Content automatic generation system and method based on user data

Abstract

The invention relates to the field of content generation, and particularly discloses an automatic content generation system and method based on user data, which comprises the following steps: firstly, acquiring keyword text data input by a user and a history browsing news manuscript, then performing feature extraction and correlation analysis on the keyword text data and the history browsing news manuscript by using a deep learning technology, and finally, generating a content through a generator. According to the method, the user personalized news manuscript recommendation is generated, so that the requirements and interests of the user are more accurately understood, the problem of information overload is effectively reduced, and the user experience and the information acquisition efficiency are improved.

Description

Technical Field

[0001] The present application relates to the field of content generation, and more specifically, to a system and method for automatically generating content based on user data. Background Art

[0002] With the rapid development and widespread use of the Internet, we are facing the severe challenge of information overload. Users are often confused in the massive amount of information because it is difficult for them to find content that really meets their needs. Take search engines as an example. On Chinese Google alone, more than 10 million results can be found.

[0003] Modern intelligent recommendation technology mainly relies on users' browsing preferences, but this often leads to insufficient breadth of recommendations or too many documents, thus affecting the quality of news recommendation content generation and the overall user experience.

[0004] Therefore, a system and method for automatically generating content based on user data 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 system and method for automatically generating content based on user data, which first obtains the keyword text data and historically browsed news manuscripts input by the user, then uses deep learning technology to perform feature extraction and correlation analysis on the two, and finally generates personalized news manuscript recommendations for users through a generator, thereby more accurately understanding the needs and interests of users, thereby effectively reducing the problem of information overload and improving user experience and information acquisition efficiency.

[0006] According to one aspect of the present application, a system for automatically generating content based on user data is provided, comprising:

[0007] The news user data content acquisition module is used to obtain the keyword text data entered by the user and the historically browsed news manuscripts;

[0008] A news user data content extraction module, used to extract a semantic feature vector associated with the user input news keyword text and optimize the semantic understanding feature vector of the historically browsed news manuscript from the keyword text data input by the user and the historically browsed news manuscript;

[0009] The news manuscript personalized recommendation generation module is used to generate user personalized news manuscript recommendations based on the semantic feature vector associated with the user-input news keyword text and the optimized historical browsing news manuscript semantic understanding feature vector.

[0010] According to another aspect of the present application, a method for automatically generating content based on user data is provided, which includes:

[0011] Obtain keyword text data and historically browsed news manuscripts entered by the user;

[0012] Extracting a semantic feature vector associated with the user input news keyword text and optimizing a semantic understanding feature vector of the historically browsed news manuscript from the keyword text data input by the user and the historically browsed news manuscript;

[0013] Based on the semantic feature vector associated with the news keyword text input by the user and the semantic understanding feature vector of the optimized historical browsing news manuscripts, a user personalized news manuscript recommendation is generated.

[0014] Compared with the prior art, the present application provides a system and method for automatic content generation based on user data, which first obtains the keyword text data input by the user and the historically browsed news manuscripts, and then uses deep learning technology to perform feature extraction and correlation analysis on the two, and finally generates personalized news manuscript recommendations for the user through a generator, thereby more accurately understanding the user's needs and interests, thereby effectively reducing the problem of information overload and improving user experience and information acquisition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] 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.

[0016] Figure 1 A block diagram of a system for automatically generating content based on user data according to an embodiment of the present application.

[0017] Figure 2 It is a block diagram of a news user data content extraction module in a user data-based content automatic generation system according to an embodiment of the present application.

[0018] Figure 3 It is a block diagram of a keyword text data feature extraction unit in a content automatic generation system based on user data according to an embodiment of the present application.

[0019] Figure 4 The present invention is a block diagram of a historically browsed news manuscript feature extraction unit in a content automatic generation system based on user data according to an embodiment of the present application.

[0020] Figure 5 The block diagram is a news manuscript personalized recommendation generation module in the automatic content generation system based on user data according to an embodiment of the present application.

[0021] Figure 6The present invention is a flowchart of a method for automatically generating content based on user data according to an embodiment of the present application.

[0022] Figure 7 is a block diagram of an electronic device according to an embodiment of the present application. DETAILED DESCRIPTION

[0023] Below, the exemplary embodiments according to the present application will be described in detail with reference to the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present application, rather than all the embodiments of the present application, and it should be understood that the present application is not limited to the exemplary embodiments described here.

[0024] Figure 1 Schematic diagram of a system for automatically generating content based on user data in an embodiment of the present application. Figure 1 As shown, according to the user data-based content automatic generation system 100 according to the embodiment of the present application, it includes: a news user data content acquisition module 110, which is used to acquire the keyword text data input by the user and the historically browsed news manuscripts; a news user data content extraction module 120, which is used to extract the user-input news keyword text-associated semantic feature vector and the optimized historically browsed news manuscript semantic understanding feature vector from the user-input keyword text data and the historically browsed news manuscripts; a news manuscript personalized recommendation generation module 130, which is used to generate user personalized news manuscript recommendations based on the user-input news keyword text-associated semantic feature vector and the optimized historically browsed news manuscript semantic understanding feature vector.

