Intelligent public opinion analysis method and device, terminal equipment and storage medium

Abstract

The application discloses an intelligent public opinion analysis method and device, terminal equipment and a storage medium. The method comprises the following steps: collecting user video comment data; generating a concerned video image based on the collected user video comment data, and generating an intelligent public opinion preliminary analysis result based on the generated concerned video image; generating comment analysis information based on the collected user video comment data and the intelligent public opinion preliminary analysis result; and performing video public opinion analysis based on the intelligent public opinion preliminary analysis result and the comment analysis information. The intelligent public opinion analysis method based on comment collection has the advantages of lower cost, more comprehensive coverage, higher efficiency and higher accuracy compared with traditional methods.

Description

Technical Field

[0001] This invention relates to the field of mobile communication technology, and in particular to an intelligent public opinion analysis method, apparatus, terminal device, and storage medium. Background Technology

[0002] Currently, the mining, analysis, and processing of public opinion information mainly rely on user surveys and traditional text analysis methods. By analyzing browsing and commenting data on platforms such as questionnaires, Weibo, Baidu Tieba, and Zhihu, we can identify and statistically analyze keywords in comment texts and implement public opinion analysis functions such as event tag classification and hot word search recommendations.

[0003] Existing analysis methods have the following drawbacks:

[0004] User surveys have limited coverage, single dimensions, and are insufficient to reach a large number of users, exhibiting a significant time lag. Traditional text analysis methods, compared to video bullet comments, lack clear temporal characteristics, real-time interactivity, and sharing. Furthermore, bullet comments are shorter, use more internet slang, are less comprehensive, and are strongly correlated with the current video context. Therefore, traditional text analysis methods struggle to accurately capture the characteristics of bullet comment data. In addition, traditional methods offer limited user profiling (typically only gender, approximate age, and provincial / regional distribution), making them unsuitable for precise guidance in public opinion analysis. Summary of the Invention

[0005] The main objective of this invention is to provide an intelligent public opinion analysis method, device, terminal equipment, and storage medium, aiming to achieve accurate analysis and intelligent and efficient processing of social public opinion information, and improve the accuracy of public opinion information.

[0006] To achieve the above objectives, embodiments of the present invention provide an intelligent public opinion analysis method, the method comprising the following steps:

[0007] Collect user video comment data;

[0008] Based on the collected user video bullet screen data, a profile of the videos of interest is generated, and based on the generated profile of the videos of interest, preliminary intelligent public opinion analysis results are generated.

[0009] Based on the collected user video bullet screen data and the preliminary results of the intelligent public opinion analysis, bullet screen analysis information is generated;

[0010] Based on the preliminary results of the intelligent public opinion analysis and the information from the bullet screen analysis, video public opinion analysis was conducted.

[0011] Optionally, the steps of generating a profile of the videos of interest based on the collected user video bullet comment data, and generating preliminary intelligent public opinion analysis results based on the generated profile of the videos of interest, include:

[0012] Based on the collected user video bullet screen data, a video description model is trained for different categories to obtain video description models for different categories.

[0013] Based on the video description model, a video profile model is established;

[0014] Based on the video profiling model, a profile of the video of interest is generated. Based on the generated profile of the video of interest, a preliminary intelligent public opinion analysis result is generated.

[0015] Optionally, the step of generating a profile of the video of interest based on the video profile model, and generating preliminary intelligent public opinion analysis results based on the generated profile of the video of interest, includes:

[0016] Based on the video profiling model, the public opinion monitoring coefficient of the video of interest is obtained;

[0017] Select a predetermined number of videos with the highest public opinion monitoring coefficients as video profiles of interest and generate corresponding video description information by category;

[0018] A list of public opinion monitoring videos is generated based on the corresponding video description information, serving as the preliminary result of intelligent public opinion analysis.

[0019] Optionally, after the step of generating a list of public opinion monitoring videos based on the generated corresponding video description information, the method further includes:

[0020] For each video profile of interest corresponding to the public opinion monitoring video list, compare the corresponding video description information to obtain the similarity correlation of each video;

[0021] For multiple videos with a similarity correlation higher than a set threshold, the video with the highest public opinion monitoring coefficient is retained, the remaining similar videos are deleted, and replacement videos are sequentially selected from the preset number of videos according to the public opinion monitoring coefficient until the public opinion monitoring video list no longer changes, thus obtaining the final public opinion monitoring video list as the preliminary result of intelligent public opinion analysis.

[0022] Optionally, the step of generating bullet screen analysis information based on the collected user video bullet screen data and the preliminary intelligent public opinion analysis results includes:

[0023] For the videos in the aforementioned public opinion monitoring video list, read the corresponding user video comment data;

[0024] Based on the read user video bullet comment data, generate basic attribute analysis results of bullet comments, as well as bullet comment sentiment analysis information;

[0025] Based on the sentiment analysis information of the bullet comments, bullet comments with sentiment scores lower than a preset threshold are selected to generate negative bullet comment user profile results.

[0026] The generated basic attribute analysis results of bullet comments, bullet comment sentiment analysis information, and negative bullet comment user profile results are integrated to generate bullet comment analysis information.

