Big data-based network live broadcast e-commerce marketing management system and method

Abstract

The invention relates to the field of marketing management, and particularly discloses a network live broadcast e-commerce marketing management system and method based on big data, and the method comprises the steps: firstly obtaining user historical watching product information data, user comment area interaction data and user basic information of a user exceeding a preset browsing and staying time length, and then employing a deep learning technology to carry out the user historical watching product information data, user comment area interaction data and user basic information; and feature extraction and correlation analysis are carried out on the three parties, and finally, the user portrait of the target user is generated through the generator, so that the target audience is more accurately positioned, and a personalized network live broadcast e-commerce marketing strategy is formulated.

Description

Technical Field

[0001] The present application relates to the field of marketing management, and more specifically, to a big data-based online live streaming e-commerce marketing management system and method. Background Art

[0002] With the rapid development of mobile Internet technology and the continuous upgrading of terminal equipment, online live broadcasting has become one of the core ways of modern information dissemination. Various live broadcasting platforms have sprung up, promoting the diversified development of the live broadcasting industry. Live broadcasting marketing has achieved significant growth in recent years due to its low cost, high efficiency, wide dissemination and superior traffic monetization capabilities. In particular, the video trend of social platforms has further strengthened the zero-distance interactive characteristics of live broadcasting marketing, giving it unique advantages in information dissemination and user interaction.

[0003] Although livestreaming marketing has shown great potential, existing technologies still face many challenges. The most important problem is that the livestreaming audience is large and diverse, making it difficult for brands to effectively locate and reach potential users when conducting precision marketing. This problem limits the accuracy and efficiency of brand marketing, making it difficult to fully leverage the advantages of livestreaming marketing.

[0004] Therefore, a big data-based online live streaming e-commerce marketing management system and method is desired. Summary of the invention

[0005] In order to solve the above technical problems, this application is proposed. The embodiment of this application provides a network live broadcast e-commerce marketing management system and method based on big data, which first obtains the user's historical viewing product information data, user comment area interaction data and user basic information of users who exceed the predetermined browsing stay time, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally generates a user portrait of the target user through a generator, so as to more accurately locate the target audience and formulate a personalized network live broadcast e-commerce marketing strategy.

[0006] According to one aspect of the present application, a big data-based live streaming e-commerce marketing management system is provided, which includes: The online live broadcast marketing data acquisition module is used to obtain the user's historical viewing product information data, user comment area interaction data and user basic information of users who have exceeded the predetermined browsing stay time; A network live broadcast marketing data extraction module, used to extract optimized user behavior information multimodal semantic feature vectors and user basic information semantic understanding feature vectors from the user's historical viewing product information data, the user comment area interaction data and the user's basic information; The target user portrait generation module is used to generate a user portrait of the target user based on the optimized multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information.

[0007] According to another aspect of the present application, a method for managing online live broadcast e-commerce marketing based on big data is provided, which includes: Obtain the user's historical viewing product information data, user comment area interaction data and user basic information of users who have stayed in the browsing area for a predetermined period of time; Extracting an optimized user behavior information multimodal semantic feature vector and a user basic information semantic understanding feature vector from the user's historical product viewing information data, the user comment area interaction data, and the user's basic information; Based on the optimized multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information, a user profile of the target user is generated.

[0008] Compared with the prior art, the present application provides a big data-based online live streaming e-commerce marketing management system and method, which first obtains the user's historical product viewing information data, user comment area interaction data and user basic information of users who exceed the predetermined browsing stay time, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally uses a generator to generate a user portrait of the target user, so as to more accurately locate the target audience and formulate a personalized online live streaming e-commerce marketing strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0009] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0010] Figure 1 It is a block diagram of a big data-based online live streaming e-commerce marketing management system according to an embodiment of the present application.

[0011] Figure 2 It is a block diagram of the online live marketing data extraction module in the online live e-commerce marketing management system based on big data according to an embodiment of the present application.

[0012] Figure 3 It is a block diagram of a user basic information feature extraction unit in a big data-based online live broadcast e-commerce marketing management system according to an embodiment of the present application.

[0013] Figure 4It is a block diagram of a target user portrait generation module in a big data-based online live broadcast e-commerce marketing management system according to an embodiment of the present application.

[0014] Figure 5 This is a flowchart of a big data-based online live streaming e-commerce marketing management method according to an embodiment of the present application. DETAILED DESCRIPTION

[0015] Various exemplary embodiments, features and aspects of the present application 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.

[0016] The word “exemplary” is used exclusively herein to mean “serving as an example, example, or illustration.” Any embodiment described herein as “exemplary” is not necessarily to be construed as preferred or advantageous over other embodiments.

[0017] In addition, in order to better illustrate the present application, numerous specific details are given in the following specific embodiments. It should be understood by those skilled in the art that the present application 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 application.

[0018] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0019] Figure 1 This is a block diagram of a big data-based online live broadcast e-commerce marketing management system according to an embodiment of the present application. Figure 1 As shown, according to the big data-based online live streaming e-commerce marketing management system 100 of the embodiment of the present application, it includes: an online live streaming marketing data acquisition module 110, which is used to obtain the user's historical viewing product information data, user comment area interaction data and user basic information of users who exceed the predetermined browsing residence time; an online live streaming marketing data extraction module 120, which is used to extract optimized user behavior information multimodal semantic feature vectors and user basic information semantic understanding feature vectors from the user's historical viewing product information data, the user comment area interaction data and the user basic information; a target user portrait generation module 130, which is used to generate a user portrait of the target user based on the optimized user behavior information multimodal semantic feature vector and the user basic information semantic understanding feature vector.