[0025] In the above-mentioned content automatic generation system 100 based on user data, the news user data content acquisition module 110 is used to obtain the keyword text data and historical browsing news manuscripts input by the user. It should be understood that with the rapid popularization and widespread application of the Internet, users are facing the severe challenge of information overload. Users often feel confused in a large amount of information because they find it difficult to find content that really meets their needs. For example, searching on Chinese Google can only get more than 10 million results. Modern intelligent recommendation technology usually relies on the user's browsing preferences, but this often leads to too limited recommended content or too many documents, thereby affecting the quality of news recommendation generation and the overall experience of the user. Therefore, in the technical solution of the present application, by obtaining the keyword text data and historical browsing news manuscripts input by the user, and combining deep learning technology, by generating personalized news manuscript recommendations, the needs and interests of the user can be more accurately grasped, thereby effectively responding to the challenge of information overload and improving the user's experience and efficiency of information acquisition.

[0026] Specifically, in the technical solution of the present application, the keyword text data represents the current interests and needs of the user. By analyzing these keywords, the system can understand the topics and themes that the user is paying attention to. This data includes not only the keywords explicitly input by the user, but also may cover search history, instant input and other semantically related information. These data are summarized to help the system better understand the user's intention. Historical browsing news manuscripts provide important clues to the content and reading preferences of the user in the past. By analyzing the news manuscripts that the user has browsed in the past, the system can identify the user's long-term interest areas, preferred news types, and possible depth and breadth needs. This data is not limited to a single keyword, but reflects deeper interests and preferences through reading behavior and selection behavior, so it is essential to provide personalized, accurate and targeted content recommendations. Obtaining the keyword text data and historical browsing news manuscripts input by the user is to enable the content automatic generation system to more accurately understand and predict the user's information needs, so as to provide news content recommendation services that meet their personalized interests and preferences. This method not only enhances the user experience, but also improves the practicality and intelligence level of the system, so that users can be more efficient and satisfied in the process of information acquisition.

[0027] In the above-mentioned automatic content generation system 100 based on user data, the news user data content extraction module 120 is used to extract the semantic feature vector associated with the user-input news keyword text and optimize the semantic understanding feature vector of the historical browsing news manuscript from the keyword text data input by the user and the historical browsing news manuscript. It should be understood that in this way, the accuracy and personalization of content recommendations can be optimized. This method not only enables the system to adapt to changes in user needs in real time and dynamically, but also provides users with news content recommendations that are more in line with their interests and preferences through in-depth analysis and semantic understanding. By effectively utilizing these feature vectors, the automatic content generation system can accurately predict and meet users' information needs in the context of large amounts of information and strong diversity, thereby improving overall user experience and satisfaction.

[0028] Figure 2 FIG. 1 is a block diagram of a news user data content extraction module in a content automatic generation system based on user data according to an embodiment of the present application. Figure 2 As shown, in a specific embodiment of the present application, the news user data content extraction module 120 includes: a keyword text data feature extraction unit 121, which is used to perform feature extraction on the keyword text data input by the user to obtain a semantic feature vector associated with the user-input news keyword text; a historical browsing news manuscript feature extraction unit 122, which is used to perform feature extraction on the historical browsing news manuscript to obtain the optimized historical browsing news manuscript semantic understanding feature vector.

[0029] It should be understood that keyword text data may contain rich semantic information, such as word meaning, context, emotional color, etc., which is crucial to accurately grasp user intentions. Through feature extraction, the system can filter out the most representative and information-rich features from a large amount of text data, thereby effectively capturing the themes and topics that users are concerned about. In the process of feature extraction, various natural language processing techniques are usually applied, such as word vector models, semantic analysis, and sentiment recognition. These technologies can deeply analyze the meaning and relevance of each keyword in a specific context, and then convert it into a mathematical representation to form a semantic feature vector associated with the user's input news keyword text. These vectors not only reflect the semantic relationship between words, but may also take into account their importance and influence in a specific topic or field, which helps the system to more accurately match and recommend relevant content.

[0030] Furthermore, feature extraction of historically browsed news manuscripts is to deepen the system's understanding of user interests and preferences, and thus provide more personalized and accurate services. Historically browsed manuscripts record users' past reading and browsing behaviors, which reflect users' long-term interests and information acquisition preferences. By extracting features from these manuscripts, the system can analyze the key information in the manuscripts, including topics, content, emotional colors, etc., to form a representative and information-rich feature vector. The feature extraction process usually involves a variety of natural language processing technologies, such as text vectorization, topic modeling, entity recognition, and sentiment analysis. These technologies can help the system understand the semantic structure and content association of the manuscript, and identify important features and patterns therein. Through mathematization and vectorization, the system can convert complex text information into a form that can be understood and processed by machines, thereby providing strong support for subsequent recommendation and response tasks.