[0027] Optionally, the step of selecting bullet comments with sentiment scores lower than a preset threshold based on the bullet comment sentiment analysis information and generating negative bullet comment user profile results includes:

[0028] Based on the aforementioned bullet screen sentiment analysis information, read video bullet screen data of users with negative bullet screens whose sentiment scores are below a preset threshold;

[0029] The negative bullet comment data of users is read and associated with a pre-established user scenario information database to obtain a bullet comment user location database.

[0030] Based on the aforementioned danmaku user location database, a detailed profile is created for the negative danmaku users of interest identified through analysis and location, generating negative danmaku user profile results.

[0031] Optionally, the step of performing video public opinion analysis based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information includes:

[0032] The priority for handling key public opinion events is calculated based on the list of public opinion monitoring videos and the public opinion monitoring coefficients of the corresponding videos.

[0033] For videos whose public opinion processing priority reaches a preset threshold, a comprehensive intelligent public opinion analysis scheme is formed based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information, and the public opinion analysis results are output and / or pushed to external parties.

[0034] Optionally, the step of performing video public opinion analysis based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information further includes:

[0035] For videos whose public opinion processing priority reaches a preset threshold, the public opinion is presented in a contextualized geographic manner based on a pre-established user scenario information database.

[0036] Optionally, the step of collecting user video bullet screen data includes: automatically collecting user video bullet screen data through web crawlers, public API interfaces or SDK embedding points, and saving it by category.

[0037] Optionally, the step of collecting user video bullet screen data further includes: collecting user XDR data, MDT data, and MR data through platform integration; collecting engineering parameter data; and / or obtaining scene border data through web crawlers or API public interfaces.

[0038] Optionally, the step of collecting user video bullet screen data further includes:

[0039] Processing the collected user video bullet screen data includes: cleaning, merging and associating the collected user video bullet screen data to form a user network information database;

[0040] Based on the user network information database and latitude and longitude matching, and by matching the collected scene border data, a user scene information database containing user network information and geographic information is established.

[0041] Furthermore, this invention also proposes an intelligent public opinion analysis device, which includes:

[0042] The data acquisition module is used to collect user video comment data;

[0043] The list generation module is used to generate a profile of the videos of interest based on the collected user video bullet screen data, and to generate preliminary intelligent public opinion analysis results based on the generated profile of the videos of interest.

[0044] The information generation module is used to generate bullet screen analysis information based on the collected user video bullet screen data and the preliminary intelligent public opinion analysis results;

[0045] The analysis module is used to perform video-based public opinion analysis based on the preliminary results of the intelligent public opinion analysis and the bullet screen analysis information.

[0046] Furthermore, this embodiment of the invention also proposes a terminal device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the intelligent public opinion analysis method described above.

[0047] Furthermore, this embodiment of the invention also proposes a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the intelligent public opinion analysis method described above.

[0048] Furthermore, this embodiment of the invention also proposes a computer program product, which includes a computer program that, when executed by a processor, implements the intelligent public opinion analysis method described above.

[0049] The intelligent public opinion analysis method, device, terminal equipment, and storage medium proposed in this invention collect user video bullet screen data; based on the collected user video bullet screen data, generate a profile of the video of interest; based on the generated profile of the video of interest, generate preliminary intelligent public opinion analysis results; generate bullet screen analysis information based on the collected user video bullet screen data and the preliminary intelligent public opinion analysis results; and perform video public opinion analysis based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information. Therefore, it can automatically collect massive amounts of user video bullet screen data, generate profiles of the video of interest based on the collected user video bullet screen data, and generate preliminary intelligent public opinion analysis results. It can realize analytical functions such as detailed profiling of key users, identification of key hot information, prediction of dissemination and evaluation trends, and early warning of emergencies, as well as GIS presentation of the results. This enables efficient and accurate automatic monitoring and precise intelligent analysis of public opinion, achieving intelligent mining, analysis, and positioning of social public opinion, improving the accuracy of public opinion information, and providing an efficient and accurate public opinion solution for monitoring, tracking, and preventative maintenance of social hot events. The intelligent public opinion analysis method based on bullet screen collection proposed in this invention is lower in cost, more comprehensive in coverage, more efficient, and more accurate than traditional methods. Attached Figure Description

[0050] Figure 1 This is a schematic diagram of the functional modules of the terminal device to which the intelligent public opinion analysis device of this invention belongs;

[0051] Figure 2 This is a flowchart illustrating an exemplary embodiment of the intelligent public opinion analysis method of the present invention;

[0052] Figure 3 This is a schematic diagram illustrating the specific implementation process of intelligent public opinion analysis in an embodiment of the present invention;

[0053] Figure 4 This is a schematic diagram of the data collection process in an embodiment of the intelligent public opinion analysis method of the present invention;

[0054] Figure 5 This is a schematic diagram of the data processing flow in an embodiment of the intelligent public opinion analysis method of the present invention;

[0055] Figure 6 This is a schematic diagram of the process for generating a profile of the video of interest and preliminary intelligent public opinion analysis results in an embodiment of the intelligent public opinion analysis method of the present invention;

[0056] Figure 7 This is a schematic diagram of the process for generating bullet screen analysis information in an embodiment of the intelligent public opinion analysis method of the present invention;

[0057] Figure 8 This is a schematic diagram illustrating the process of intelligent analysis, positioning, and presentation of key public opinion in an embodiment of the intelligent public opinion analysis method of the present invention.