[0020] In the above-mentioned live broadcast e-commerce marketing management system 100 based on big data, the live broadcast marketing data acquisition module 110 is used to obtain the user's historical viewing product information data, user comment area interaction data and user basic information of users who exceed the predetermined browsing stay time. It should be understood that with the rapid development of mobile Internet technology and the continuous upgrading of terminal equipment, live broadcasting has become one of the key means of modern information dissemination. Various live broadcast platforms have sprung up like mushrooms after rain, promoting the diversified progress of the live broadcast industry. Live broadcast marketing has achieved significant growth in recent years due to its low cost, high efficiency, wide dissemination and superior traffic monetization capabilities. In particular, the video trend of social platforms has further strengthened the real-time interactive characteristics of live broadcast marketing, making it show unique advantages in information dissemination and user interaction. However, despite the huge potential of live broadcast marketing, current technology still faces multiple challenges. The most important problem is that the audience group of live broadcast is large and diverse, and it is difficult for brands to effectively locate and reach potential users when conducting precision marketing. This problem limits the accuracy and efficiency of brand marketing, making it difficult to give full play to the potential advantages of live broadcast marketing. Therefore, in the technical solution of the present application, the user's historical product viewing information data, user comment area interaction data and user basic information of users who exceed the predetermined browsing stay time are obtained, and combined with deep learning technology, user portraits of target users are generated to accurately lock in the target audience and formulate personalized online live e-commerce marketing strategies.

[0021] Specifically, obtaining the historical product information data, user comment area interaction data and user basic information of users who have stayed for more than the predetermined browsing time is to deeply understand the user's interest preferences, behavior patterns and personal characteristics, so as to accurately implement the online live e-commerce marketing strategy. By analyzing the user's historical viewing data, we can understand the product categories, preference trends and purchasing tendencies that they are concerned about, and provide a basis for the recommendation system and personalized promotion. At the same time, collecting the user's interactive data in the comment area can capture their real feedback and social interaction patterns on the product, and provide real-time reference for improving the quality of products and services. In addition, obtaining basic information about users such as age, gender, and geographic location will help build a more comprehensive user portrait and further optimize marketing goals and strategy design. In summary, meticulous data collection and analysis can not only help improve user experience and satisfaction, but also improve marketing effectiveness and brand influence, and provide solid data support and strategic foundation for the sustainable development of online live e-commerce.

[0022] In the above-mentioned live streaming e-commerce marketing management system 100 based on big data, the live streaming marketing data extraction module 120 is used to extract optimized multimodal semantic feature vectors of user behavior information and semantic understanding feature vectors of user basic information from the user's historical product information data, the user comment area interaction data and the user's basic information. In this way, not only the understanding and prediction capabilities of user behavior are improved, but also data-driven decision-making and precise marketing operations are supported. This way of comprehensive analysis and use of data not only promotes interaction and trust between enterprises and users, but also lays a solid foundation for the long-term development and competitive advantage of e-commerce platforms.

[0023] Figure 2 FIG. 1 is a block diagram of a network live broadcast marketing data extraction module in a network live broadcast e-commerce marketing management system based on big data according to an embodiment of the present application. Figure 2 As shown, in a specific embodiment of the present application, the online live marketing data extraction module 120 includes: a historical viewing information feature extraction unit 121, which is used to perform feature extraction on the user's historical viewing product information data to obtain a semantic feature vector associated with the user's historical viewing product information; a comment area interaction feature extraction unit 122, which is used to perform feature extraction on the user comment area interaction data to obtain a user comment area interaction multi-scale text understanding feature vector; a user behavior information multimodal construction unit 123, which is used to construct the user's historical viewing product information associated semantic feature vector and the user comment area interaction multi-scale text understanding feature vector to obtain the optimized user behavior information multimodal semantic feature vector; a user basic information feature extraction unit 124, which is used to perform feature extraction on the user basic information to obtain the user basic information semantic understanding feature vector.

[0024] It should be understood that by analyzing the user's historical viewing records, the user's favorite product categories, brand preferences and purchasing trends can be identified, thereby providing an accurate basis for the personalized recommendation system. Among them, the feature extraction process covers multi-dimensional analysis from user viewing time, frequency to viewing content. Through machine learning and deep learning technology, this information is converted into semantic feature vectors that can be quantified and understood. These feature vectors can not only help e-commerce platforms accurately predict users' purchase intentions and behavioral conversion paths, but also optimize product recommendations, personalized marketing activities and advertising effects.

[0025] Furthermore, feature extraction of interactive data in user comment areas is performed in order to fully understand users’ attitudes, emotions, and interactive behaviors toward products. As a direct communication channel between users and products, interaction in comment areas records users’ actual experience and emotional feedback on products. Through feature extraction, users’ emotional tendencies, opinions, and product-specific evaluation details can be captured from comment texts, which are crucial for product improvement and market feedback. In the feature extraction process, not only the emotional polarity and emotional intensity of comments are considered, but also the themes and keywords of comments can be analyzed to identify users’ preference patterns and potential needs. By continuously improving and optimizing the feature extraction technology of interactive data in comment areas, e-commerce platforms can achieve more accurate user emotion management and marketing strategies, thereby enhancing the interaction and trust between brands and users.

[0026] Furthermore, the semantic feature vector of the user's historical product viewing information captures the user's preference for different products and their relevance, while the multi-scale text understanding feature vector of the user's comment area interaction provides the emotions, opinions, and interaction patterns expressed by the user in the comment. By constructing these two vectors, the accuracy of the recommendation system can be improved to better meet the personalized needs of users. This optimization not only improves the understanding of user behavior, but also improves the accuracy of product recommendations, thereby improving the user experience.

