Online Analysis Method and Device for New Media Data

Abstract

The present application discloses a method and device for online analysis of new media data. The server method of the BI system includes: simultaneously collecting user interaction data about a target product from multiple new media platforms through a preset distributed data collection architecture; extracting emotional description information that can reflect the emotional tendency of users towards the target product from the user interaction data of each new media platform to obtain multi-source emotional description information; inputting the multi-source emotional description information into a preset new media data online analysis component to output the user emotional fusion characteristics corresponding to the target product; and mining the user behavior pattern information and preference information of the users towards the target product within a preset period according to the user emotional fusion characteristics and a pre-established feature knowledge base, and storing them in the database of the BI system. Therefore, by adopting the embodiments of the present application, cross-platform data integration can be achieved, and at the same time, user behavior patterns and preferences can be discovered from a large amount of data.

Description

Technical Field

[0001] This application relates to the field of computer technology, and particularly to a method and device for online analysis of new media data. Background Art

[0002] In the new media era, enterprises and content creators increasingly rely on user interaction data on various new media platforms to guide product development and content creation. Such data includes, but is not limited to, comments on video content by users, live broadcast interactions, and evaluation feedback on products. However, how to effectively utilize this data to obtain valuable insights remains a challenge.

[0003] In related technologies, a centralized data management method is usually adopted, which mainly relies on a single database to store and analyze data collected from specific new media platforms. Due to system design limitations, these methods can usually only process data from a single platform and cannot achieve cross-platform data integration, resulting in one-sided analysis results. At the same time, these systems usually lack advanced data analysis and data mining functions and are difficult to discover deep user behavior patterns and preferences from a large amount of data. Summary of the Invention

[0004] Embodiments of this application provide a method and device for online analysis of new media data. To provide a basic understanding of some aspects of the disclosed embodiments, a simple summary is given below. This summary part is not a general review, nor is it intended to identify key / important constituent elements or delineate the protection scope of these embodiments. Its sole purpose is to present some concepts in a simple form as a preamble to the subsequent detailed description.

[0005] In a first aspect, embodiments of this application provide a method for online analysis of new media data, which is applied to the server side of a BI system. The method includes:

[0006] When the duration between the current moment and the last data analysis moment is equal to a preset period, collect user interaction data regarding a target product simultaneously from multiple new media platforms through a preset distributed data collection architecture. The preset distributed data collection architecture includes multiple data collection channels, and each data collection channel is associated with a data collection interface provided by each new media platform;

[0007] Extract emotional description information that can reflect the emotional tendency of users towards the target product from the user interaction data of each new media platform to obtain multi-source emotional description information;

[0008] Input the multi-source emotional description information into a preset online analysis component for new media data and output the user emotional fusion feature corresponding to the target product;

[0009] Based on the user emotion fusion features and the pre-established feature knowledge base, mine the user behavior pattern information and preference information of the target product within the preset period;

[0010] In the database of the BI system, store the ternary mapping relationship between the target product, the behavior pattern information, and the preference information.

[0011] In a second aspect, an embodiment of the present application provides a new media data online analysis device, which includes:

[0012] A data collection module, configured to, when the duration between the current moment and the previous data analysis moment is equal to the preset period, collect user interaction data about the target product from multiple new media platforms simultaneously through a preset distributed data collection architecture. The preset distributed data collection architecture includes multiple data collection channels, and each data collection channel is associated with the data collection interfaces provided by each new media platform one by one;

[0013] An emotion description information extraction module, configured to extract emotion description information that can reflect the user's emotional tendency towards the target product from the user interaction data of each new media platform to obtain multi-source emotion description information;

[0014] A data input module, configured to input the multi-source emotion description information into a preset new media data online analysis component and output the user emotion fusion features corresponding to the target product;

[0015] A data output module, configured to, based on the user emotion fusion features and the pre-established feature knowledge base, mine the user behavior pattern information and preference information of the target product within the preset period;

[0016] A ternary mapping relationship storage module, configured to store the ternary mapping relationship between the target product, the behavior pattern information, and the preference information in the database of the BI system.

[0017] The technical solution provided by the embodiment of the present application may include the following beneficial effects:

[0018] In the embodiments of the present application, on the one hand, data collection from multiple platforms is performed through a preset distributed data collection architecture. Since this architecture includes multiple data collection channels, each channel is associated with the data collection interfaces of each new media platform one by one, so that cross-platform data collection can be realized simultaneously, improving the coverage and diversity of data. On the other hand, emotional description information is extracted from the user interaction data of each new media platform and input into a preset online new media data analysis component. By combining a pre-established feature knowledge base, behavior pattern information and preference information are mined. This emotional description information can comprehensively represent the emotional tendency of users, and this preset online new media data analysis component can process multi-modal data, effectively integrating the complex interaction relationships existing between different types of data, extracting effective emotional features, so as to cooperate with the pre-established feature knowledge base to deeply discover user behavior patterns and preferences, and further provide guiding data for product development and content creation at the business layer.

[0019] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The accompanying drawings herein are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0021] Figure 1 It is a schematic flowchart of a method for online analysis of new media data provided by an embodiment of the present application;

[0022] Figure 2 It is a data schematic diagram of emotional description information reflecting the emotional tendency of users towards a target product provided by an embodiment of the present application;

[0023] Figure 3 It is a schematic diagram of the component architecture of a preset online new media data analysis component provided by an embodiment of the present application;

[0024] Figure 4 It is a data schematic diagram of a data table of the comprehensive emotional intensity value of each new media platform and each tendency theme provided by an embodiment of the present application;

[0025] Figure 5 It is a data schematic diagram of a pre-established feature knowledge base provided by an embodiment of the present application;

[0026] Figure 6 It is a data schematic diagram of a ternary mapping relationship provided by an embodiment of the present application;

[0027] Figure 7 It is a schematic flowchart of a method for training an emotional tendency recognition model provided by an embodiment of the present application;

[0028] Figure 8 is a schematic diagram of the model architecture of an emotion tendency recognition model provided by an embodiment of the present application;

[0029] Figure 9 is a schematic diagram of the structure of an online analysis device for new media data provided by an embodiment of the present application;

[0030] Figure 10 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present application. Detailed implementation manners

[0031] The following description and the accompanying drawings fully illustrate the specific implementation manners of the present application, enabling those skilled in the art to practice them.