[0058] The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0059] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0060] The main solution of this invention is as follows: First, collect user video bullet comment data. Second, based on the collected user video bullet comment data, generate a profile of the video of interest. Third, based on the generated profile of the video of interest, generate preliminary intelligent public opinion analysis results. Fourth, based on the collected user video bullet comment data and the preliminary intelligent public opinion analysis results, generate bullet comment analysis information. Fifth, perform video public opinion analysis based on the preliminary intelligent public opinion analysis results and the bullet comment analysis information. This allows for the automatic collection of massive amounts of user video bullet comment data, the generation of profiles of the video of interest based on the collected user video bullet comment data, and the generation of preliminary intelligent public opinion analysis results. It enables detailed profiling of key users, identification of key hot topics, prediction of dissemination and evaluation trends, and early warning of emergencies, along with GIS presentation of the results. This achieves efficient and accurate automatic monitoring and precise intelligent analysis of public opinion, enabling intelligent mining, analysis, and positioning of social public opinion, improving the accuracy of public opinion information, and providing an efficient and accurate public opinion solution for monitoring, tracking, and preventative maintenance of social hot topics. The intelligent public opinion analysis method based on bullet comment data collection proposed in this invention is lower in cost, has a wider coverage, higher efficiency, and higher accuracy compared to traditional methods.

[0061] This invention addresses the fact that existing solutions for mining and analyzing public opinion information primarily rely on user surveys and traditional text analysis methods. These methods analyze browsing and commenting on platforms such as questionnaires, Weibo, Baidu Tieba, and Zhihu to identify and statistically analyze comment text keywords, and implement public opinion analysis functions such as event tag classification and hot word search recommendations. However, these methods have several drawbacks: user surveys have limited coverage, a single dimension, and are insufficient to cover a large number of users, exhibiting significant time lag; traditional text analysis methods, compared to video bullet comments, lack clear temporal characteristics, real-time interactivity, and sharing, and bullet comment text is shorter, uses more internet slang, is incomplete, and is strongly correlated with the current video context. Therefore, traditional text analysis methods struggle to accurately address the characteristics of bullet comment data; furthermore, traditional methods offer limited user profiling (generally only including gender, approximate age, and provincial / regional distribution), making it difficult to accurately guide public opinion analysis.

[0062] Therefore, the solutions proposed in this invention can achieve accurate analysis and intelligent and efficient processing of public opinion information, thereby improving the accuracy of public opinion information.

[0063] This solution proposes an intelligent public opinion analysis method based on bullet screen (danmu) data collection, including: 1) collecting bullet screen information and related user information; generating a user scenario information database containing user network usage information and geographic information; 2) training video description models for different video categories, establishing video profile models based on the video description models, which include video profile tags and tag weights for each profile dimension; calculating public opinion monitoring coefficients based on the tag weights for each dimension; selecting videos with high public opinion monitoring coefficients as public opinion monitoring videos and generating corresponding description information for each category, statistically analyzing the most frequent words in the description information of each video as keywords for that video, and storing the video description information and keywords together in the public opinion monitoring video list; comparing the description information of the selected video set, and identifying videos with similarity correlations higher than a set threshold. 1) For each video, only the video with the highest public opinion monitoring coefficient is retained, and the remaining similar videos are deleted. Replacement videos are selected in order according to the public opinion monitoring coefficient until the list of public opinion monitoring videos no longer changes; 2) For the videos in the list of public opinion monitoring videos, the corresponding user video bullet screen data is read and the basic attribute analysis results of the video bullet screen are generated. The basic attribute information of the bullet screen includes the bullet screen keywords that appear most frequently in the video; a bullet screen sentiment value score is generated for the bullet screen in the video. The sentiment value scores of all bullet screens sent by the same user are weighted to obtain the comprehensive sentiment score of the user; bullet screens with sentiment value scores lower than the preset value are selected. Based on the user information of the user corresponding to the selected bullet screen, a negative bullet screen user profile is generated; the generated bullet screen basic attribute analysis results, bullet screen sentiment analysis information, and negative bullet screen user profile information are integrated to generate bullet screen analysis information and stored;

[0064] 4) Combine the list of public opinion monitoring videos and the public opinion monitoring coefficient to determine the priority of handling key public opinion issues; for videos whose public opinion handling priority reaches the preset threshold, conduct public opinion analysis based on the bullet screen analysis information and the list of public opinion monitoring videos, generate corresponding handling plans, and push them to the target audience for public opinion handling.

[0065] Specifically, refer to Figure 1 , Figure 1 This is a schematic diagram of the functional modules of the terminal device to which the intelligent public opinion analysis device of the present invention belongs. This intelligent public opinion analysis device can be an independent device capable of intelligent public opinion analysis, and it can be implemented on the terminal device in the form of hardware or software. The terminal device can be a smart mobile terminal such as a mobile phone or tablet computer, or a network device such as a server.