[0031] Figure 3 FIG. 1 is a block diagram of a keyword text data feature extraction unit in a content automatic generation system based on user data according to an embodiment of the present application. Figure 3 As shown, in a specific embodiment of the present application, the keyword text data feature extraction unit 121 includes: a text data keyword semantic processing subunit 1211, used to perform keyword semantic processing on the keyword text data input by the user to obtain a two-dimensional semantic feature matrix of the user-input keyword text; a keyword text two-dimensional convolution subunit 1212, used to pass the user-input keyword text two-dimensional convolutional neural network as a feature encoder through the user-input keyword text two-dimensional convolutional neural network to obtain a semantic feature vector associated with the user-input news keyword text.

[0032] It should be understood that keyword semantic processing involves a variety of natural language processing technologies, such as word vector models, semantic analysis, and context understanding. These technologies can map each keyword into a high-dimensional semantic space, where each dimension represents a semantic feature or context information. The generation of a two-dimensional semantic feature matrix usually includes two main steps: first, semantic representation of keywords, that is, converting each keyword into a vector that captures the semantic information of the keyword; second, organizing these vectors into a matrix form so that each row or column represents a specific semantic dimension or semantic relationship. The establishment of this matrix can help the system understand and analyze the text data input by users more comprehensively, and improve the accuracy and efficiency of the system in understanding and responding to user needs. In practical applications, the two-dimensional semantic feature matrix can be used for a variety of tasks, such as information retrieval, content recommendation, and natural language generation. Through this matrix, the system can achieve structured representation and semantic association analysis of text data, thereby optimizing the effect of personalized recommendation and customized services.

[0033] Furthermore, the two-dimensional semantic feature matrix of the keyword text input by the user is passed through the two-dimensional convolutional neural network of the user input keyword text as a feature encoder in order to further extract and process the deep semantic information between the keywords. As an effective deep learning model, the convolutional neural network has been widely used in the field of natural language processing, and it performs well in text feature extraction and semantic understanding. The two-dimensional semantic feature matrix is ​​input into the convolutional neural network, and the local features and global semantic information between the keywords can be effectively captured by using the structures such as the convolutional layer and the pooling layer of the convolutional neural network. The convolution operation can extract the semantic features of keywords at different scales, thereby more comprehensively expressing the semantic meaning and relevance of the text. The pooling operation helps to reduce the feature dimension and extract the most significant features, so that the model can process and understand the text data more efficiently. By using the convolutional neural network as a feature encoder, the original two-dimensional semantic feature matrix can be converted into a compact and information-rich semantic feature vector. These vectors not only contain the semantic relationship between keywords, but also take into account their importance and influence in the context, thereby more accurately reflecting the text association and needs input by the user. Specifically, each layer of the two-dimensional convolutional neural network for user-input keyword text as a feature encoder performs 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, so that the last layer of the two-dimensional convolutional neural network for user-input keyword text as a feature encoder outputs the semantic feature vector associated with the user-input news keyword text, wherein the input of the two-dimensional convolutional neural network for user-input keyword text as a feature encoder is the two-dimensional semantic feature matrix of the user-input keyword text.

[0034] In a specific embodiment of the present application, the text data keyword semantic processing subunit 1211 includes: passing the keyword text data input by the user through a user-input keyword text bidirectional long short-term memory neural network to obtain a plurality of user-input keyword text semantic feature vectors; and arranging the plurality of user-input keyword text semantic feature vectors in two dimensions to obtain a two-dimensional semantic feature matrix of the user-input keyword text.

[0035] It should be understandable that the keyword text entered by the user contains the user's specific needs and interests, but there may be multiple ways of expression and semantic meanings. By introducing the bidirectional LSTM network, the system can simultaneously consider the information transmission of the text sequence in the forward and reverse directions, thereby fully capturing the semantic background and importance of each keyword in the context. This bidirectional learning capability makes the generated semantic feature vector more comprehensive and accurate. The bidirectional LSTM network can effectively capture the semantic coherence and contextual information between keywords by learning the long-term dependencies in the text sequence. This is especially important for processing longer text sequences, because it can distinguish the semantic changes and importance of keywords in different contexts, thereby improving the performance of the model in complex semantic analysis tasks.

[0036] Furthermore, the semantic feature vectors of keyword texts input by multiple users are arranged in two dimensions, aiming to integrate and structure the semantic information of each text input for further information processing and analysis. This matrix representation helps to more efficiently understand and utilize the semantic content of text data, thereby improving the accuracy and efficiency of various natural language processing tasks. Arranging the semantic feature vectors into a two-dimensional matrix can effectively preserve the relative position and relationship of each text input in the semantic space. This is very important for analyzing the semantic similarity between texts, building semantic graphs, or conducting content-based recommendation systems, because it provides a clear structured framework that makes the semantic information of text data easier to understand and process. The two-dimensional semantic feature matrix can help the system better capture the complex semantic associations between multiple text inputs. By arranging each semantic feature vector in a certain order, the matrix not only reflects the individual semantic features of each text input, but also shows their position and importance in the overall context. This global perspective helps to deeply analyze and understand the overall semantic structure of the text collection.