[0027] In particular, user basic information includes but is not limited to key data such as age, gender, geographic location, educational background, and occupation. This information can help e-commerce platforms build user portraits and identify user group characteristics and market segments. Through feature extraction technology, this basic information can be converted into structured semantic understanding feature vectors, making them more operational and analytically valuable. These feature vectors can not only help companies understand users' consumption habits and purchasing power, but also support the optimization of personalized recommendation systems and the formulation of customized marketing strategies. In the feature extraction process, technical means such as data mining, machine learning, and statistical analysis are used to ensure that the feature vectors extracted from user basic information have a high degree of accuracy and predictive power. By continuously updating and optimizing the feature extraction method of user basic information, e-commerce platforms can continuously optimize user experience management, improve service response speed and the accuracy of personalized recommendations, thereby enhancing user satisfaction and loyalty.

[0028] In a specific embodiment of the present application, the historical viewing information feature extraction unit 121 includes: passing the user's historical viewing product information data through a user's historical viewing product information bidirectional long short-term memory neural network to obtain a plurality of user's historical viewing product information semantic feature vectors; arranging the plurality of user's historical viewing product information semantic feature vectors in two dimensions into a user's historical viewing product information two-dimensional semantic feature matrix; passing the user's historical viewing product information two-dimensional semantic feature matrix through a user's historical viewing product information two-dimensional semantic feature filter to obtain the user's historical viewing product information associated semantic feature vector.

[0029] It should be understood that the user's historical product viewing information data is processed through the user's historical product viewing information bidirectional long short-term memory neural network (BiLSTM) to comprehensively capture the user's viewing behavior characteristics from multiple time scales and semantic levels, thereby generating multiple user historical product viewing information semantic feature vectors. Although the traditional unidirectional LSTM network can capture the dependencies on the time series, it cannot effectively handle long-distance dependencies and capture global information. BiLSTM can more accurately model the dependencies and semantic features in time series data by considering both past and future information. Through BiLSTM, the user's historical product viewing data can be effectively processed, not only considering the viewing behavior at each time point, but also encoding these viewing behaviors on the time series, thereby forming semantic feature vectors of multiple time steps. These feature vectors not only contain basic information such as product category and viewing frequency, but also can capture the user's interest evolution and behavior pattern changes through deep representation learning. In practical applications, by inputting the user's historical product viewing information data into the BiLSTM network, the user's behavior pattern can be deeply analyzed and understood, providing a data-driven decision-making basis for e-commerce platforms. By continuously optimizing and adjusting the network structure, the efficiency and accuracy of feature extraction can be improved, and the understanding of the motivations and trends behind user behavior can be further enhanced.

[0030] Furthermore, each row of the two-dimensional semantic feature matrix of user historical product viewing information represents a user, and each column represents a specific time point or dimension of viewing features. This arrangement can intuitively display the viewing behavior of different users in different time periods and their semantic feature expressions. This structured representation not only makes data easier to manage and analyze, but also enables deeper user behavior pattern mining and analysis through matrix operations and data mining techniques.

[0031] Furthermore, the two-dimensional semantic feature matrix of the user's historical viewing product information is obtained by structuring the historical behavior data of multiple users into a matrix form, and each element represents the semantic information at a specific time point or feature dimension. However, directly using these raw data may result in information overload and unnecessary complexity. Therefore, a two-dimensional semantic feature filter of the user's historical viewing product information is introduced, and its main task is to screen and extract the most representative and predictive features through a series of algorithms and models to construct the associated semantic feature vector of the user's behavior. The design of the filter usually involves technologies such as feature selection, data dimensionality reduction, and pattern recognition, through which the potential patterns and association laws behind the user's behavior can be effectively captured. Specifically, each layer of the two-dimensional semantic feature filter of the user's historical viewing product information is used to perform convolution processing, mean pooling processing based on the local feature matrix, and nonlinear activation processing on the input data in the forward pass of the layer to output the user's historical viewing product information associated semantic feature vector by the last layer of the two-dimensional semantic feature filter of the user's historical viewing product information, wherein the input of the two-dimensional semantic feature filter of the user's historical viewing product information is the two-dimensional semantic feature matrix of the user's historical viewing product information.

[0032] In a specific embodiment of the present application, the comment area interaction feature extraction unit 122 includes: passing the user comment area interaction data through a user comment area interaction context semantic encoder including an embedding layer to obtain a plurality of user comment area interaction semantic feature vectors; arranging the plurality of user comment area interaction semantic feature vectors in one dimension into a user comment area interaction one-dimensional feature vector; passing the user comment area interaction one-dimensional feature vector through a user comment area interaction multi-scale neighborhood feature encoding module to obtain the user comment area interaction multi-scale text understanding feature vector.