[0066] In this embodiment, the terminal device to which the intelligent public opinion analysis device belongs includes at least an output module 110, a processor 120, a memory 130, and a communication module 140.

[0067] The memory 130 stores the operating system and intelligent public opinion analysis program; the output module 110 can be a display screen, speaker, etc. The communication module 140 can include a WIFI module, a mobile communication module, and a Bluetooth module, etc., and communicates with external devices or servers through the communication module 140.

[0068] In one embodiment, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it performs the following steps:

[0069] Collect user video comment data;

[0070] Based on the collected user video bullet screen data, a profile of the videos of interest is generated, and based on the generated profile of the videos of interest, preliminary intelligent public opinion analysis results are generated.

[0071] Based on the collected user video bullet screen data and the preliminary results of the intelligent public opinion analysis, bullet screen analysis information is generated;

[0072] Based on the preliminary results of the intelligent public opinion analysis and the information from the bullet screen analysis, video public opinion analysis was conducted.

[0073] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0074] Based on the collected user video bullet screen data, a video description model is trained for different categories to obtain video description models for different categories.

[0075] Based on the video description model, a video profile model is established;

[0076] Based on the video profiling model, a profile of the video of interest is generated. Based on the generated profile of the video of interest, a preliminary intelligent public opinion analysis result is generated.

[0077] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0078] Based on the video profiling model, the public opinion monitoring coefficient of the video of interest is obtained;

[0079] Select a predetermined number of videos with the highest public opinion monitoring coefficients as video profiles of interest and generate corresponding video description information by category;

[0080] A list of public opinion monitoring videos is generated based on the corresponding video description information, serving as the preliminary result of intelligent public opinion analysis.

[0081] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0082] For each video profile of interest corresponding to the public opinion monitoring video list, compare the corresponding video description information to obtain the similarity correlation of each video;

[0083] For multiple videos with a similarity correlation higher than a set threshold, the video with the highest public opinion monitoring coefficient is retained, the remaining similar videos are deleted, and replacement videos are sequentially selected from the preset number of videos according to the public opinion monitoring coefficient until the public opinion monitoring video list no longer changes, thus obtaining the final public opinion monitoring video list as the preliminary result of intelligent public opinion analysis.

[0084] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0085] For the videos in the aforementioned public opinion monitoring video list, read the corresponding user video comment data;

[0086] Based on the read user video bullet comment data, generate basic attribute analysis results of bullet comments, as well as bullet comment sentiment analysis information;

[0087] Based on the sentiment analysis information of the bullet comments, bullet comments with sentiment scores lower than a preset threshold are selected to generate negative bullet comment user profile results.

[0088] The generated basic attribute analysis results of bullet comments, bullet comment sentiment analysis information, and negative bullet comment user profile results are integrated to generate bullet comment analysis information.

[0089] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0090] Based on the aforementioned bullet screen sentiment analysis information, read video bullet screen data of users with negative bullet screens whose sentiment scores are below a preset threshold;

[0091] The negative bullet comment data of users is read and associated with a pre-established user scenario information database to obtain a bullet comment user location database.

[0092] Based on the aforementioned danmaku user location database, a detailed profile is created for the negative danmaku users of interest identified through analysis and location, generating negative danmaku user profile results.

[0093] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0094] The priority for handling key public opinion events is calculated based on the list of public opinion monitoring videos and the public opinion monitoring coefficients of the corresponding videos.

[0095] For videos whose public opinion processing priority reaches a preset threshold, a comprehensive intelligent public opinion analysis scheme is formed based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information, and the public opinion analysis results are output and / or pushed to external parties.

[0096] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0097] For videos whose public opinion processing priority reaches a preset threshold, the public opinion is presented in a contextualized geographic manner based on a pre-established user scenario information database.

[0098] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0099] Automatically collect user video bullet comment data through web crawlers, public API interfaces, or SDK embedding, and save it by category;

[0100] Collect user XDR data, MDT data, and MR data through platform integration;

[0101] Collect engineering parameter data;

[0102] Obtain scene border data through web crawling or public API interfaces.

[0103] Furthermore, when the intelligent public opinion analysis program in memory 130 is executed by the processor, it also performs the following steps:

[0104] The collected video bullet screen data is processed, including cleaning, merging and associating the collected video bullet screen data to form a user network information database;

[0105] Based on the user network information database and latitude and longitude matching, and by matching the scene border data, a user scene information database containing user network information and geographic information is established.