[0032] It should be clear that the described embodiments are only a part of the embodiments of the present application, rather than all of them. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0033] When the following description refers to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all the implementation manners consistent with the present application. On the contrary, they are only examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0034] In the description of the present application, it should be understood that the terms "first", "second", etc. are only used for descriptive purposes and cannot be understood as indicating or implying relative importance. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to specific circumstances. In addition, in the description of the present application, unless otherwise stated, "a plurality of" means two or more. "And / or" describes the association relationship of associated objects and indicates that three relationships can exist. For example, A and / or B can represent: A exists alone, A and B exist simultaneously, and B exists alone. The character " / " generally represents an "or" relationship between the front and back associated objects.

[0035] Currently, the creation of video content mainly relies on centralized data management methods, which mainly rely on a single database to store and analyze data collected from specific new media platforms.

[0036] The inventors have realized that these methods can usually only process data from a single platform and cannot achieve cross-platform data integration, resulting in one-sided analysis results. At the same time, these systems usually lack advanced data analysis and data mining functions and are difficult to discover deep user behavior patterns and preferences from a large amount of data.

[0037] To solve the above problems, the present application provides a method and device for online analysis of new media data to address the issues existing in the above related technical problems. In an embodiment of the present application, on the one hand, data collection from multiple platforms is performed through a preset distributed data collection architecture. Since this architecture includes multiple data collection channels, each channel is associated one by one with the data collection interfaces of each new media platform, enabling simultaneous cross-platform data collection, thereby improving the coverage and diversity of data. On the other hand, emotional description information is extracted from the user interaction data of each new media platform and input into a preset online analysis component for new media data. In combination with a pre-established feature knowledge base, behavioral pattern information and preference information are mined. This emotional description information can comprehensively represent the emotional tendency of users, and the preset online analysis component for new media data can process multi-modal data, effectively integrating the complex interaction relationships existing between different types of data and extracting effective emotional features. Thus, in cooperation with the pre-established feature knowledge base, it is possible to deeply discover user behavioral patterns and preferences, and further provide guiding data for product development and content creation at the business layer. The following will be described in detail using exemplary embodiments.

[0038] The following will Figure 1 - be combined with the Figure 8 accompanying drawings to introduce in detail the method for online analysis of new media data provided in the embodiments of the present application. This method can be implemented relying on a computer program and can run on a new media data online analysis device based on the von Neumann architecture. This computer program can be integrated into an application or run as an independent tool-type application.

[0039] Please refer to Figure 1 , which is a schematic flowchart of a method for online analysis of new media data provided in an embodiment of the present application and is applied to the server side of a BI system. As Figure 1 shown, the method of the embodiment of the present application may include the following steps:

[0040] S101, when the duration between the current moment and the previous data analysis moment is equal to a preset period, collect user interaction data regarding a target product from multiple new media platforms simultaneously through a preset distributed data collection architecture. The preset distributed data collection architecture includes multiple data collection channels, and each data collection channel is associated with the data collection interface provided by each new media platform;

[0041] Among them, the preset period refers to the set time interval, which is used to specify when to collect data from the new media platform. For example, the preset period can be once a day, once a week, or once a month. The distributed data collection architecture is a design of a data collection system that allows data to be collected simultaneously from multiple data sources. This architecture is usually used to process large amounts of data and improve the efficiency of data collection. In the distributed data collection architecture, each data collection channel is an independent data collection path responsible for collecting data from a specific new media platform. New media platform: refers to an online platform where users can interact and communicate, such as social media, video sharing websites, blogs, etc. The data collection interface is a technical interface provided by the new media platform that allows external systems to collect data according to the platform's specifications and restrictions.

[0042] In some embodiments of the present application, the preset period can be set to once a week, that is, the user interaction data is collected once a week. A distributed data collection architecture is designed, which includes multiple data collection channels, and each channel is responsible for collecting data from different new media platforms. A data collection channel is set for each new media platform (such as Weibo, Douyin, TikTok, YouTube, etc.). Each data collection channel is associated with the data collection interface of the corresponding platform to ensure that data can be collected according to the platform's regulations.

[0043] In some embodiments of the present application, the server side of the BI system determines the target product (such as a wheelchair), and calculates the duration between the current moment and the last data analysis moment in real time. When the duration is equal to the preset period, user interaction data videos, comments, live comments, product feedback, etc. about the wheelchair are collected simultaneously from multiple new media platforms.

[0044] For example, when the preset period arrives (such as 8:00 am on Monday), the distributed data collection architecture is automatically started, and all data collection channels start working simultaneously. Each channel collects user interaction data about the target smartphone model from the associated new media platform, including comments, likes, shares, etc.

[0045] S102, extract the emotional description information that can reflect the user's emotional tendency towards the target product from the user interaction data of each new media platform to obtain multi-source emotional description information;

[0046] Among them, the user interaction data of each new media platform includes video comments, live comments, and product feedback.

[0047] In some embodiments of the present application, the specific process of extracting emotional description information that can reflect the emotional tendency of users towards the target product from the user interaction data of each new media platform includes: performing data preprocessing on the video comments, live comments, and product opinion feedback included in the user interaction data of each new media platform to obtain the vocabulary sequence and phrase sequence of each new media platform; the preprocessing includes data cleaning, data deduplication, data formatting, and data tokenization; inputting the vocabulary sequence and phrase sequence of each new media platform into a pre-trained emotional tendency recognition model, and outputting multiple emotional words corresponding to the user interaction data of each new media platform and the intensity of each emotional word; calculating the emotional intensity value of the user interaction data of each new media platform according to the intensity of each emotional word and the number of emotional words; performing topic analysis on the vocabulary sequence and phrase sequence of each new media platform to extract the tendency topics involved in the user interaction data of each new media platform; taking the emotional intensity value and tendency topic of the user interaction data of each new media platform as the emotional description information that can reflect the emotional tendency of users towards the target product.

[0048] Among them, video comments are the written feedback of users on video content, which may contain emotional expressions and evaluations of the video content. Live comments are the comments posted by users in real time during the live broadcast, and these comments usually reflect the immediate reactions of users to the live broadcast content. Product opinion feedback is the feedback of users on the product usage experience, which may include suggestions, complaints, or praises, and is an important source of information for product improvement. The emotional tendency recognition model is a data analysis model used to automatically identify and classify the emotional tendency in text data, such as positive, negative, or neutral. Emotional words are the words that express specific emotions in the text, such as "like" and "disappointed". Emotional intensity is the degree of intensity of the emotion expressed by the emotional word, which can be weak, medium, or strong. The emotional intensity value is a numerical value used to quantify the degree of intensity of the emotion in the text data. The tendency topic is the main concern or discussion focus reflected in the text data, representing the main interests or attention areas of users.