[0033] It should be understood that the interactive data in the user comment area includes user comments, replies, likes and other behaviors, which usually have complex contexts and multi-level semantic information. It may be difficult to extract valuable information by directly analyzing these raw data, so they need to be processed through embedding layers and contextual semantic encoders. The main function of the embedding layer is to convert the discrete words and symbols in the interactive data in the user comment area into dense vector representations, which can effectively capture the semantic relationship and contextual information between words. For example, through word embedding technologies such as Word2Vec or GloVe, the words in the comments are converted into low-dimensional vectors, so that words with similar meanings are closer in the vector space. The interactive context semantic encoder in the user comment area further uses the feature vectors generated by these embedding layers to understand the contextual relationship and semantic structure of the comment content through deep learning models (such as LSTM, Transformer, etc.). These models can take into account the contextual information of the comments, such as the order of comments, the coherence of replies, the emotional tendency of users, etc., to generate feature vectors with rich semantic information. These semantic feature vectors not only contain the content information of the comments, but also reflect the emotional state and interaction intention of the commenter. Through this processing method, the multiple user comment area interaction semantic feature vectors obtained can reflect the user's real opinions, emotional fluctuations and interaction patterns in the comment area, thereby providing more accurate input for the content recommendation system. These vectors can help the system identify the user's interests, predict the user's needs, and thus achieve more personalized recommendations and interactions. Specifically, the user comment area interaction data is segmented to obtain a comment interaction word sequence; the embedding layer of the user comment area interaction context semantic encoder containing the embedding layer is used to map each comment interaction word in the comment interaction word sequence to a word embedding vector to obtain a sequence of comment interaction word embedding vectors; the converter-based Bert model of the user comment area interaction context semantic encoder containing the embedding layer is used to perform global context semantic encoding on the sequence of comment interaction word embedding vectors to obtain multiple user comment area interaction semantic feature vectors.

[0034] Furthermore, multiple user comment area interaction semantic feature vectors are usually extracted through deep learning models or other text processing technologies, and each vector reflects a series of user behaviors and emotional characteristics in the comment area. These feature vectors may cover a variety of information such as sentiment polarity, keyword frequency, interaction frequency, etc., but their dimensions are high and their structures are complex. By arranging these vectors into one-dimensional feature vectors of user comment area interactions, each user's comment interaction behavior can be concisely represented as a vector sequence, with each element corresponding to a specific feature or dimension. The form of one-dimensional feature vectors of user comment area interactions makes data easier to store, process, and analyze.

[0035] Furthermore, the one-dimensional feature vector of the user comment area interaction is processed by the multi-scale neighborhood feature encoding module of the user comment area interaction to obtain the multi-scale text understanding feature vector of the user comment area interaction, in order to more comprehensively and deeply understand the interactive behavior and speech content of the user in the comment area. This process combines multi-scale feature extraction and text understanding technology, which can capture the rich semantic information and behavior patterns in the interaction of the comment area, thereby improving the accurate understanding of user intentions and emotions. Among them, the one-dimensional feature vector of the user comment area interaction usually contains the basic characteristics of the user in the comment, reply, like and other behaviors, such as emotional tendency, keyword frequency, etc. However, a single feature vector may not fully reflect the complexity and diversity of user interaction. Therefore, the introduction of the multi-scale neighborhood feature encoding module aims to analyze and understand the interaction of the user comment area more carefully through the contextual information and contextual relationships of different scales. The multi-scale neighborhood feature encoding module can include a variety of technologies and models, such as convolutional neural network (CNN), recurrent neural network (RNN), self-attention mechanism (Transformer), etc. These modules can process comment texts at different scales, and gradually understand and encode text features from local to global. For example, phrases and local relationships can be captured through convolution operations, RNN can handle long-term dependencies and sequence information, and the self-attention mechanism can effectively capture global semantic associations. By integrating multi-scale feature encoding, the obtained multi-scale text understanding feature vector of user comment area interaction has higher information richness and expression ability. These feature vectors can not only more accurately express the semantic meaning and emotional trend of the comments, but also identify the potential patterns and user behavior trends hidden behind complex interactions. This is of great significance for improving the accuracy of content recommendations, optimizing user interaction experience, and improving brand management strategies. Specifically, the user comment area interactive multi-scale neighborhood feature encoding module includes: a first convolutional layer, a second convolutional layer parallel to the first convolutional layer, and a cascade layer connected to the first convolutional layer and the second convolutional layer, wherein the first convolutional layer uses a one-dimensional convolutional kernel with a first scale, and the second convolutional layer uses a one-dimensional convolutional kernel with a second scale. More specifically, the first convolution layer of the user comment area interaction multi-scale neighborhood feature encoding module is used to perform one-dimensional convolution encoding on the one-dimensional feature vector of the user comment area interaction to obtain a first-scale feature vector; the second convolution layer of the user comment area interaction multi-scale neighborhood feature encoding module is used to perform one-dimensional convolution encoding on the one-dimensional feature vector of the user comment area interaction to obtain a second-scale feature vector; the first-scale feature vector and the second-scale feature vector are cascaded to obtain the multi-scale text understanding feature vector of the user comment area interaction.

[0036] In a specific embodiment of the present application, the user behavior information multimodal construction unit 123 includes: associating the semantic feature vector associated with the user's historical product viewing information and the user comment area interactive multi-scale text understanding feature vector to obtain a user behavior information multimodal semantic feature vector; based on the user basic information semantic understanding feature vector, performing linear embedding modulation compensation on the user behavior information multimodal semantic feature vector based on fine-grained difference features to obtain an optimized user behavior information multimodal semantic feature vector.

[0037] It should be understandable that the semantic feature vector of the user's historical product viewing information reflects the user's preferences and purchase intentions for different categories of products. At the same time, the interactive multi-scale text understanding feature vector of the user comment area includes the user's emotional attitude towards the product, specific usage experience and suggestions. Through sentiment analysis and text mining technology, we can deeply understand the user's attitude and feedback content. The process of associating these two types of feature vectors can not only enrich and supplement each other's information semantically, but also help reveal the deep motivations and decision-making factors behind user behavior.