[0106] This embodiment collects user video bullet comment data through the above-described scheme; based on the collected user video bullet comment data, it generates a profile of the videos of interest, and based on the generated profile of the videos of interest, it generates preliminary intelligent public opinion analysis results; based on the collected user video bullet comment data and the preliminary intelligent public opinion analysis results, it generates bullet comment analysis information; and based on the preliminary intelligent public opinion analysis results and the bullet comment analysis information, it performs video public opinion analysis. Thus, it can automatically collect massive amounts of user video bullet comment data, generate profiles of the videos of interest, and generate preliminary intelligent public opinion analysis results. It can achieve detailed profiling of key users, identification of key hot information, prediction of dissemination and evaluation trends, and early warning of emergencies, as well as GIS presentation of the results. This enables efficient and accurate automatic monitoring and precise intelligent analysis of public opinion, achieving intelligent mining, analysis, and positioning of social public opinion, improving the accuracy of public opinion information, and providing an efficient and accurate public opinion solution for monitoring, tracking, and preventative maintenance of social hot events. The intelligent public opinion analysis method based on bullet comment collection proposed in this embodiment of the invention is lower in cost, more comprehensive in coverage, more efficient, and more accurate than traditional methods.

[0107] Based on, but not limited to, the terminal device architecture described above, embodiments of the method of the present invention are proposed.

[0108] Reference Figure 2 , Figure 2 This is a flowchart illustrating an exemplary embodiment of the intelligent public opinion analysis method of the present invention.

[0109] like Figure 2 As shown in the figure, this invention proposes an intelligent public opinion analysis method, which includes the following steps:

[0110] Step S10: Collect user video bullet comment data;

[0111] The collection of user video bullet comment data can include: automatically collecting user video bullet comment data through web crawlers, public API interfaces or SDK embedding points, and saving it by category.

[0112] The user video comment data mainly includes video information, comment information, and user information. The video information mainly includes: video category, video URL, video ID (the video ID is a general term for video IDs on various websites; for example, for website 1, this field is cid, for website 2, this field is video_id, etc.), video title, number of video views, video upload time, and video uploader information.

[0113] The bullet screen information mainly includes: video ID, timestamp of bullet screen appearance in video, bullet screen mode, bullet screen font size, bullet screen color, bullet screen Unix timestamp, bullet screen sender information, bullet screen text, IMEI, device model, etc.

[0114] The user information (including video uploaders and bullet screen commenters) mainly includes user ID, user nickname, user level, total number of video views, number of followers, following list, and friend list.

[0115] Furthermore, collecting user video bullet screen data can also be linked to collect other related data, including: collecting user XDR data, MDT data, and MR data through platform integration; collecting engineering parameter data; and obtaining scene border data through web crawlers or API public interfaces, so as to conduct intelligent public opinion analysis by combining the collected user video bullet screen data with scene border data, user XDR data, MDT data, MR data, engineering parameter data, etc.

[0116] Furthermore, after collecting user video comment data, the collected user video comment data can be processed, specifically including:

[0117] The collected user video bullet screen data is cleaned, stored, and fused together to form a user network information database for subsequent analysis.

[0118] Based on the user network information database and latitude and longitude matching, and further matching the scene border data, a user scene information database containing user network information and geographic information is established.

[0119] Step S20: Based on the collected user video bullet screen data, generate a profile of the video of interest, and based on the generated profile of the video of interest, generate preliminary intelligent public opinion analysis results.

[0120] In practice, as one implementation method, the following scheme can be adopted:

[0121] Based on the collected user video bullet screen data, a video description model is trained for different categories to obtain video description models for different categories.

[0122] Based on the video description model, a video profile model is established;

[0123] Based on the video profiling model, a profile of the video of interest is generated. Based on the generated profile of the video of interest, preliminary intelligent public opinion analysis results are generated. Specifically, this may include:

[0124] Based on the video profiling model, the public opinion monitoring coefficient of the video of interest is obtained;

[0125] Select a predetermined number of videos with the highest public opinion monitoring coefficients as video profiles of interest and generate corresponding video description information by category;

[0126] A list of public opinion monitoring videos is generated based on the corresponding video description information, serving as the preliminary result of intelligent public opinion analysis.

[0127] Furthermore, after generating the list of public opinion monitoring videos based on the generated corresponding video description information, it may also include:

[0128] For each video profile of interest corresponding to the public opinion monitoring video list, compare the corresponding video description information to obtain the similarity correlation of each video;

[0129] For multiple videos with a similarity correlation higher than a set threshold, the video with the highest public opinion monitoring coefficient is retained, the remaining similar videos are deleted, and replacement videos are sequentially selected from the preset number of videos according to the public opinion monitoring coefficient until the public opinion monitoring video list no longer changes, thus obtaining the final public opinion monitoring video list as the preliminary result of intelligent public opinion analysis.

[0130] Step S30: Generate bullet screen analysis information based on the collected user video bullet screen data and the preliminary intelligent public opinion analysis results;

[0131] Specifically, as one implementation method, the following scheme can be adopted:

[0132] For the videos of interest in the public opinion monitoring video list, read the corresponding user video comment data;

[0133] Based on the read user video bullet comment data, generate basic attribute analysis results for bullet comments and generate bullet comment sentiment analysis information;

[0134] Based on the aforementioned bullet screen sentiment analysis information, bullet screens with sentiment scores below a preset threshold are selected to generate negative bullet screen user profile results; specifically, this may include:

[0135] Based on the aforementioned bullet screen sentiment analysis information, read video bullet screen data of users with negative bullet screens whose sentiment scores are below a preset threshold;

[0136] The negative bullet comment data of users is read and associated with the previously established user scenario information database to obtain a bullet comment user location database.