[0049] Among them, the emotional description information that reflects the emotional tendency of users towards the target product is, for example Figure 2 as shown. The user ID is the unique identifier of the user participating in the interaction. The interaction content is the specific content of the user interaction, usually in text form, reflecting the views and feelings of the user towards the target product. The emotional intensity value is the numerical value obtained after performing emotional analysis on the user interaction content, indicating the degree of intensity of the emotion, and the higher the value, the stronger the emotion. The tendency topic is the main concern or discussion focus reflected in the user interaction content, representing the user's attention to specific aspects of the product.

[0050] In the embodiments of the present application, user interaction data is collected and analyzed from new media platforms to deeply understand the emotional tendency and focus of users on products. Data preprocessing ensures the quality of the data, and the emotional tendency recognition model and topic analysis provide in-depth insights into users' emotions and interests.

[0051] S103, input the multi-source emotional description information into a preset new media data online analysis component, and output the user emotional fusion feature corresponding to the target product;

[0052] Among them, the multi-source emotional description information includes the emotional intensity value and the tendency theme of the user interaction data of each new media platform, and the tendency theme is used to reflect the focus of users or the focus of user discussions.

[0053] For example Figure 3 As shown, the preset new media data online analysis component includes an emotional intensity value mapping module, a parameter acquisition module, a weight allocation module, a comprehensive emotional intensity value calculation module, and a user emotional fusion feature generation module.

[0054] In some embodiments of the present application, the specific process of inputting the multi-source emotional description information into the preset new media data online analysis component and outputting the user emotional fusion feature corresponding to the target product includes: the emotional intensity value mapping module standardizes the emotional intensity value to map the emotional intensity value to a unified scoring range to obtain the standardized emotional intensity of each new media platform; the parameter acquisition module acquires the user base, influence quantification parameter of each new media platform, and the comprehensive attention of the target product on each platform; the weight allocation module allocates weights to each new media platform according to the user base, influence quantification parameter, and comprehensive attention; the comprehensive emotional intensity value calculation module calculates the comprehensive emotional intensity value of each new media platform based on the weight allocated to each new media platform and the standardized emotional intensity of each new media platform; the user emotional fusion feature generation module generates the user emotional fusion feature corresponding to the target product according to the comprehensive emotional intensity value of each new media platform and the tendency theme of each new media platform.

[0055] Among them, the emotional intensity value refers to the value obtained through sentiment analysis, indicating the intensity of the user's emotion, usually ranging from negative values (very negative) to positive values (very positive). Standardization refers to the process of converting data into a standard range, such as 0 to 1 or -1 to 1, for facilitating comparison between different data sets. The user base refers to the number of users who follow or use the target product on a specific new media platform. The influence power quantification parameter refers to an indicator for measuring the influence of a new media platform, such as user activity, content sharing rate, etc. The comprehensive attention degree refers to the overall attention degree that the target product receives on a specific new media platform, including the interaction frequency and intensity of users. Weight allocation refers to allocating a weight reflecting its importance to each new media platform according to the user base, influence power quantification parameter, and comprehensive attention degree. The comprehensive emotional intensity value refers to the standardized emotional intensity value after considering the weight, reflecting the weighted average of the users' emotions on each platform. The user emotional fusion feature refers to the comprehensive feature that synthesizes the users' emotions on all new media platforms, used to reflect the overall emotional tendency of users towards the target product.

[0056] For example, perform standardization processing on the emotional intensity values and map them to a unified scoring range of 0 to 1. Allocate weights to each platform according to the user base, influence power quantification parameter, and comprehensive attention degree. For example, if Weibo has a large user base and high activity, it may be allocated a higher weight. Based on the weights allocated to each platform and the standardized emotional intensity values, calculate the comprehensive emotional intensity value of each platform. For example, if the standardized emotional intensity value of Weibo is 0.7 and the weight is 0.5, then its comprehensive emotional intensity value is 0.35. Generate the user emotional fusion feature corresponding to the target product according to the comprehensive emotional intensity value of each platform and the tendency themes (such as camera quality, battery life, etc.). This feature can be a comprehensive score or a classification label, reflecting the overall emotional tendency of users towards the smartphone model.

[0057] In some embodiments of the present application, the specific process of generating the user emotional fusion feature corresponding to the target product according to the comprehensive emotional intensity value of each new media platform and the tendency theme of each new media platform includes: associating the comprehensive emotional intensity value of each new media platform with each tendency theme to obtain multiple comprehensive emotional intensity values for each tendency theme; summing up the multiple comprehensive emotional intensity values of each tendency theme to obtain the overall score of each tendency theme; performing feature conversion on the tendency themes whose overall scores are greater than the preset threshold to obtain a tendency theme feature sequence; splicing the tendency theme feature sequence into a whole as the user emotional fusion feature corresponding to the target product.

[0058] For example, if the target product is a smart watch, at this time, the multiple comprehensive sentiment intensity values of each tendency theme are summed to obtain the overall score of each tendency theme. The tendency themes with the overall score greater than the preset threshold (e.g., 0.5) are subjected to feature transformation to obtain a tendency theme feature sequence. The tendency theme feature sequences are concatenated into a whole as the user emotion fusion feature corresponding to the smart watch.

[0059] In some embodiments, the comprehensive sentiment intensity value of each new media platform and the data table of each tendency theme are, for example Figure 4 As shown, the target product is a smart watch. At this time, the battery life is 0.6 (Weibo) + 0.5 (YouTube) = 1.1, the design aesthetics is 0.8 (Douyin), and the functional practicality is 0.7 (TikTok). All the overall scores are greater than the preset threshold of 0.5. Therefore, all tendency themes are converted into features. At this time, the tendency theme feature sequence is [battery life (1.1), design aesthetics (0.8), functional practicality (0.7)].

[0060] S104. According to the user emotion fusion feature and the pre-established feature knowledge base, mine the user's behavior pattern information and preference information for the target product within a preset period;

[0061] Among them, the user emotion fusion feature is the tendency theme feature sequence, and the pre-established feature knowledge base includes the association relationship between the original features and the behavior pattern information and preference information.