[0038] In particular, in the technical solution of the present application, the multimodal semantic feature vector of user behavior information is constructed by integrating the user's historical viewing product information and comment area interaction data. These data have dynamic and temporal properties, and are processed by a bidirectional long short-term memory neural network and a multi-scale neighborhood feature encoding module, capturing the complex patterns and dynamic changes of user behavior, focusing on the user's behavior patterns and interaction details, and having rich temporal and contextual information in a high-dimensional space. In contrast, the semantic understanding feature vector of user basic information is processed by data cleaning, word embedding model, and text semantic encoder based on a converter. Among them, user basic information such as age, gender, and geographic location are static and structured. These features mainly focus on the user's basic identity information, and focus on extracting information at the text semantic and lexical levels during processing. The dimensions are usually low, and they mainly reflect the basic attributes of the user, rather than their behavioral dynamics or interaction patterns. This difference in feature processing leads to inconsistency in the fine-grainedness of the multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information in the feature space. That is, the feature distribution of user behavior information has temporal and complex contextual relationships, while the feature distribution of user basic information is more structured and static. When these two features are fused, due to the difference in distribution on the feature manifold, the fused user portrait generation feature vector may appear abnormally stacked on the boundary of the feature manifold. Therefore, in the technical solution of the present application, based on the user basic information semantic understanding feature vector, the user behavior information multimodal semantic feature vector is linearly embedded and modulated based on fine-grained difference features to obtain an optimized user behavior information multimodal semantic feature vector.

[0039] Among them, based on the user basic information semantic understanding feature vector, the user behavior information multimodal semantic feature vector is linearly embedded and modulated based on fine-grained difference features to obtain an optimized user behavior information multimodal semantic feature vector, including: calculating the positional difference between the user behavior information multimodal semantic feature vector and the user basic information semantic understanding feature vector to obtain an e-commerce marketing information shifted positional difference representation vector; after cascading the user behavior information multimodal semantic feature vector and the user basic information semantic understanding feature vector into an e-commerce marketing information joint feature vector, the e-commerce marketing information joint feature vector is input into a one-dimensional convolutional layer to obtain an e-commerce marketing information shifted convolutional coding representation vector; after using the first e-commerce marketing information linear embedding weight matrix to perform e-commerce marketing information linear embedding modulation coding on the user behavior information multimodal semantic feature vector to obtain a user behavior information multimodal semantic feature vector linear modulation vector, Calculate the positional difference between the linear modulation vector of the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information to obtain the shifted modulation differential representation vector of the e-commerce marketing information; use the shifted positional differential representation vector of the e-commerce marketing information as the value vector, the shifted convolutional coding representation vector of the e-commerce marketing information and the shifted modulation differential representation vector of the e-commerce marketing information as the query vector and the key vector, input the shifted positional differential representation vector of the e-commerce marketing information, the shifted convolutional coding representation vector of the e-commerce marketing information and the shifted modulation differential representation vector of the e-commerce marketing information into a shift corrector based on an emulated transformer structure to obtain the multi-scale hierarchical fusion shifted modulation weight vector of the e-commerce marketing information; calculate the positional point multiplication between the multi-scale hierarchical fusion shifted modulation weight vector of the e-commerce marketing information and the multimodal semantic feature vector of the user behavior information to obtain the optimized multi-modal semantic feature vector of the user behavior information.

[0040] The following optimization formula is used to express that the linear embedding modulation compensation based on the fine-grained difference feature of the user behavior information multimodal semantic feature vector is performed based on the user basic information semantic understanding feature vector to obtain an optimized user behavior information multimodal semantic feature vector; Wherein, the optimization formula is: ; ; ; ; ;in, Indicates the shift of e-commerce marketing information by position difference to represent the vector, Represents the multimodal semantic feature vector of user behavior information, Represents the semantic understanding feature vector of user basic information, Represents the shift convolutional coding representation vector of e-commerce marketing information, represents the shift modulation difference representation vector of e-commerce marketing information, represents the linear embedding weight matrix of the first e-commerce marketing information, represents the e-commerce marketing information modulation bias vector, represents the multi-scale hierarchical fusion shift modulation weight vector of e-commerce marketing information, Represents the transposed vector of the shift-modulated differential representation vector of e-commerce marketing information, The length of the shift convolutional coding representation vector representing e-commerce marketing information represents the difference by position, represents a one-dimensional convolutional layer, represents a cascade of the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information, represents the linear rectification function, represents matrix multiplication, represents the normalized exponential function, It means point multiplication by position. Represents the optimized multimodal semantic feature vector of user behavior information.

[0041] In the technical solution of the present application, in the process of fusing the multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information, the fine-grained inconsistency of feature distribution between the multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information is taken into consideration, which will cause abnormal stacking of the fused user portrait generation feature vector on the boundary of the feature manifold. Based on this, in the technical solution of the present application, before fusing the multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information to obtain the user portrait generation feature vector, linear embedding modulation compensation based on fine-grained difference features is performed on the multimodal semantic feature vector of user behavior information based on the semantic understanding feature vector of user basic information.

[0042] In the technical solution of the present application, a linear embedding modulation compensation method based on fine-grained difference features is proposed to address the problem of fine-grained inconsistency in feature distribution that arises in the process of fusing multimodal semantic feature vectors of user behavior information and semantic understanding feature vectors of user basic information. The method aims to reduce the abnormal stacking phenomenon of feature vectors generated by user portraits on the boundary of feature manifolds, so as to improve the robustness and classification performance of the model.