[0137] Based on the aforementioned danmaku user location database, a detailed profile is created for the negative danmaku users of interest identified through analysis and location, generating negative danmaku user profile results.

[0138] Finally, the generated basic attribute analysis results of bullet comments, bullet comment sentiment analysis information, and negative bullet comment user profile results are integrated to generate bullet comment analysis information.

[0139] Step S40: Perform video public opinion analysis based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information.

[0140] Specifically, as one implementation method, the priority of handling key public opinion events can be calculated based on the list of public opinion monitoring videos and the public opinion monitoring coefficients of the corresponding videos;

[0141] For videos whose public opinion processing priority reaches a preset threshold, a comprehensive intelligent public opinion analysis scheme is formed based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information, and the public opinion analysis results are output and / or pushed to external parties.

[0142] Furthermore, for videos whose public opinion processing priority reaches a preset threshold, a contextualized geographic presentation of public opinion can be achieved based on the previously established user scenario information database.

[0143] This embodiment collects user video bullet comment data through the above-described scheme; based on the collected user video bullet comment data, it generates a profile of the videos of interest, and based on the generated profile of the videos of interest, it generates preliminary intelligent public opinion analysis results; based on the collected user video bullet comment data and the preliminary intelligent public opinion analysis results, it generates bullet comment analysis information; and based on the preliminary intelligent public opinion analysis results and the bullet comment analysis information, it performs video public opinion analysis. Thus, it can automatically collect massive amounts of user video bullet comment data, generate profiles of the videos of interest, and generate preliminary intelligent public opinion analysis results. It can achieve detailed profiling of key users, identification of key hot information, prediction of dissemination and evaluation trends, and early warning of emergencies, as well as GIS presentation of the results. This enables efficient and accurate automatic monitoring and precise intelligent analysis of public opinion, achieving intelligent mining, analysis, and positioning of social public opinion, improving the accuracy of public opinion information, and providing an efficient and accurate public opinion solution for monitoring, tracking, and preventative maintenance of social hot events. The intelligent public opinion analysis method based on bullet comment collection proposed in this embodiment of the invention is lower in cost, more comprehensive in coverage, more efficient, and more accurate than traditional methods.

[0144] The following is a detailed description of the solution in this embodiment:

[0145] The principle behind intelligent public opinion analysis in this invention is as follows: It automatically collects massive amounts of user video comment data, generates video description information through machine learning algorithms, further establishes a multi-dimensional video profile model, and generates a list of public opinion monitoring videos. It also automatically collects and correlates XDR data, MDT data, MR data, engineering parameter data, and geographic scene border data to establish a user scene information database. For trending videos in the public opinion monitoring video list, it further uses machine learning algorithms combined with the user scene information database to generate comment analysis information. This further enables detailed profiling of key users, identification of key hot information, prediction of dissemination and evaluation trends, and early warning of emergencies, along with GIS presentation of the results. This achieves efficient and accurate automatic monitoring and precise intelligent analysis of public opinion.

[0146] The specific implementation process of intelligent public opinion analysis in this invention embodiment can be referred to Figure 3 As shown. The specific process includes:

[0147] Step 100, Data Acquisition;

[0148] Step 200, Data Processing;

[0149] Step 300: Generate a profile of the video of interest and preliminary intelligent public opinion analysis results;

[0150] Step 400: Generate bullet screen analysis information;

[0151] Step 500: Intelligent analysis, positioning, and presentation of key public opinion.

[0152] Specifically, the data acquisition scheme for step 100 above is described as follows:

[0153] With the development of 5G networks and internet technology, video applications have gradually become a major mode of entertainment and social interaction. Bullet comments, as a new type of comment mode, are characterized by high real-time performance and strong interactivity, and have become a feature of the vast majority of video websites / apps. Compared to traditional text-based comments, bullet comments better reflect users' real-time opinions and emotions, while also rapidly increasing the speed and scope of information dissemination. Furthermore, when users watch videos and other applications online, they leave behind massive amounts of trace data, which can accurately reflect detailed user characteristics. To achieve subsequent analysis, massive amounts of data need to be collected, mainly including user video bullet comment data, scene border data, user XDR data, MDT data, MR data, and engineering parameter data. The specific steps for data collection can be as follows: Figure 4 As shown, it specifically includes:

[0154] Step 101: Collect user video bullet comment data;

[0155] User video comment data is automatically collected through web crawlers, public API interfaces, or SDK tracking, and saved in categories. The collected data mainly includes video information, comment information, and user information. The video information primarily includes: video category, video URL, video ID (the video ID is a general term for video IDs across various websites, e.g., cid for website 1, video_id for website 2, etc.), video title, number of video views, video upload time, and uploader information. The comment information primarily includes: video ID, timestamp of the comment appearing in the video, comment mode, comment font size, comment color, comment Unix timestamp, comment sender information, comment text, IMEI, device model, etc. The user information (including video uploaders and comment senders) primarily includes user ID, user nickname, user level, total video views, number of followers, following list, and friend list, etc.