[0062] In some embodiments of the present application, the specific process of mining the user's behavior pattern information and preference information for the target product within a preset period according to the user emotion fusion feature and the pre-established feature knowledge base is as follows: calculate the feature similarity between the tendency theme feature sequence and the original features; use the original features with the similarity greater than the preset similarity threshold as the target features; obtain the pattern information and preference information corresponding to the target features from the association relationship; use the pattern information and preference information corresponding to the target features as the user's behavior pattern information and preference information for the target product within a preset period.

[0063] For example, the feature similarity between the tendency theme feature sequence and the original features is calculated by the cosine similarity calculation method, and the original features with the similarity greater than the preset similarity threshold (e.g., 0.5) are used as the target features. If the similarity of "energy-saving effect" is 0.6, then "energy-saving effect" is the target feature. The pattern information and preference information corresponding to the target feature are obtained from the pre-established feature knowledge base. If "energy-saving effect" is the target feature, the knowledge base contains the information that users have a higher willingness to purchase products with significant energy-saving effects, which can be used as the behavior pattern information and preference information.

[0064] In some embodiments of the present application, the specific process of generating a pre - established feature knowledge base includes: collecting original user interaction data from multiple new media platforms; using natural language processing technology to analyze the original user interaction data to extract emotional features and tendency theme features in the original user interaction data. The emotional features include positive emotional features, negative emotional features, and neutral emotional features. The positive emotional features are used to represent the user's optimistic mood, the negative emotional features are used to represent the user's negative mood, the neutral emotional features are used to represent that the user has no emotion, and the tendency theme features are used to reflect the user's discussion focus or concern; according to the emotional features, analyzing the mode information and preference information of the user for the product at the current moment; storing the tendency theme features and the mode information and preference information of the user for different products to obtain a pre - established feature knowledge base. The pre - established feature knowledge base is, for example Figure 5 as shown

[0065] S105, in the database of the BI system, store the ternary mapping relationship between the target product, the behavior mode information, and the preference information

[0066] In some embodiments of the present application, when the target product is determined to be a newly released smart watch, after obtaining the behavior mode information and preference information of the user for the target product within a preset period based on the pre - established feature knowledge base, a ternary mapping relationship can be established for the target product (smart watch), the behavior mode information (purchase tendency), and the preference information (long battery life, beautiful design). The ternary mapping relationship is, for example Figure 6 as shown

[0067] In the embodiments of the present application, on the one hand, data collection from multiple platforms is carried out through a preset distributed data collection architecture. Since this architecture includes multiple data collection channels, and each channel is associated with the data collection interfaces of each new media platform one by one, simultaneous cross - platform data collection can be achieved, improving the coverage and diversity of data. On the other hand, extract emotional description information from the user interaction data of each new media platform and input it into a preset new media data online analysis component. Combine it with the pre - established feature knowledge base to mine the behavior mode information and preference information. This emotional description information can comprehensively represent the emotional tendency of the user, and this preset new media data online analysis component can process multi - modal data, effectively integrating the complex interaction relationships existing between different types of data, extracting effective emotional features, so as to cooperate with the pre - established feature knowledge base to deeply discover the user behavior mode and preference, and further provide guiding data for product development and content creation at the business layer

[0068] Please refer to Figure 7 , which is a schematic flowchart of a method for training an emotional tendency recognition model provided by an embodiment of the present application. As Figure 7 shown, the method of the embodiment of the present application may include the following steps

[0069] S201, Collect historical video comments, historical live comments, and historical product feedback covering different fields and sentiment tendencies to obtain historical text data;

[0070] S202, Perform sentiment annotation on the collected historical text data to classify the historical text data into positive sentiment data, negative sentiment data, and neutral sentiment data; Positive sentiment data is data labeled to represent users' optimistic emotions, negative sentiment data is data labeled to represent users' negative emotions, and neutral sentiment data is data labeled to represent users' lack of emotions;

[0071] In some embodiments, identify and mark the sentiment tendency of each comment. Positive sentiment may be represented by "1", negative sentiment by "-1", and neutral sentiment by "0" to obtain positive sentiment data, negative sentiment data, and neutral sentiment data.

[0072] S203, Create a sentiment tendency recognition model;

[0073] For example Figure 8 As shown, the sentiment tendency recognition model includes a word embedding network, a sequence embedding layer, a position embedding layer, a self-attention layer, a feature fusion module, and a loss function.

[0074] S204, Input the positive sentiment data, negative sentiment data, and neutral sentiment data into the sentiment tendency recognition model and output the loss value of the model;

[0075] In some embodiments of the present application, the specific process of inputting the positive sentiment data, negative sentiment data, and neutral sentiment data into the sentiment tendency recognition model and outputting the loss value of the model includes: performing data cleaning and text tokenization on the positive sentiment data, negative sentiment data, and neutral sentiment data to obtain tokenized text data; inputting the tokenized text data into the word embedding network to capture the semantic relationships and context information between words for word conversion and obtain word vectors in the text sequence; inputting the word vectors in the text sequence into the sequence embedding layer to capture the sequential information and context relationships in the text sequence and obtain sequence vectors; obtaining the position information of the tokenized text data and inputting the obtained position information of the text data into the position embedding layer to capture the position information of the words in the text and obtain position vectors; inputting the sequence vectors and position vectors into the self-attention layer to capture the mutual relationships between different words and obtain global features; inputting the word vectors, sequence vectors, position vectors, and global features into the feature fusion module to integrate feature information at different levels and obtain a comprehensive feature vector; and outputting the loss value of the model according to the comprehensive feature vector and the loss function.

[0076] Specifically, the specific process of outputting the loss value of the model based on the comprehensive feature vector and the loss function includes: performing a linear transformation on the comprehensive feature vector to obtain a score vector for each sentiment category; converting the score vector for each sentiment category into a probability distribution to obtain the predicted sentiment category probability distribution; inputting the predicted sentiment category probability distribution and the labeled sentiment label corresponding to the predicted sentiment category probability distribution into the loss function to output the loss value of the model;

[0077] Among them, the calculation formula for the score vector is:

[0078]

[0079] Among them, is the score vector, is the weight matrix, is the comprehensive feature vector, is the bias vector;

[0080] Among them, the calculation formula for the probability distribution is:

[0081]

[0082] Among them, is the probability distribution of the th sentiment category, is the score vector of the th sentiment category, is the total number of sentiment categories, is the sum of the score vectors of all sentiment categories;

[0083] Among them, the loss function is:

[0084]

[0085] Among them, is the forward propagation result of the sentiment recognition model at time step , is the model update parameter, is that the optimization objective is to maximize the loss value of the parameter set , is the predicted sentiment category probability distribution, is the labeled sentiment label corresponding to the predicted sentiment category probability distribution, is the probability distribution of all predicted sentiment categories, represents the quantization value of the labeled sentiment label , is the logarithmic probability, is the original parameter of the model, is the model parameter after update, is the time step label

[0086] S205: When the loss value reaches the minimum, generate a pre-trained sentiment tendency recognition model; or, when the loss value does not reach the minimum, forward propagate the model loss value to update the model update parameters of the sentiment tendency recognition model, and continue to execute the step of inputting positive sentiment data, negative sentiment data, and neutral sentiment data into the sentiment tendency recognition model until the loss value reaches the minimum.