[0037] Figure 4 FIG. 1 is a block diagram of a feature extraction unit for historically browsed news articles in a content automatic generation system based on user data according to an embodiment of the present application. Figure 4As shown, in a specific embodiment of the present application, the historical browsing news manuscript feature extraction unit 122 includes: a historical browsing news manuscript word segmentation subunit 1221, used to segment the historical browsing news manuscript to obtain multiple historical browsing news manuscript data items; a historical browsing news manuscript semantic encoding subunit 1222, used to semantically encode the multiple historical browsing news manuscript data items to obtain the optimized historical browsing news manuscript semantic understanding feature vector.

[0038] It should be understood that word segmentation is one of the basic steps in natural language processing, and its purpose is to divide a continuous text sequence into meaningful words or phrases, each of which can independently express certain semantic content and information as a data item. Word segmentation can help the system understand the specific content and theme of each news manuscript. By segmenting the text according to words, the system can identify key words and phrases in the document, thereby accurately capturing the core information and elements of the document. In addition, by performing word segmentation on a large number of historical browsing manuscripts, the system can collect and organize all the words that appear in the document and their frequency statistics. These data items not only help analyze the semantic structure and relevance of the document, but also support the semantic understanding and information retrieval functions of the vocabulary, so that the system can better respond to the diverse information needs of users.

[0039] Furthermore, semantic encoding is performed on multiple historically browsed news document data items, aiming to convert text information into a numerical representation with semantic richness and information accuracy, thereby deepening the system's ability to understand document content and topics. Among them, semantic encoding involves mapping natural language text into mathematical space, allowing computers to process and analyze text information more effectively. Through semantic encoding, the system is able to capture the hidden semantic structure and relevance in the document. Traditional text data items are discrete words or phrases, while semantic encoding converts each word or phrase into a continuous vector representation. The position and direction of these vectors in high-dimensional space reflect the semantic similarity and relationship between words, thereby more comprehensively expressing the semantic content and information content of the document. Semantic encoding helps to improve the accuracy and efficiency of the system in information retrieval and text understanding tasks. By comparing the semantically encoded vectors, the system can quickly calculate the similarity between documents and accurately locate the content of interest to users.

[0040] In a specific embodiment of the present application, the historical browsing news manuscript semantic encoding subunit 1222 includes: passing the multiple historical browsing news manuscript data items through a historical browsing news manuscript word embedding model to obtain multiple historical browsing news manuscript data item word embedding vectors; passing the multiple historical browsing news manuscript data item word embedding vectors through a historical browsing news manuscript semantic encoder based on a context encoder to obtain a historical browsing news manuscript semantic understanding feature vector; based on the semantic feature vector associated with the user-input news keyword text, performing backward machine learning viewpoint metric perception on the historical browsing news manuscript semantic understanding feature vector to obtain an optimized historical browsing news manuscript semantic understanding feature vector.

[0041] It should be understood that the word embedding vector of the historical browsing news manuscript data item can more accurately express the semantic content and topic features in each manuscript. Compared with traditional word indexing or one-hot encoding, the word embedding vector can represent words in a continuous manner and can retain the relative relationship between words in the semantic space. This representation method enables the system to better understand the meaning and context of the text, thereby improving the accuracy of text processing and information retrieval. The word embedding vector of the historical browsing news manuscript data item helps to realize the semantic similarity calculation and correlation analysis of the text. By comparing the word embedding vectors of different words in the manuscript, the system can quickly determine the semantic similarity between texts, thereby supporting applications such as information retrieval, text clustering and content recommendation. This capability is of great significance for building intelligent recommendation systems and personalized information services, and can improve user experience and satisfaction. The application of the word embedding model of historical browsing news manuscripts can also support text generation and natural language processing tasks.

[0042] Furthermore, the word embedding vectors of multiple historically browsed news manuscript data items are passed through the historically browsed news manuscript semantic encoder based on the context encoder, in order to further enhance the system's deep semantic understanding and precise expression capabilities of text information. This method combines word embedding technology and context understanding model, which can effectively capture the complex semantic structure and information meaning in the document, thereby providing key support for intelligent text analysis and application. The historically browsed news manuscript semantic encoder based on the context encoder can consider the context and role of each word embedding vector in its context. Compared with simple static word embedding, context encoders such as recurrent neural networks (RNN), long short-term memory networks (LSTM) or Transformer can dynamically adjust the representation of word embedding by learning the dependency and sequence information in the text, so that the vector of each word can more accurately reflect its semantic role and contextual relationship in the text. The historically browsed news manuscript semantic encoder generates richer and more advanced semantic understanding feature vectors by comprehensively considering the overall structure and semantic features of the text. These vectors not only contain semantic information at the word level, but also cover the semantic relationship and content expression at the document level, enabling the system to perform text understanding and information reasoning in a wider context. Among them, the historical browsing news manuscript semantic encoder based on the context encoder can also support multimodal data processing and cross-document semantic analysis. By integrating text information with other modal data (such as images, videos, etc.), the system can achieve more comprehensive and in-depth semantic inference, and provide technical support and solutions for cross-domain intelligent applications. Specifically, the converter-based Bert model of the historical browsing news manuscript semantic encoder based on the context encoder is used to perform global context semantic encoding on the word embedding vectors of the multiple historical browsing news manuscript data items to obtain multiple feature vectors; and, the multiple feature vectors are cascaded to obtain a historical browsing news manuscript semantic understanding feature vector.