[0156] Step 102, User XDR data acquisition;

[0157] User XDR data is collected through platform integration. The main data collected includes: S1-U / S11 and S1-MME raw bitstreams, MME UE S1AP ID, IMSI / IMEI (International Mobile Subscriber Identity / International Mobile Equipment Identity), LocalProvince (current province), LocalCity (current city), longitude, latitude, First-Play-Time, Procedure Start Time, Procedure End Time, URL (Uniform Resource Locator, also known as web address), AppName, etc.

[0158] Step 103, MDT data acquisition;

[0159] MDT data is automatically collected through platform integration. The collected data mainly includes: timestamp, Cell ID, IMSI / IMEI, longitude, latitude, MME UE S1AP ID, etc.

[0160] Step 104, MR data acquisition;

[0161] MR data can be automatically collected through platform integration. The collected data mainly includes: timestamp, MME UE S1APID, eNB ID, Cell ID, Location-longitude, Location-latitude, etc.

[0162] Step 105, Engineering parameter data acquisition;

[0163] The main data to be collected for engineering parameters include: province, city, district / county, network element name, cell name, CGI, CellID, coverage type, and coverage scenario.

[0164] Step 106, Scene border data acquisition;

[0165] Scene boundary data is obtained through web scraping or public API interfaces. Specifically, this includes obtaining the POI identifier (uid), and further parsing the POI name, POI boundary coordinate set, POI industry classification, and other information based on the obtained POI uid. The parsed POI boundary coordinate set is then converted into latitude and longitude format through coordinate transformation to form the scene boundary data.

[0166] Furthermore, the collected data can be cleaned and stored, such as... Figure 4 As shown, it also includes:

[0167] Step 107, Data Cleaning;

[0168] Data cleaning mainly involves classifying and parsing the massive amounts of data collected in the above steps, converting formats, normalizing fields with the same physical meaning, unifying capitalization, and deleting meaningless characters such as spaces.

[0169] Step 108, data storage;

[0170] The massive amounts of data output from the above steps, which can be used for analysis and application formats, are stored in the data unit.

[0171] It should be noted that the data cleaning in step 107 above can also be categorized under the following data processing steps, and no specific limitations are made thereto.

[0172] Specifically, the data processing scheme for step 200 above is described as follows:

[0173] Before proceeding with further analysis, the massive amounts of collected data need to be fused and correlated for optimal results. Figure 5 As shown, the main steps include the following:

[0174] Step 201, data fusion and association;

[0175] By associating user XDR data, MDT data, MR data, and engineering parameter data with timestamps, IMEI, IMSI, MME, UE S1AP ID, Cell ID, and eNB ID, a user network usage information database is formed, which includes timestamps, serving cell information, GPS latitude and longitude information, user identifiers (IMSI / IMEI), and user network behavior information (url).

[0176] Step 202: Establish a user scenario information database;

[0177] Based on the user network information database and latitude and longitude matching, the scene border data is further matched to form a user scene information database containing user network information and geographic information such as POI.

[0178] Specifically, the solution for generating the video profile and preliminary intelligent public opinion analysis results for step 300 above is described below:

[0179] This solution takes into account that the video data collected in the above steps may have inherent connections such as similar content, similar background, and causal relationships. Furthermore, differences in video category, publisher influence, etc., are also important indicators for public opinion analysis. Therefore, this embodiment establishes a video profiling model based on video description information, and further generates preliminary intelligent public opinion analysis results based on this model. Figure 6 As shown, the specific steps are as follows:

[0180] Step 301: Train a video description model for classification based on the collected user video bullet screen data.

[0181] Specifically, the video information collected in the aforementioned steps is divided into multiple subsets according to different video categories, and each subset is further trained using multiple machine learning algorithms. The performance of different algorithms for each subset is then evaluated and compared, and the optimal description model for each subset is selected.

[0182] Optionally, the video description algorithm can be selected from OA-BTG, MARN, MLE+HybridDis, etc.

[0183] Step 302: Based on the video description models for different categories obtained in the above steps, establish video profile models.

[0184] Specifically, the video profiling model is shown in Table 1 below:

[0185]

[0186] Table 1. Video Profiling Model

[0187] The correlation degree of similar videos can be calculated using algorithms such as PSNR and SSIM, and the weights of each tag can be obtained using algorithms such as TF-IDF, correlation coefficient matrix weight classification, and support vector machine.

[0188] Furthermore, based on the video profiling model, the public opinion monitoring coefficient Vy = L(Vb, Vs, Vc) of the video of interest can be calculated using the following formula, where L(·) can be selected from algorithms such as linear regression and convolutional neural networks.

[0189] Step 303: Generate and store preliminary intelligent public opinion analysis results. This specifically includes:

[0190] Step 3031: Based on the video description model and video profile model obtained in the preceding steps, select the videos with the highest public opinion monitoring coefficient ranking (e.g., the top 50) as public opinion monitoring videos (video identifiers and their URLs) and generate corresponding description information. Segment the generated description information using a word segmentation tool (e.g., jieba tool), and further perform word frequency statistics and sorting. Then, use the words with the highest word frequency (e.g., the top 5) as keywords for the video. Store the keywords of the video and the description information of the video together in the public opinion monitoring video list.