[0087] In the embodiments of the present application, on the one hand, data collection from multiple platforms is performed through a preset distributed data collection architecture. Since this architecture includes multiple data collection channels, each channel is associated one by one with the data collection interfaces of each new media platform, so that cross-platform data collection can be achieved simultaneously, improving the coverage and diversity of data. On the other hand, extract sentiment description information from the user interaction data of each new media platform and input it into a preset new media data online analysis component, and combine the pre-established feature knowledge base to mine behavior pattern information and preference information. This sentiment description information can comprehensively represent the user's sentiment tendency, and this preset new media data online analysis component can process multi-modal data, effectively fuse the complex interaction relationships existing between different types of data, and extract effective sentiment features, so as to cooperate with the pre-established feature knowledge base to deeply discover user behavior patterns and preferences, and then provide guiding data for product development and content creation at the business layer.

[0088] The following is an embodiment of the apparatus of the present application, which can be used to execute the method embodiment of the present application. For details not disclosed in the embodiment of the apparatus of the present application, please refer to the method embodiment of the present application.

[0089] Please refer to Figure 9 , which shows a schematic structural diagram of a new media data online analysis device provided by an exemplary embodiment of the present application. This new media data online analysis device can be implemented as all or part of an electronic device through software, hardware, or a combination of both. The device 1 includes a data collection module 10, a sentiment description information extraction module 20, a data input module 30, a data output module 40, and a ternary mapping relationship storage module 50.

[0090] The data collection module 10 is used to simultaneously collect user interaction data about the target product from multiple new media platforms through a preset distributed data collection architecture when the duration between the current moment and the previous data analysis moment is equal to a preset period. The preset distributed data collection architecture includes multiple data collection channels, and each data collection channel is associated one by one with the data collection interfaces provided by each new media platform;

[0091] The emotional description information extraction module 20 is used to extract emotional description information that can reflect the emotional tendency of users towards the target product from the user interaction data of each new media platform, so as to obtain multi-source emotional description information;

[0092] The data input module 30 is used to input the multi-source emotional description information into a preset new media data online analysis component and output the user emotional fusion features corresponding to the target product;

[0093] The data output module 40 is used to mine the behavior pattern information and preference information of users towards the target product within a preset period according to the user emotional fusion features and a pre-established feature knowledge base;

[0094] The ternary mapping relationship storage module 50 is used to store the ternary mapping relationship between the target product, the behavior pattern information, and the preference information in the database of the BI system.

[0095] It should be noted that when the new media data online analysis device provided in the above embodiments executes the new media data online analysis method, only the above-mentioned division of each functional module is used for illustration. In practical applications, the above functions can be allocated to different functional modules according to needs, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the new media data online analysis device provided in the above embodiments and the embodiments of the new media data online analysis method belong to the same concept. The implementation process is detailed in the method embodiments and will not be repeated here.

[0096] The serial numbers of the embodiments of the present application above are only for description and do not represent the advantages and disadvantages of the embodiments.

[0097] In the embodiments of the present application, on the one hand, data collection from multiple platforms is performed through a preset distributed data collection architecture. Since this architecture includes multiple data collection channels, each channel is associated with the data collection interfaces of each new media platform one by one, so that cross-platform data collection can be realized simultaneously, improving the coverage and diversity of data. On the other hand, emotional description information is extracted from the user interaction data of each new media platform and input into a preset new media data online analysis component, and behavior pattern information and preference information are mined in combination with a pre-established feature knowledge base. This emotional description information can comprehensively represent the emotional tendency of users, and this preset new media data online analysis component can process multi-modal data, effectively fuse the complex interaction relationships existing between different types of data, and extract effective emotional features, so as to cooperate with the pre-established feature knowledge base to deeply discover user behavior patterns and preferences, and further provide guiding data for product development and content creation at the business layer.

[0098] The present application also provides a computer-readable medium, on which program instructions are stored. When the program instructions are executed by a processor, the online analysis method for new media data provided by each of the above method embodiments is implemented.

[0099] The present application also provides a computer program product containing instructions. When it runs on a computer, it causes the computer to execute the online analysis method for new media data provided by each of the above method embodiments.

[0100] Please refer to Figure 10 , which is a schematic structural diagram of an electronic device provided by an embodiment of the present application. As Figure 10 shown, the electronic device 1000 may include: at least one processor 1001, at least one network interface 1004, a user interface 1003, a memory 1005, and at least one communication bus 1002.

[0101] Among them, the communication bus 1002 is used to realize the connection and communication between these components.

[0102] Among them, the user interface 1003 may include a display screen (Display) and a camera (Camera). Optionally, the user interface 1003 may further include a standard wired interface and a wireless interface.

[0103] Among them, the network interface 1004 may optionally include a standard wired interface and a wireless interface (such as a WI-FI interface).

[0104] Among them, the processor 1001 may include one or more processing cores. The processor 1001 connects various parts within the entire electronic device 1000 through various interfaces and lines. By running or executing instructions, programs, code sets, or instruction sets stored in the memory 1005, and by invoking the data stored in the memory 1005, it executes various functions of the electronic device 1000 and processes data. Optionally, the processor 1001 may be implemented in at least one hardware form of digital signal processing (DSP), field-programmable gate array (FPGA), or programmable logic array (PLA). The processor 1001 may integrate one or a combination of several of a central processing unit (CPU), a graphics processing unit (GPU), and a modem, etc. Among them, the CPU mainly processes the operating system, user interface, application programs, etc.; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communication. It can be understood that the above-mentioned modem may not be integrated into the processor 1001 and may be implemented separately by a single chip.