[0043] Furthermore, in the technical solution of the present application, it is considered that the keyword text entered by the user and the news manuscripts browsed historically may differ in content and theme. The keywords entered by the user may reflect the user's current interests more specifically and directly, while the historical browsing data reflects the user's past interests and preferences. The user's interests may change over time. The historical browsing data may not fully capture the user's current interest changes, so when generating recommendations, if only relying on historical data, the user's latest interests may be ignored. The semantic understanding feature vector of the historical browsing news manuscript may rely more on the contextual information of the news manuscript, while the keywords entered by the user may focus more on the semantics of the keywords themselves. There may be differences in semantic understanding between the two, which leads to the lack of a posteriori-priori causal association between the semantic understanding feature vector of the historical browsing news manuscript relative to the semantic feature vector associated with the user-input news keyword text. In order to solve this problem, in the technical solution of the present application, based on the semantic feature vector associated with the user-input news keyword text, the viewpoint metric perception of the semantic understanding feature vector of the historical browsing news manuscript is backward machine-learned to obtain an optimized semantic understanding feature vector of the historical browsing news manuscript.

[0044] Among them, based on the semantic feature vector associated with the user-input news keyword text, the viewpoint metric perception of the semantic understanding feature vector of the historically browsed news manuscripts is performed by backward machine learning to obtain the optimized semantic understanding feature vector of the historically browsed news manuscripts, including: calculating the positional difference between the semantic understanding feature vector of the historically browsed news manuscripts and the semantic feature vector associated with the user-input news keyword text to obtain a first viewpoint metric representation vector; calculating the positional point multiplication between the semantic understanding feature vector of the historically browsed news manuscripts and the semantic feature vector associated with the user-input news keyword text to obtain a second viewpoint metric representation vector; calculating the positional point multiplication between the semantic understanding feature vector of the historically browsed news manuscripts and the semantic feature vector associated with the user-input news keyword text to obtain a second viewpoint metric representation vector. The positions of the quantities are summed up to obtain a third viewpoint measurement representation vector; the gradient amplitude of each position in the first viewpoint measurement representation vector, the second viewpoint measurement representation vector and the third viewpoint measurement representation vector is calculated to obtain a first viewpoint measurement perception vector, a second viewpoint measurement perception vector and a third viewpoint measurement perception vector; the first viewpoint measurement perception vector, the second viewpoint measurement perception vector and the third viewpoint measurement perception vector are input into a backward machine learning reverse multi-viewpoint perception optimization module to obtain a reverse multi-viewpoint perception optimization representation vector; the positional point multiplication between the reverse multi-viewpoint perception optimization representation vector and the historical browsing news manuscript semantic understanding feature vector is calculated to obtain the optimized historical browsing news manuscript semantic understanding feature vector.

[0045] The backward machine learning reverse multi-view perception optimization module is expressed as follows:

[0046]

[0047] in, and represents the eigenvalues ​​of the i-th position of the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector, respectively, and μ represents the three eigenvalues ​​of the corresponding positions in the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector , σ is the standard deviation of the three eigenvalues ​​at corresponding positions in the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector, represents the maximum value of the three eigenvalues ​​at the corresponding positions in the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector, ReLU represents a linear rectification function, softmax represents a normalized exponential function, and v ri Represents the eigenvalue of the i-th position of the inverse multi-view perceptual optimization representation vector.

[0048] Specifically, in the technical solution of the present application, based on the semantic feature vector associated with the user-input news keyword text, the semantic understanding feature vector of the historical browsing news manuscript is optimized by backward machine learning. The process includes first calculating the position difference, dot product and sum between the semantic understanding feature vector of the historical browsing news manuscript and the semantic feature vector associated with the user-input news keyword text through multi-view analysis (Multi-View Analysis) to obtain three viewpoint measurement representation vectors, and quantify the relationship between features from different angles. Then, the gradient amplitude of each position in these viewpoint measurement representation vectors is calculated through backward error perception to identify key change points to provide a basis for feature optimization.

[0049] Then, the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector are input into the backward machine learning reverse multi-viewpoint perception optimization module, wherein the backward machine learning reverse multi-viewpoint perception optimization module captures the significant comprehensive ability factor of the feature importance perception (Feature Importance Perception) of the feature sub-dimension distribution of the corresponding positions in the first to third viewpoint metric perception vectors by adopting a squeeze-incentive mechanism composed of -functions. Then, the position point multiplication between the reverse multi-viewpoint perception optimization representation vector and the historical browsing news manuscript semantic understanding feature vector is calculated to perform backward optimization perception (Backward Optimization Perception) on the historical browsing news manuscript semantic understanding feature vector to adjust the historical browsing news manuscript semantic understanding feature vector so as to better associate it with the semantic feature vector associated with the user input news keyword text, thereby enhancing the causal relationship between the semantic understanding feature vector of the historical browsing news manuscript and the semantic feature vector associated with the user input news keyword text.