[0043] Specifically, the method first calculates the element-by-element difference between the multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information to obtain the shifted element-by-element differential representation vector of e-commerce marketing information, thereby capturing the difference between the two feature vectors. Then, the multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information are cascaded into an e-commerce marketing information joint feature vector and input into a one-dimensional convolutional layer to generate a shifted convolutional coding representation vector of e-commerce marketing information. That is, the deep-level shifted implicit information between the multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information is captured by one-dimensional convolutional coding. At the same time, the multimodal semantic feature vector of user behavior information is linearly embedded and modulated by the first e-commerce marketing information linear embedding weight matrix to obtain a linear modulation vector of the multimodal semantic feature vector of user behavior information, and the element-by-element difference between the linear modulation vector of the multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information is calculated to obtain a shifted modulation differential representation vector of e-commerce marketing information. The e-commerce marketing information shift modulation differential representation vector is used to represent the positional expression of the shift information of the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information in the linear space. That is, in the technical solution of the present application, the positional difference of the shift of the e-commerce marketing information, the shift convolutional coding representation vector of the e-commerce marketing information and the shift modulation differential representation vector of the e-commerce marketing information respectively represent the explicit expression of the fine-grained distribution difference shift information between the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information in different feature spaces.

[0044] Next, the shifted element-by-element differential representation vector of e-commerce marketing information is used as the value vector, the shifted convolutional coding representation vector of e-commerce marketing information and the shifted modulation differential representation vector of e-commerce marketing information are used as the query vector and the key vector respectively, and these vectors are input into the shift corrector based on the simulated transformer structure to generate the multi-scale hierarchical fusion shift modulation weight vector of e-commerce marketing information. In this way, the importance of features at different scales is evaluated by the attention mechanism, that is, the attention mechanism is used to simulate the hierarchical precipitation expression of the fine-grained difference feature distribution between the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information in different feature spaces. Finally, the element-by-element product of the multi-scale hierarchical fusion shift modulation weight vector of e-commerce marketing information and the multimodal semantic feature vector of user behavior information is calculated to obtain the optimized multimodal semantic feature vector of user behavior information, that is, the weighted operation is used to assign weights to each element in the multimodal semantic feature vector of user behavior information, which reduces the influence of local disturbances and improves the quality of the feature vector.

[0045] In summary, in the method, the position difference of the shift of the e-commerce marketing information, the shift convolutional coding representation vector of the e-commerce marketing information, and the shift modulation difference representation vector of the e-commerce marketing information respectively represent the explicit expression of the fine-grained distribution difference shift information between the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information in different feature spaces. The generation of the multi-scale hierarchical fusion shift modulation weight vector of the e-commerce marketing information uses the attention mechanism to evaluate the importance of features at different scales and provide weight information for feature shift correction. Ultimately, the generation of the optimized multimodal semantic feature vector of user behavior information reduces the impact of disturbances, improves the robustness of the model to local disturbances in the input data, enhances the model's adaptability to noisy or variant data sets, improves the fusion quality of the feature vector, and improves the overall classification performance and training efficiency of the model.

[0046] Figure 3 Schematic diagram of a block diagram of a user basic information feature extraction unit in a live broadcast e-commerce marketing management system based on big data according to an embodiment of the present application. Figure 3 As shown, in a specific embodiment of the present application, the user basic information feature extraction unit 124 includes: a user basic information preprocessing subunit 1241, used to perform data preprocessing on the user basic information to obtain a plurality of user basic information data item word embedding vectors; a user basic information convolutional encoding subunit 1242, used to pass the plurality of user basic information data item word embedding vectors through a converter-based user basic information text semantic encoder to obtain the user basic information semantic understanding feature vector.

[0047] It should be understandable that data preprocessing and word embedding conversion of basic user information can significantly improve data quality and model performance, and provide a solid data foundation for intelligent decision-making of enterprises and platforms.

[0048] Furthermore, the word embedding vectors of multiple user basic information data items generally provide a static numerical representation of each data item, such as gender, age, geographic location, etc. However, these vectors may be difficult to fully reflect the multi-dimensional characteristics and dynamic changes of users when they exist alone. Transformer-based text semantic encoders, such as BERT, GPT and other models, can extract deeper semantic information and contextual relationships from text data through pre-training and fine-tuning. Specifically, the converter-based Bert model of the converter-based user basic information text semantic encoder is used to perform global contextual semantic encoding on the word embedding vectors of the multiple user basic information data items to obtain multiple basic information feature vectors; and, the multiple basic information feature vectors are cascaded to obtain the user basic information semantic understanding feature vector.

[0049] In a specific embodiment of the present application, the user basic information preprocessing subunit 1241 includes: performing data cleaning on the user basic information to obtain multiple user basic information data items; and passing the multiple user basic information data items through a user basic information word embedding model to obtain word embedding vectors of the multiple user basic information data items.

[0050] It should be understood that by performing data cleaning, it is possible to ensure that the quality of multiple basic user information data items meets credible standards, providing reliable data support for data-driven application scenarios such as personalized recommendations, precision marketing, user behavior analysis, etc. This processing not only improves the quality and availability of data, but also provides a solid foundation for enterprises and institutions to gain advantages in a highly competitive market environment.

[0051] Furthermore, multiple user basic information data items are processed through the user basic information word embedding model to obtain the user basic information data item word embedding vector, in order to convert the original user data into numerical features that can be efficiently used by the machine learning model. This process involves converting discrete, usually categorized basic information, such as gender, age, geographic location, etc., into dense vector representations, thereby achieving more efficient and accurate data processing.

[0052] In the above-mentioned live streaming e-commerce marketing management system 100 based on big data, the target user portrait generation module 130 is used to generate a user portrait of the target user based on the optimized multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information. It should be understood that by combining these two feature vectors, a user portrait of the target user can be generated. This user portrait not only includes the basic attributes of the user, such as age and gender, but also integrates the user's behavioral habits, interest preferences and interaction patterns. The generated user portrait can fully display the personalized characteristics of the user, and provides strong support for applications such as personalized recommendations, precision marketing, and customer relationship management.