[0105] Among them, the memory 1005 may include random access memory (RAM) and may also include read-only memory. Optionally, the memory 1005 includes a non-transitory computer-readable storage medium. The memory 1005 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 1005 may include a program storage area and a data storage area. Among them, the program storage area may store instructions for implementing the operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-mentioned various method embodiments, etc.; the data storage area may store the data involved in the above-mentioned various method embodiments. Optionally, the memory 1005 may also be at least one storage system located far from the aforementioned processor 1001. As Figure 10 shown, the memory 1005, as a computer storage medium, may include an operating system, a network communication module, a user interface module, and a new media data online analysis application program.

[0106] In Figure 10In the electronic device 1000 shown, the user interface 1003 is mainly used to provide an interface for the user to input and obtain the data input by the user; while the processor 1001 can be used to call the new media data online analysis application program stored in the memory 1005 and specifically perform the following operations:

[0107] When the duration between the current moment and the last data analysis moment is equal to the preset period, through the preset distributed data acquisition architecture, simultaneously collect the user interaction data about the target product from multiple new media platforms. The preset distributed data acquisition architecture includes multiple data acquisition channels, and each data acquisition channel is associated with the data acquisition interface provided by each new media platform;

[0108] Extract the emotional description information that can reflect the user's emotional tendency towards the target product from the user interaction data of each new media platform to obtain multi-source emotional description information;

[0109] Input the multi-source emotional description information into the preset new media data online analysis component and output the user emotional fusion characteristics corresponding to the target product;

[0110] According to the user emotional fusion characteristics and the pre-established feature knowledge base, mine the user's behavior pattern information and preference information for the target product within the preset period;

[0111] In the database of the BI system, store the ternary mapping relationship between the target product, the behavior pattern information, and the preference information.

[0112] In one embodiment, when the processor 1001 executes inputting the multi-source emotional description information into the preset new media data online analysis component and outputting the user emotional fusion characteristics corresponding to the target product, it specifically performs the following operations:

[0113] The emotional intensity value mapping module standardizes the emotional intensity value to map the emotional intensity value to a unified scoring range to obtain the standardized emotional intensity of each new media platform;

[0114] The parameter acquisition module acquires the user base, influence quantification parameter, and the comprehensive attention of the target product on each platform of each new media platform;

[0115] The weight allocation module allocates weights to each new media platform according to the user base, influence quantification parameter, and comprehensive attention;

[0116] The comprehensive emotional intensity value calculation module calculates the comprehensive emotional intensity value of each new media platform based on the weights allocated to each new media platform and the standardized emotional intensity of each new media platform;

[0117] The user emotion fusion feature generation module generates user emotion fusion features corresponding to the target product according to the comprehensive emotion intensity value of each new media platform and the tendency theme of each new media platform.

[0118] In one embodiment, when the processor 1001 executes to generate user emotion fusion features corresponding to the target product according to the comprehensive emotion intensity value of each new media platform and the tendency theme of each new media platform, the following operations are specifically performed:

[0119] Associate the comprehensive emotion intensity value of each new media platform with each tendency theme to obtain multiple comprehensive emotion intensity values for each tendency theme;

[0120] Sum the multiple comprehensive emotion intensity values of each tendency theme to obtain the overall score of each tendency theme;

[0121] Perform feature conversion on the tendency themes with the overall score greater than the preset threshold to obtain a tendency theme feature sequence;

[0122] Concatenate the tendency theme feature sequences into a whole as the user emotion fusion features corresponding to the target product.

[0123] In one embodiment, when the processor 1001 executes to mine the behavior pattern information and preference information of the user for the target product within a preset period according to the user emotion fusion features and the pre-established feature knowledge base, the following operations are specifically performed:

[0124] Calculate the feature similarity between the tendency theme feature sequence and the original features;

[0125] Take the original features with the similarity greater than the preset similarity threshold as the target features;

[0126] Obtain the pattern information and preference information corresponding to the target features from the association relationship;

[0127] Take the pattern information and preference information corresponding to the target features as the behavior pattern information and preference information of the user for the target product within a preset period.

[0128] In one embodiment, when the processor 1001 executes to generate the pre-established feature knowledge base, the following operations are specifically performed:

[0129] Collect original user interaction data from multiple new media platforms;

[0130] Analyze the original user interaction data using natural language processing techniques to extract the emotional features and tendency theme features in the original user interaction data. The emotional features include positive emotional features, negative emotional features, and neutral emotional features. Positive emotional features are used to represent users' optimistic emotions, negative emotional features are used to represent users' negative emotions, neutral emotional features are used to represent users' lack of emotions, and tendency theme features are used to reflect users' discussion focuses or concerns.

[0131] According to the emotional features, analyze the mode information and preference information of users for the product at the current moment.

[0132] Store the tendency theme features and the mode information and preference information of users for different products to obtain a pre-established feature knowledge base.

[0133] In one embodiment, when the processor 1001 executes to extract the emotional description information that can reflect the emotional tendency of users towards the target product from the user interaction data of each new media platform, it specifically performs the following operations:

[0134] Perform data preprocessing on the video comments, live comments, and product opinion feedback included in the user interaction data of each new media platform to obtain the vocabulary sequence and phrase sequence of each new media platform. The preprocessing includes data cleaning, data deduplication, data formatting, and data tokenization.

[0135] Input the vocabulary sequence and phrase sequence of each new media platform into a pre-trained emotional tendency recognition model, and output multiple emotional words corresponding to the user interaction data of each new media platform and the intensity of each emotional word.

[0136] Calculate the emotional intensity value of the user interaction data of each new media platform according to the intensity and quantity of each emotional word.

[0137] Perform topic analysis on the vocabulary sequence and phrase sequence of each new media platform to extract the tendency themes involved in the user interaction data of each new media platform.

[0138] Use the emotional intensity value and tendency theme of the user interaction data of each new media platform as the emotional description information that can reflect the emotional tendency of users towards the target product.

[0139] In one embodiment, when the processor 1001 executes to generate a pre-trained emotional tendency recognition model, it specifically performs the following operations:

[0140] Collect historical video comments, historical live comments, and historical product opinion feedback covering different fields and emotional tendencies to obtain historical text data.