[0050] In the above-mentioned user data-based content automatic generation system 100, the news manuscript personalized recommendation generation module 130 is used to generate user personalized news manuscript recommendations based on the semantic feature vector associated with the news keyword text input by the user and the semantic understanding feature vector of the optimized historical browsing news manuscript. It should be understood that through the semantic feature vector associated with the news keyword text input by the user, the system can understand the user's current focus and needs. After semantic encoding and feature extraction, these keywords can accurately reflect the user's interest topics and information preferences, providing a basis for the precise positioning of personalized recommendations. Using the semantic understanding feature vector of optimized historical browsing news manuscripts, the system can analyze and understand the user's past browsing behavior and preferences. By comparing the user's current keyword vector and historical semantic vector, the system can identify news topics and content types that are highly matched with the user's interests, thereby optimizing the personalization of the recommendation results. The personalized recommendation system dynamically adjusts and optimizes the recommendation results by comprehensively considering factors such as keyword matching, semantic similarity and user historical preferences. This recommendation method based on multi-dimensional feature vectors not only improves the recommendation accuracy and user satisfaction, but also enhances the system's ability to adapt to changes in user needs in a complex information environment. By providing personalized information services, the system can enhance the frequency and depth of users' access to diverse news reports, thereby promoting a more comprehensive understanding and engagement of users on a wide range of topics.

[0051] Figure 5 FIG. 1 is a block diagram of a personalized news article recommendation generation module in a content automatic generation system based on user data according to an embodiment of the present application. Figure 5As shown, in a specific embodiment of the present application, the news article personalized recommendation generation module 130 includes: a news article data feature fusion unit 131, which is used to fuse the semantic feature vector associated with the user-input news keyword text and the optimized historical browsing news article semantic understanding feature vector to obtain a news article recommendation generation feature vector; a news article recommendation personalized generation unit 132, which is used to pass the news article recommendation generation feature vector through a generator to generate user personalized news article recommendations.

[0052] It should be understood that the semantic feature vector of the user's current input and historical browsing data is integrated to generate a recommendation feature vector that integrates short-term and long-term user interests, enhancing the personalization and adaptability of the recommendation results. This fusion method not only improves the accuracy of the recommendation system, but also optimizes the user experience, making the recommended news articles closer to the user's actual needs and interests. By adopting deep learning and machine learning technologies, the system can process and analyze user input and historical data in real time in a big data environment, and quickly generate highly adaptable and responsive recommendation results.

[0053] Furthermore, the feature vector generated by the news manuscript recommendation is used by the generator to generate personalized news manuscript recommendations for users, in order to utilize the ability of the generation model to generate more targeted and attractive recommendation content according to the personalized needs and historical behavior data of users. This method combines deep learning generation technology with a personalized recommendation system, and can achieve efficient and intelligent information filtering and customized services in a big data environment. By generating news manuscript recommendations through the generator, the system can generate feature vectors based on user-specific recommendations, and combine generation models such as variational autoencoders (VAE), generative adversarial networks (GANs) or autoregressive models (such as GPT), so as to maintain consistency in semantics and syntax and effectively create new recommended content. This generation process not only takes into account the user's interests and behavior data, but also comprehensively considers the diversity and creativity of the text in the generation process, providing users with a more attractive and rich recommendation experience. This method not only meets the needs of personalized recommendations, but also opens up new possibilities and prospects for information dissemination and intelligent services in the digital age.

[0054] In summary, the embodiment of the present application first obtains the keyword text data input by the user and the historically browsed news manuscripts, then uses deep learning technology to perform feature extraction and correlation analysis on the two, and finally generates personalized news manuscript recommendations for the user through a generator, thereby more accurately understanding the user's needs and interests, thereby effectively reducing the problem of information overload and improving user experience and information acquisition efficiency.

[0055] As described above, the automatic content generation system 100 based on user data according to the embodiment of the present application can be implemented in various terminal devices. In one example, the automatic content generation system 100 based on user data can be integrated into the terminal device as a software module and / or a hardware module. For example, the automatic content generation system 100 based on user data can be a software module in the operating system of the terminal device, or can be an application developed for the terminal device; of course, the automatic content generation system 100 based on user data can also be one of the many hardware modules of the terminal device.

[0056] Alternatively, in another example, the user data-based content automatic generation system 100 and the terminal device may also be separate devices, and the user data-based content automatic generation 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.

