on big data according to claim 1 is characterized in that: Performing semantic analysis on the user basic information coding vector, the user interaction coding vector, and the user consumption coding vector to obtain a multi-scale fusion correlation feature vector of user information, including: Extracting user information global correlation features between the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain a user information global correlation feature vector; Extracting contextual semantic association features between the user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector to obtain a user information contextual semantic association feature vector; and The user information global association feature vector and the user information context semantic association feature vector are fused to obtain the user information multi-scale fused association feature vector.
3. The method for generating crowd portraits in a live broadcast room based on big data according to claim 2 is characterized in that: Extracting the user information global correlation feature between the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain the user information global correlation feature vector includes: The user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector are passed through a user information global correlation feature extractor based on a fully connected layer to obtain the user information global correlation feature vector.
4. The method for generating crowd portraits in a live broadcast room based on big data according to claim 3 is characterized in that: Extracting contextual semantic association features between the user basic information encoding vector, the user interaction encoding vector, and the user consumption encoding vector to obtain a user information contextual semantic association feature vector, including: The user basic information encoding vector, the user interaction encoding vector and the user consumption encoding vector are passed through a converter-based user information context semantic association feature extractor to obtain the user information context semantic association feature vector.
5. The method for generating crowd portraits in a live broadcast room based on big data according to claim 4 is characterized in that: Fusion of the user information global correlation feature vector and the user information context semantic correlation feature vector to obtain the user information multi-scale fusion correlation feature vector includes: The mutual gain between the user information global association feature vector and the user information context semantic association feature vector is calculated to obtain the user information multi-scale fusion association feature vector.
6. The method for generating crowd portraits in a live broadcast room based on big data according to claim 5 is characterized in that: Based on the multi-scale fusion correlation feature vector of the user information, a crowd portrait is generated, including: The multi-scale fusion associated feature vector of the user information is passed through a crowd portrait generator based on AIGC to generate a crowd portrait.
7. The method for generating crowd portraits in a live broadcast room based on big data according to claim 6 is characterized in that: The method further includes a training step of: training the user information global association feature extractor based on the fully connected layer, the user information contextual semantic association feature extractor based on the converter, and the crowd portrait generator based on AIGC; Wherein, the training step includes: Acquire training data, wherein the training data includes basic training information, training interaction data, and training consumption data of the live broadcast user being analyzed, as well as the true value of the crowd portrait; Performing data cleaning and structured representation on the training basic information, training interaction data and training consumption data of the analyzed live broadcast user to obtain a training user basic information coding vector, a training user interaction coding vector and a training user consumption coding vector; The training user basic information encoding vector, the training user interaction encoding vector and the training user consumption encoding vector are passed through the training user information global correlation feature extractor based on the fully connected layer to obtain a training user information global correlation feature vector; Passing the training user basic information encoding vector, the training user interaction encoding vector and the training user consumption encoding vector through the converter-based user information context semantic association feature extractor to obtain a training user information context semantic association feature vector; Calculating the mutual gain between the training user information global association feature vector and the training user information context semantic association feature vector to obtain the training user information multi-scale fusion association feature vector; Performing feature distribution optimization on the training user information multi-scale fusion associated feature vector to obtain an optimized training user information multi-scale fusion associated feature vector; Passing the optimized training user information multi-scale fusion associated feature vector through the AIGC-based crowd portrait generator to obtain a training generated crowd portrait; Calculating the cross entropy function value between the training generated crowd portrait and the true value of the crowd portrait; and The cross entropy function value is used as the loss function value to train the fully connected layer-based user information global association feature extractor, the converter-based user information contextual semantic association feature extractor, and the AIGC-based crowd portrait generator.
8. The method for generating crowd portraits in a live broadcast room based on big data according to claim 7 is characterized in that: The feature distribution of the training user information multi-scale fusion associated feature vector is optimized to obtain an optimized training user information multi-scale fusion associated feature vector, including: The feature distribution of the training user information multi-scale fusion associated feature vector is optimized using the following optimization formula to obtain the optimized training user information multi-scale fusion associated feature vector; Wherein, the optimization formula is: Where V is the multi-scale fusion associated feature vector of the training user information, v i is the i-th eigenvalue of the multi-scale fusion associated feature vector of the training user information, ||V||2 represents the bi-norm of the multi-scale fusion associated feature vector of the training user information, log is the logarithmic function with base 2, and α is the weight hyperparameter, v' i It is the i-th eigenvalue of the multi-scale fusion associated feature vector of the optimized training user information.
9. A live broadcast room crowd portrait generation system based on big data, characterized in that: include: The data acquisition module is used to obtain the basic information, interaction data and consumption data of the live broadcast user being analyzed; A data cleaning and structured representation module, used for performing data cleaning and structured representation on the basic information, interaction data and consumption data of the analyzed live broadcast user to obtain a user basic information coding vector, a user interaction coding vector and a user consumption coding vector; A semantic analysis module, used for performing semantic analysis on the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain a multi-scale fusion correlation feature vector of user information; as well as The crowd portrait generation module is used to generate a crowd portrait based on the multi-scale fusion associated feature vector of the user information.
10. The live broadcast room crowd portrait generation system based on big data according to claim 9 is characterized in that: The semantic analysis module comprises: A global correlation feature extraction unit, used to extract the user information global correlation feature between the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain a user information global correlation feature vector; A semantic association feature extraction unit, used to extract contextual semantic association features between the user basic information coding vector, the user interaction coding vector and the user consumption coding vector to obtain a user information contextual semantic association feature vector; and A fusion unit is used to fuse the user information global correlation feature vector and the user information context semantic correlation feature vector to obtain the user information multi-scale fused correlation feature vector.
Citation Information
Patent Citations
Live broadcast platform portrait construction method and device, equipment and storage medium
CN113010727A
User portrait generation method and device, equipment and storage medium
CN113611427A
Marketing method and system based on Internet e-commerce live broadcast platform
CN113657975A
Data processing method for directional analysis and prediction of e-commerce platform
CN115879972A
Pedestrian credit information processing method and system based on LSTM (Long Short Term Memory) model
CN116308737A
Cited By
Radar model identification method and system based on intelligent reorganization and semantic coding features
CN117473404A
Multi-modal data-based micro-movie user portrait analysis system and multi-modal data-based micro-movie user portrait analysis method
CN120354262A