[0141] Perform sentiment annotation on the collected historical text data to classify the historical text data into positive sentiment data, negative sentiment data, and neutral sentiment data; positive sentiment data is data labeled to represent the user's optimistic mood, negative sentiment data is data labeled to represent the user's negative mood, and neutral sentiment data is data labeled to represent the user's lack of mood;

[0142] Create a sentiment tendency recognition model;

[0143] Input the positive sentiment data, negative sentiment data, and neutral sentiment data into the sentiment tendency recognition model, and output the loss value of the model;

[0144] When the loss value reaches the minimum, generate a pre-trained sentiment tendency recognition model; or, when the loss value does not reach the minimum, forward propagate the model loss value to update the model update parameters of the sentiment tendency recognition model, and continue to execute the step of inputting the positive sentiment data, negative sentiment data, and neutral sentiment data into the sentiment tendency recognition model until the loss value reaches the minimum.

[0145] In one embodiment, when the processor 1001 executes inputting the positive sentiment data, negative sentiment data, and neutral sentiment data into the sentiment tendency recognition model and outputting the loss value of the model, it specifically performs the following operations:

[0146] Perform data cleaning and text tokenization on the positive sentiment data, negative sentiment data, and neutral sentiment data to obtain the tokenized text data;

[0147] Input the tokenized text data into the word embedding network to capture the semantic relationships and context information between words for word conversion, and obtain the word vectors in the text sequence;

[0148] Input the word vectors in the text sequence into the sequence embedding layer to capture the sequential information and context relationships in the text sequence, and obtain the sequence vectors;

[0149] Obtain the position information of the tokenized text data, and input the obtained position information of the text data into the position embedding layer to capture the position information of the words in the text, and obtain the position vectors;

[0150] Input the sequence vectors and position vectors into the self-attention layer to capture the mutual relationships between different words, and obtain the global features;

[0151] Input the word vectors, sequence vectors, position vectors, and global features into the feature fusion module to integrate the feature information at different levels, and obtain the comprehensive feature vectors;

[0152] According to the comprehensive feature vectors and the loss function, output the loss value of the model.

[0153] In one embodiment, when the processor 1001 executes to output the loss value of the model according to the comprehensive feature vector and the loss function, the following operations are specifically performed:

[0154] Perform a linear transformation on the comprehensive feature vector to obtain a score vector for each sentiment category;

[0155] Convert the score vector for each sentiment category into a probability distribution to obtain the predicted sentiment category probability distribution;

[0156] Input the predicted sentiment category probability distribution and the labeled sentiment label corresponding to the predicted sentiment category probability distribution into the loss function to output the loss value of the model.

[0157] In the embodiments of the present application, on the one hand, data collection from multiple platforms is performed through a preset distributed data collection architecture. Since this architecture includes multiple data collection channels, and each channel is associated with the data collection interfaces of each new media platform one by one, simultaneous cross-platform data collection can be achieved, improving the coverage and diversity of data. On the other hand, emotional description information is extracted from the user interaction data of each new media platform and input into a preset new media data online analysis component. In combination with a pre-established feature knowledge base, behavior pattern information and preference information are mined. This emotional description information can comprehensively represent the emotional tendency of users, and this preset new media data online analysis component can process multi-modal data, effectively integrating the complex interaction relationships existing between different types of data, extracting effective emotional features, so as to cooperate with the pre-established feature knowledge base to deeply discover user behavior patterns and preferences, and further provide guiding data for product development and content creation at the business layer.

[0158] Those of ordinary skill in the art can understand that all or part of the processes of implementing the methods in the above embodiments can be completed by instructing relevant hardware through a computer program. The program for new media data online analysis can be stored in a computer-readable storage medium. When this program is executed, it can include the processes of the embodiments of the above methods. Among them, the storage medium of the program for new media data online analysis can be a magnetic disk, an optical disc, a read-only memory, or a random access memory, etc.

[0159] The above-disclosed are only the preferred embodiments of the present application. Of course, the scope of the rights of the present application cannot be limited thereby. Therefore, equivalent changes made according to the claims of the present application still fall within the scope covered by the present application.

Claims

1. A new media data online analysis method, characterized in that: Applied to the server side of the BI system, the method includes: When the time between the current moment and the last data analysis moment is equal to the preset period, user interaction data on the target product is collected from multiple new media platforms simultaneously through a preset distributed data collection architecture, wherein the preset distributed data collection architecture includes multiple data collection channels, each data collection channel is associated with a data collection interface provided by each new media platform; Extracting emotional description information that can reflect the user's emotional inclination towards the target product from the user interaction data of each new media platform to obtain multi-source emotional description information; the multi-source emotional description information includes the emotional intensity value and inclination theme of the user interaction data of each new media platform, and the inclination theme is used to reflect the user's focus or the focus of user discussion; Input the multi-source emotion description information into a preset new media data online analysis component, and output the user emotion fusion feature corresponding to the target product; the preset new media data online analysis component includes an emotion intensity value mapping module, a parameter acquisition module, a weight allocation module, a comprehensive emotion intensity value calculation module, and a user emotion fusion feature generation module; The step of inputting the multi-source emotion description information into a preset new media data online analysis component and outputting the user emotion fusion features corresponding to the target product includes: The emotion intensity value mapping module performs standardization processing on the emotion intensity value to map the emotion intensity value to a unified scoring range to obtain the standardized emotion intensity of each new media platform; The parameter acquisition module acquires the user base, influence quantification parameters and comprehensive attention of the target product on each platform of each new media platform; The weight allocation module allocates a weight to each new media platform according to the user base, the influence quantification parameter and the comprehensive attention; The comprehensive emotion intensity value calculation module calculates the comprehensive emotion intensity value of each new media platform based on the weight assigned to each new media platform and the standardized emotion intensity of each new media platform; The user emotion fusion feature generation module generates a user emotion fusion feature corresponding to the target product according to the comprehensive emotion intensity value of each new media platform and the tendency theme of each new media platform; Mining the user's behavior pattern information and preference information for the target product within a preset period based on the user's emotion fusion features and a pre-established feature knowledge base; In the database of the BI system, a ternary mapping relationship between the target product and the behavior pattern information and the preference information is stored.

2. The method according to claim 1, characterized in that Generating the user emotion fusion feature corresponding to the target product according to the comprehensive emotion intensity value of each new media platform and the tendency theme of each new media platform includes: Associating the comprehensive emotion intensity value of each new media platform with each tendency topic to obtain multiple comprehensive emotion intensity values ​​of each tendency topic; Sum up the multiple comprehensive sentiment intensity values ​​of each tendency topic to obtain the overall score of each tendency topic; Performing feature conversion on the tendency topics whose overall scores are greater than a preset threshold to obtain a tendency topic feature sequence; The tendency theme feature sequences are spliced ​​into a whole as the user emotion fusion feature corresponding to the target product.