[0057] Figure 6 Flow chart of a method for automatically generating content based on user data according to an embodiment of the present application. Figure 6 As shown, according to the automatic content generation method based on user data of an embodiment of the present application, it includes: S110, obtaining keyword text data input by the user and historically browsed news manuscripts; S120, extracting semantic feature vectors associated with the user-input news keyword text and optimized semantic understanding feature vectors of historically browsed news manuscripts from the keyword text data input by the user and the historically browsed news manuscripts; S130, generating user personalized news manuscript recommendations based on the semantic feature vectors associated with the user-input news keyword text and the optimized semantic understanding feature vectors of historically browsed news manuscripts.

[0058] Here, those skilled in the art can understand that the specific operations of each step in the above-mentioned method for automatically generating content based on user data have been described in the above reference. Figures 1 to 5 The description of the automatic content generation system based on user data has been introduced in detail, and therefore, its repeated description will be omitted.

[0059] Below, reference Figure 7 To describe an electronic device according to an embodiment of the present application.

[0060] Figure 7 A block diagram of an electronic device according to an embodiment of the present application is illustrated.

[0061] like Figure 7As shown, the electronic device 10 includes an input device 11, an input interface 12, a central processing unit 13, a memory 14, an output interface 15, an output device 16, and a bus 17. The input interface 12, the central processing unit 13, the memory 14, and the output interface 15 are connected to each other through the bus 17, and the input device 11 and the output device 16 are connected to the bus 17 through the input interface 14 and the output interface 15, respectively, and then connected to other components of the electronic device 10.

[0062] Specifically, the input device 11 receives input information from the outside and transmits the input information to the central processing unit 13 through the input interface 12; the central processing unit 13 processes the input information based on the computer executable instructions stored in the memory 14 to generate output information, temporarily or permanently stores the output information in the memory 14, and then transmits the output information to the output device 16 through the output interface 15; the output device 16 outputs the output information to the outside of the electronic device 10 for user use.

[0063] In one embodiment, Figure 7 The electronic device 10 shown can be implemented as a network device, which may include: a memory configured to store programs; a processor configured to run the programs stored in the memory to execute any one of the content automatic generation methods based on user data described in the above embodiments.

[0064] According to an embodiment of the present application, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present application includes a computer program product, which includes a computer program tangibly contained on a machine-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from a network, and / or installed from a removable storage medium.

[0065] It will be appreciated by those skilled in the art that all or some of the steps, systems, and functional modules / units in the methods disclosed above may be implemented as software, firmware, hardware, and appropriate combinations thereof. In hardware implementations, the division between the functional modules / units mentioned in the above description does not necessarily correspond to the division of physical components; for example, a physical component may have multiple functions, or a function or step may be performed by several physical components in cooperation. Some or all physical components may be implemented as software executed by a processor, such as a central processing unit, a digital signal processor, or a microprocessor, or implemented as hardware, or implemented as an integrated circuit, such as an application-specific integrated circuit. Such software may be distributed on a computer-readable medium, which may include a computer storage medium (or non-transitory medium) and a communication medium (or temporary medium). As known to those skilled in the art, the term computer storage medium includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storing information (such as computer-readable instructions, data structures, program modules, or other data). Computer storage media include, but are not limited to, RAM, ROM, EEPROM, flash memory or other memory technology, CD-ROM, digital versatile disks (DVD) or other optical disk storage, magnetic cassettes, magnetic tapes, magnetic disk storage or other magnetic storage devices, or any other medium that can be used to store the desired information and can be accessed by a computer. In addition, it is well known to those skilled in the art that communication media typically contain computer-readable instructions, data structures, program modules, or other data in a modulated data signal such as a carrier wave or other transport mechanism, and may include any information delivery media.

[0066] It is to be understood that the above embodiments are merely exemplary embodiments used to illustrate the principles of the present application, but the present application is not limited thereto. For those skilled in the art, various modifications and improvements can be made without departing from the spirit and substance of the present application, and these modifications and improvements are also considered to be within the scope of protection of the present application.

Claims

1. A content automatic generation system based on user data, characterized in that: include: The news user data content acquisition module is used to obtain the keyword text data entered by the user and the historically browsed news manuscripts; A news user data content extraction module, used to extract a semantic feature vector associated with the user input news keyword text and optimize the semantic understanding feature vector of the historically browsed news manuscript from the keyword text data input by the user and the historically browsed news manuscript; The news manuscript personalized recommendation generation module is used to generate user personalized news manuscript recommendations based on the semantic feature vector associated with the user-input news keyword text and the optimized historical browsing news manuscript semantic understanding feature vector.

2. The automatic content generation system based on user data according to claim 1 is characterized in that: The news user data content extraction module includes: A keyword text data feature extraction unit, used for performing feature extraction on the keyword text data input by the user to obtain a semantic feature vector associated with the news keyword text input by the user; The historical browsing news manuscript feature extraction unit is used to extract features from the historical browsing news manuscript to obtain the optimized historical browsing news manuscript semantic understanding feature vector.