[0053] Figure 4 Schematic diagram of a block diagram of a target user portrait generation module in a live streaming e-commerce marketing management system based on big data according to an embodiment of the present application. Figure 4 As shown, in a specific embodiment of the present application, the target user portrait generation module 130 includes: a user information feature fusion unit 131, used to fuse the optimized user behavior information multimodal semantic feature vector and the user basic information semantic understanding feature vector to obtain a user portrait generation feature vector; a user portrait generation unit 132, used to pass the user portrait generation feature vector through a generator to generate a user portrait of the target user.

[0054] It should be understood that behavioral characteristics can reveal the actual needs and preferences of users, while basic information provides the background and context for these behaviors. By integrating these two types of feature vectors, enterprises can build more accurate user portraits, achieve more personalized services and optimize decision support.

[0055] Furthermore, the user portrait generation feature vector is a comprehensive feature description obtained by fusing the semantic feature vector of user behavior information and basic information. These feature vectors contain information on multiple aspects such as user behavior habits, preferences, social attributes, etc., but there may still be incomplete information or more refined expressions are needed. As a generative model, the generator can generate new user feature vectors with similar distribution and characteristics by learning and analyzing a large amount of real data. The working principle of the generator is usually based on deep learning technology, such as generative adversarial networks (GAN) or variational autoencoders (VAE). These models can learn the high-level features and potential structures of the data, so that they can generate outputs that conform to the distribution of real data when a new feature vector is input. For the user portrait generation task, the generator can generate a new user portrait that matches the characteristics of the target user based on the input user feature vector. The user portrait generated by the generator can not only make up for the missing or insufficient information in the original data, but also provide more detailed and in-depth user insights.

[0056] In summary, the embodiment of the present application first obtains the user's historical product viewing information data, user comment area interaction data and user basic information of users who exceed the predetermined browsing stay time, and then uses deep learning technology to perform feature extraction and correlation analysis on the three, and finally uses a generator to generate a user portrait of the target user, so as to more accurately locate the target audience and formulate a personalized online live e-commerce marketing strategy.

[0057] As described above, the network live broadcast e-commerce marketing management system 100 based on big data according to the embodiment of the present application can be implemented in various terminal devices. In one example, the network live broadcast e-commerce marketing management system 100 based on big data can be integrated into the terminal device as a software module and / or hardware module. For example, the network live broadcast e-commerce marketing management system 100 based on big data can be a software module in the operating system of the terminal device, or it can be an application developed for the terminal device; of course, the network live broadcast e-commerce marketing management system 100 based on big data can also be one of the many hardware modules of the terminal device.

[0058] Alternatively, in another example, the big data-based live streaming e-commerce marketing management system 100 and the terminal device may also be separate devices, and the big data-based live streaming e-commerce marketing management system 100 may be connected to the terminal device via a wired and / or wireless network, and transmit interactive information in accordance with an agreed data format.

[0059] Figure 5 Flow chart of the network live broadcast e-commerce marketing management method based on big data according to an embodiment of the present application. Figure 5 As shown, according to the big data-based online live streaming e-commerce marketing management method of the embodiment of the present application, it includes: S110, obtaining the user's historical viewing product information data, user comment area interaction data and user basic information of the user who exceeds the predetermined browsing stay time; S120, extracting the optimized user behavior information multimodal semantic feature vector and the user basic information semantic understanding feature vector from the user's historical viewing product information data, the user comment area interaction data and the user basic information; S130, generating a user portrait of the target user based on the optimized user behavior information multimodal semantic feature vector and the user basic information semantic understanding feature vector.

[0060] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned network live broadcast e-commerce marketing management method based on big data have been referenced above. Figures 1 to 4 The description of the big data-based live streaming e-commerce marketing management system has been introduced in detail, and therefore, its repeated description will be omitted.

[0061] In the several embodiments provided by the present invention, it should be understood that the disclosed devices, apparatuses and methods can be implemented in other ways. For example, the device embodiments described above are only schematic, for example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.

[0062] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0063] In addition, each functional module in each embodiment of the present invention may be integrated into one processing unit, each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional modules.

[0064] It is obvious to those skilled in the art that the present invention is not limited to the details of the above exemplary embodiments, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0065] Therefore, no matter from which point of view, the embodiments should be regarded as illustrative and non-restrictive, and the scope of the present invention is limited by the appended claims rather than the above description, so it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention. Any attached figure mark in the claims should not be regarded as limiting the claims involved.

[0066] In addition, it is clear that the word "comprising" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claim can also be implemented by one unit or device through software or hardware. The second and other words are used to indicate names, but not to indicate any particular order.

[0067] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solution of the present invention can be modified or replaced by equivalents without departing from the spirit of the technical solution of the present invention.

Claims

1. A live streaming e-commerce marketing management system based on big data, characterized in that: include: The online live broadcast marketing data acquisition module is used to obtain the user's historical viewing product information data, user comment area interaction data and user basic information of users who have exceeded the predetermined browsing stay time; A network live broadcast marketing data extraction module, used to extract optimized user behavior information multimodal semantic feature vectors and user basic information semantic understanding feature vectors from the user's historical viewing product information data, the user comment area interaction data and the user's basic information; The target user portrait generation module is used to generate a user portrait of the target user based on the optimized multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information.

2. According to the big data-based online live broadcast e-commerce marketing management system of claim 1, it is characterized in that: The network live broadcast marketing data extraction module includes: A historical viewing information feature extraction unit, used to extract features from the user's historical viewing product information data to obtain a semantic feature vector associated with the user's historical viewing product information; A comment area interaction feature extraction unit, used to extract features from the user comment area interaction data to obtain a user comment area interaction multi-scale text understanding feature vector; A user behavior information multimodal construction unit, used to construct the semantic feature vector associated with the user's historical product viewing information and the multi-scale text understanding feature vector of the user comment area interaction to obtain the optimized user behavior information multimodal semantic feature vector; The user basic information feature extraction unit is used to extract features from the user basic information to obtain a semantic understanding feature vector of the user basic information.