3. The method according to claim 1, characterized in that The user emotion fusion feature is a tendency theme feature sequence, and the pre-established feature knowledge base includes the association relationship between the original feature and the behavior pattern information and the preference information; The mining of the user's behavior pattern information and preference information for the target product within a preset period based on the user's emotion fusion feature and a pre-established feature knowledge base includes: Calculating the feature similarity between the tendency topic feature sequence and the original feature; The original features whose similarity is greater than a preset similarity threshold are taken as target features; From the association relationship, obtain the mode information and preference information corresponding to the target feature; The pattern information and preference information corresponding to the target feature are used as the user's behavior pattern information and preference information for the target product within a preset period.

4. The method according to claim 3, generating a pre-established feature knowledge base according to the following steps, comprising: Collect raw user interaction data from multiple new media platforms; Analyzing the original user interaction data using natural language processing technology to extract emotional features and tendency theme features in the original user interaction data, wherein the emotional features include positive emotional features, negative emotional features, and neutral emotional features; The positive emotion feature is used to characterize the user's optimistic emotion, the negative emotion feature is used to characterize the user's negative emotion, the neutral emotion feature is used to characterize the user's lack of emotion, and the tendency topic feature is used to reflect the user's discussion focus or concern; Analyze the user's current mode information and preference information for the product based on the emotional characteristics; The tendency theme features and the user's pattern information and preference information for different products are stored to obtain a pre-established feature knowledge base.

5. The method according to claim 1, characterized in that The user interaction data of each new media platform includes video comments, live broadcast comments, and product feedback; Extracting emotional description information that can reflect the user's emotional inclination toward the target product from the user interaction data of each new media platform includes: Preprocessing the user interaction data of each new media platform, including video comments, live broadcast comments, and product feedback, to obtain a vocabulary sequence and a phrase sequence of each new media platform; preprocessing includes data cleaning, data deduplication, data formatting, and data segmentation; Inputting the vocabulary sequence and phrase sequence of each new media platform into a pre-trained sentiment tendency recognition model, and outputting a plurality of sentiment words corresponding to the user interaction data of each new media platform and the strength of each sentiment word; Calculating the emotion intensity value of the user interaction data of each new media platform according to the intensity of each emotion vocabulary and the number of emotion vocabulary; Performing topic analysis on the word sequence and phrase sequence of each new media platform to extract the tendency topics involved in the user interaction data of each new media platform; The emotional intensity value of the user interaction data of each new media platform and the tendency theme are used as emotional description information that can reflect the user's emotional tendency towards the target product.

6. According to the method of claim 5, the pre-trained emotion tendency recognition model is generated according to the following steps, including: Collect historical video comments, historical live broadcast comments, and historical product feedback covering different fields and sentiment tendencies to obtain historical text data; Perform sentiment annotation on the collected historical text data to classify the historical text data into positive sentiment data, negative sentiment data and neutral sentiment data; The positive emotion data is data labeled to represent the user's optimistic emotion, the negative emotion data is data labeled to represent the user's negative emotion, and the neutral emotion data is data labeled to represent the user's lack of emotion; Create a sentiment tendency recognition model; Inputting the positive emotion data, negative emotion data and neutral emotion data into the emotion tendency recognition model, and outputting the loss value of the model; When the loss value reaches the minimum, a pre-trained emotion tendency recognition model is generated; or, when the loss value does not reach the minimum, the model loss value is forward propagated to update the model update parameters of the emotion tendency recognition model, and the step of inputting the positive emotion data, negative emotion data and neutral emotion data into the emotion tendency recognition model is continued until the loss value reaches the minimum.

7. The method according to claim 6, characterized in that The sentiment tendency recognition model includes a word embedding network, a sequence embedding layer, a position embedding layer, a self-attention layer, a feature fusion module and a loss function; The step of inputting the positive emotion data, the negative emotion data and the neutral emotion data into the emotion tendency recognition model and outputting the loss value of the model comprises: Performing data cleaning and text segmentation on the positive sentiment data, the negative sentiment data, and the neutral sentiment data to obtain text data after segmentation; Inputting the segmented text data into the word embedding network to capture the semantic relationship and context information between words for word conversion, and obtaining word vectors in the text sequence; Inputting the word vectors in the text sequence into the sequence embedding layer to capture the order information and contextual relationship in the text sequence to obtain a sequence vector; Acquire the position information of the text data after word segmentation, and input the acquired position information of the text data into the position embedding layer to capture the position information of the vocabulary in the text and obtain the position vector; Inputting the sequence vector and the position vector into the self-attention layer to capture the relationship between different words and obtain a global feature; Inputting the word vector, sequence vector, position vector and global feature into the feature fusion module to integrate feature information at different levels to obtain a comprehensive feature vector; According to the comprehensive feature vector and the loss function, the loss value of the model is output.

8. The method according to claim 7, characterized in that Outputting the loss value of the model according to the comprehensive feature vector and the loss function includes: Performing a linear transformation on the comprehensive feature vector to obtain a score vector for each emotion category; Convert the score vector of each emotion category into a probability distribution to obtain a predicted emotion category probability distribution; The predicted emotion category probability distribution and the annotated emotion label corresponding to the predicted emotion category probability distribution are input into the loss function, and the loss value of the model is output.

9. A new media data online analysis device implemented using the method described in any one of claims 1 to 8, characterized in that: The device comprises: A data collection module, for simultaneously collecting user interaction data on a target product from multiple new media platforms through a preset distributed data collection architecture when the time between the current moment and the last data analysis moment is equal to a preset period, wherein the preset distributed data collection architecture includes multiple data collection channels, and each data collection channel is associated with a data collection interface provided by each new media platform one by one; An emotion description information extraction module is used to extract emotion description information that can reflect the user's emotional inclination towards the target product from the user interaction data of each new media platform, and obtain multi-source emotion description information; A data input module, used to input the multi-source emotion description information into a preset new media data online analysis component, and output the user emotion fusion features corresponding to the target product; A data output module is used to mine the user's behavior pattern information and preference information for the target product within a preset period according to the user's emotion fusion feature and a pre-established feature knowledge base; The ternary mapping relationship storage module is used to store the ternary mapping relationship between the target product and the behavior pattern information and the preference information in the database of the BI system.

Images