3. The automatic content generation system based on user data according to claim 2 is characterized in that: The keyword text data feature extraction unit includes: A text data keyword semantic processing subunit, used for performing keyword semantic processing on the keyword text data input by the user to obtain a two-dimensional semantic feature matrix of the keyword text input by the user; The keyword text two-dimensional convolution subunit is used to pass the user-input keyword text two-dimensional semantic feature matrix through the user-input keyword text two-dimensional convolution neural network as a feature encoder to obtain the user-input news keyword text associated semantic feature vector.

4. The automatic content generation system based on user data according to claim 3 is characterized in that: The text data keyword semantic processing subunit includes: Passing the keyword text data input by the user through a user-input keyword text bidirectional long short-term memory neural network to obtain a plurality of user-input keyword text semantic feature vectors; The plurality of semantic feature vectors of the user-input keyword texts are arranged in two dimensions to obtain a two-dimensional semantic feature matrix of the user-input keyword texts.

5. The automatic content generation system based on user data according to claim 4 is characterized in that: The historical browsing news manuscript feature extraction unit includes: A historical browsing news manuscript word segmentation subunit is used to segment the historical browsing news manuscript to obtain a plurality of historical browsing news manuscript data items; The historical browsing news manuscript semantic encoding subunit is used to perform semantic encoding on the multiple historical browsing news manuscript data items to obtain the optimized historical browsing news manuscript semantic understanding feature vector.

6. The automatic content generation system based on user data according to claim 5, characterized in that: The historical browsing news manuscript semantic encoding subunit includes: Passing the plurality of historically browsed news manuscript data items through a historically browsed news manuscript word embedding model to obtain a plurality of historically browsed news manuscript data item word embedding vectors; Passing the word embedding vectors of the plurality of historical browsing news manuscript data items through a historical browsing news manuscript semantic encoder based on a context encoder to obtain a historical browsing news manuscript semantic understanding feature vector; Based on the semantic feature vector associated with the news keyword text input by the user, the viewpoint metric perception of the semantic understanding feature vector of the historically browsed news manuscripts is performed through backward machine learning to obtain the optimized semantic understanding feature vector of the historically browsed news manuscripts.

7. The automatic content generation system based on user data according to claim 6 is characterized in that: Based on the semantic feature vector associated with the news keyword text input by the user, the viewpoint metric perception of the semantic understanding feature vector of the historical browsing news manuscript is backward machine-learned to obtain the optimized semantic understanding feature vector of the historical browsing news manuscript, including: Calculating the position difference between the semantic understanding feature vector of the historically browsed news manuscript and the semantic feature vector associated with the user input news keyword text to obtain a first viewpoint metric representation vector; Calculate the position point multiplication between the semantic understanding feature vector of the historical browsed news manuscript and the semantic feature vector associated with the user input news keyword text to obtain a second viewpoint metric representation vector; Calculating the positional sum of the semantic understanding feature vector of the historically browsed news manuscript and the semantic feature vector associated with the user-input news keyword text to obtain a third viewpoint metric representation vector; Calculating the gradient magnitude of each position in the first viewpoint metric representation vector, the second viewpoint metric representation vector, and the third viewpoint metric representation vector to obtain a first viewpoint metric perception vector, a second viewpoint metric perception vector, and a third viewpoint metric perception vector; Inputting the first viewpoint metric perception vector, the second viewpoint metric perception vector and the third viewpoint metric perception vector into a backward machine learning inverse multi-view perception optimization module to obtain a reverse multi-view perception optimization representation vector; The optimized historical browsing news manuscript semantic understanding feature vector is obtained by calculating the position point multiplication between the reverse multi-view perception optimization representation vector and the historical browsing news manuscript semantic understanding feature vector.

8. The automatic content generation system based on user data according to claim 7 is characterized in that: The news manuscript personalized recommendation generation module includes: A news manuscript data feature fusion unit, used for fusing the semantic feature vector associated with the news keyword text input by the user and the semantic understanding feature vector of the optimized historical browsing news manuscript to obtain a news manuscript recommendation generation feature vector; The news article recommendation personalized generation unit is used to generate a feature vector for the news article recommendation through a generator to generate a user personalized news article recommendation.

9. A method for automatically generating content based on user data, characterized in that: include: Obtain keyword text data and historically browsed news manuscripts entered by the user; Extracting a semantic feature vector associated with the user input news keyword text and optimizing a semantic understanding feature vector of the historically browsed news manuscript from the keyword text data input by the user and the historically browsed news manuscript; Based on the semantic feature vector associated with the news keyword text input by the user and the semantic understanding feature vector of the optimized historical browsing news manuscripts, a user personalized news manuscript recommendation is generated.

10. The method for automatically generating content based on user data according to claim 9, characterized in that: Extracting a semantic feature vector associated with the user input news keyword text and optimizing a semantic understanding feature vector of the historically browsed news manuscript from the keyword text data input by the user and the historically browsed news manuscript, including: Performing feature extraction on the keyword text data input by the user to obtain a semantic feature vector associated with the news keyword text input by the user; Feature extraction is performed on the historical browsing news manuscript to obtain the optimized historical browsing news manuscript semantic understanding feature vector.

Images