3. The big data-based live streaming e-commerce marketing management system according to claim 2 is characterized in that: The historical viewing information feature extraction unit includes: Passing the user's historical product viewing information data through a user's historical product viewing information bidirectional long short-term memory neural network to obtain multiple user's historical product viewing information semantic feature vectors; Arranging the plurality of semantic feature vectors of the historically viewed product information of the user in two dimensions into a two-dimensional semantic feature matrix of the historically viewed product information of the user; The two-dimensional semantic feature matrix of the user's historically viewed product information is passed through a two-dimensional semantic feature filter of the user's historically viewed product information to obtain a semantic feature vector associated with the user's historically viewed product information.

4. The big data-based live streaming e-commerce marketing management system according to claim 3 is characterized in that: The comment area interactive feature extraction unit includes: Passing the user comment area interaction data through a user comment area interaction context semantic encoder including an embedding layer to obtain a plurality of user comment area interaction semantic feature vectors; Arranging the plurality of user comment area interaction semantic feature vectors in one dimension into a user comment area interaction one-dimensional feature vector; The one-dimensional feature vector of the user comment area interaction is passed through a multi-scale neighborhood feature encoding module of the user comment area interaction to obtain the multi-scale text understanding feature vector of the user comment area interaction.

5. The big data-based live streaming e-commerce marketing management system according to claim 4 is characterized in that: The user behavior information multimodal construction unit includes: Associating the semantic feature vector associated with the product information viewed by the user in history and the multi-scale text understanding feature vector of the user comment area interaction to obtain a multi-modal semantic feature vector of the user behavior information; Based on the semantic understanding feature vector of the user basic information, the linear embedding modulation compensation based on the fine-grained difference feature is performed on the multimodal semantic feature vector of the user behavior information to obtain an optimized multimodal semantic feature vector of the user behavior information.

6. The big data-based live streaming e-commerce marketing management system according to claim 5 is characterized in that: Based on the semantic understanding feature vector of the user basic information, the linear embedding modulation compensation based on the fine-grained difference feature is performed on the multimodal semantic feature vector of the user behavior information to obtain an optimized multimodal semantic feature vector of the user behavior information, including: Calculating the position difference between the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information to obtain the position difference representation vector of the e-commerce marketing information shift; After concatenating the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information into a joint feature vector of e-commerce marketing information, the joint feature vector of e-commerce marketing information is input into a one-dimensional convolutional layer to obtain a shifted convolutional coding representation vector of the e-commerce marketing information; After performing e-commerce marketing information linear embedding modulation encoding on the multimodal semantic feature vector of the user behavior information using the first e-commerce marketing information linear embedding weight matrix to obtain a linear modulation vector of the multimodal semantic feature vector of the user behavior information, a positional difference between the linear modulation vector of the multimodal semantic feature vector of the user behavior information and the semantic understanding feature vector of the user basic information is calculated to obtain an e-commerce marketing information shift modulation difference representation vector; Taking the shifted position differential representation vector of the e-commerce marketing information as the value vector, the shifted convolutional coding representation vector of the e-commerce marketing information and the shift modulated differential representation vector of the e-commerce marketing information as the query vector and the key vector, the shifted position differential representation vector of the e-commerce marketing information, the shifted convolutional coding representation vector of the e-commerce marketing information and the shift modulated differential representation vector of the e-commerce marketing information are input into a shift corrector based on a pseudo-converter structure to obtain a multi-scale hierarchical fusion shift modulation weight vector of the e-commerce marketing information; The optimized user behavior information multimodal semantic feature vector is obtained by calculating the point-by-point multiplication of the multi-scale hierarchical fusion shift modulation weight vector of the e-commerce marketing information and the multimodal semantic feature vector of the user behavior information.

7. The big data-based live streaming e-commerce marketing management system according to claim 6 is characterized in that: The user basic information feature extraction unit includes: A user basic information preprocessing subunit, configured to perform data preprocessing on the user basic information to obtain word embedding vectors for multiple user basic information data items; The user basic information convolutional encoding subunit is used to embed the multiple user basic information data item words into vectors through a converter-based user basic information text semantic encoder to obtain the user basic information semantic understanding feature vector.

8. The big data-based live streaming e-commerce marketing management system according to claim 7 is characterized in that: The user basic information preprocessing subunit includes: Performing data cleaning on the user basic information to obtain multiple user basic information data items; The multiple user basic information data items are passed through a user basic information word embedding model to obtain word embedding vectors of the multiple user basic information data items.

9. The big data-based live streaming e-commerce marketing management system according to claim 8 is characterized in that: The target user portrait generation module includes: A user information feature fusion unit, used to fuse the optimized user behavior information multimodal semantic feature vector and the user basic information semantic understanding feature vector to obtain a user portrait generation feature vector; The user portrait generating unit is used to generate a feature vector from the user portrait through a generator to generate a user portrait of a target user.

10. A live streaming e-commerce marketing management method based on big data, characterized in that: include: Obtain the user's historical viewing product information data, user comment area interaction data and user basic information of users who have stayed in the browsing area for a predetermined period of time; Extracting an optimized user behavior information multimodal semantic feature vector and a user basic information semantic understanding feature vector from the user's historical product viewing information data, the user comment area interaction data and the user's basic information; Based on the optimized multimodal semantic feature vector of user behavior information and the semantic understanding feature vector of user basic information, a user profile of the target user is generated.

Images