[0191] Step 3032: For the video set selected in step 3031, compare their descriptive information. For multiple videos with a similarity correlation (similar video correlation) higher than the set threshold value vt (e.g., 80%), only retain the video with the highest public opinion monitoring coefficient, delete the remaining similar videos, and select replacement videos in order according to the public opinion monitoring coefficient until the list of public opinion monitoring videos no longer changes.

[0192] Step 3033: Save the final generated list of public opinion monitoring videos (mainly including video identifiers, video URLs, video descriptions, video keywords, etc.) as the preliminary results of intelligent public opinion analysis.

[0193] Specifically, the scheme for generating bullet screen analysis information in step 400 above is described as follows:

[0194] After obtaining the list of public opinion monitoring videos and the corresponding preliminary public opinion analysis results, further analysis information is generated by combining the user video bullet screen data and user scene border data obtained in the aforementioned steps, such as... Figure 7 As shown, the specific steps are as follows:

[0195] Step 401: For videos in the public opinion monitoring video list, read the corresponding user video bullet comment data and generate bullet comment basic attribute analysis results. The bullet comment basic attribute analysis mainly includes:

[0196] Step 4011: By using the timestamps of the bullet comments appearing in the video and the Unix timestamps of the bullet comments, statistical analysis is performed on the distribution of sent bullet comments during video playback time and the distribution over natural time. High-popularity segments of the video and peak periods of user internet activity are identified and the results are saved.

[0197] Step 4012: Count the number and content (i.e., text of bullet comments) sent by each user based on the information of the bullet comment senders;

[0198] Step 4013: By segmenting the bullet screen text and counting word frequencies, select the top-ranked words (e.g., the top 5) as video bullet screen keywords.

[0199] Step 402: Generate sentiment analysis information for the bullet comments. This mainly includes:

[0200] Step 4021: For each video of interest, the collected video comment data is segmented and stop word removed, and then saved. The segmentation tools that can be used include, but are not limited to, Pymmseg-cpp, Loso, and Jieba. The stop words can be from a published Chinese stop word list, for example.

[0201] Step 4022: Convert the result of the previous step into a vector. For example, models such as Word2Vec and Bag of Words can be used.

[0202] Step 4023: Let the sentiment score of the bullet comments be df∈[-10, 10], where df∈Z. Take the vector transformation result as input and the sentiment score of the bullet comments as output. Use machine learning algorithms such as neural networks, random forests, and support vector machines to output and save the sentiment score of each bullet comment.

[0203] Step 4024, optionally, weight the emotional value score of each bullet comment in the video with the corresponding user level and the number of followers of the user to obtain the comprehensive emotional score of the video and save it; weight the emotional value scores of all bullet comments sent by the same user to obtain the comprehensive emotional score of the user and save it.

[0204] Step 403: Based on the sentiment analysis information of the bullet comments output in the previous step, select bullet comments with a sentiment score (df) lower than a preset threshold (Ne, for example, -5) to generate negative bullet comment user profile results. Specifically, this includes:

[0205] Step 4031: Read the negative video comment data of users whose sentiment score df is lower than the preset threshold Ne. By using fields such as timestamp, IMEI, and URL, associate the read user video comment data with the user scene information database to obtain a comment user location database containing timestamp, comment information, video information, user ID, user nickname, user level, total number of user video views, number of users followed, user follow list, user internet behavior (url), occupied service cell information, GPS latitude and longitude information, user identifier (IMSI / IMEI), POI geographic information, etc.

[0206] Step 4032: Based on the established bullet screen user location database, a detailed user profile can be created.

[0207] More specifically, it can analyze and locate the daily online time, frequent locations (e.g., frequent POI name, frequent POI industry category, daily activity trajectory, etc., and can further locate the indoor scene floor range through service community engineering parameter information), user's historical uploaded video information, user's main followers information (user ID, user nickname), user's friend list, etc., and at the same time, it can realize the GIS presentation of the distribution of the users of the negative bullet comments based on GPS-level precise location information.

[0208] Step 404: Integrate the generated basic attribute analysis results of bullet comments, bullet comment sentiment analysis information, and negative bullet comment user profile information to generate and store bullet comment analysis information.

[0209] Specifically, for step 500 above, the key solution for intelligent analysis, positioning, and presentation of public opinion is described as follows:

[0210] After acquiring the aforementioned bullet screen analysis information, video profiles, and preliminary intelligent public opinion analysis results, key public opinion can be intelligently analyzed, located, and the results presented, such as... Figure 8 As shown, the implementation steps are as follows:

[0211] Step 501: Prioritize key public opinion events;

[0212] The list of public opinion monitoring videos and the public opinion monitoring coefficients obtained through the above steps are further weighted by factors such as video upload time, location, category, number of bullet comments, and comprehensive sentiment score to determine the priority of handling key public opinion issues.

[0213] Step 502: For public opinion events that have reached the priority threshold, based on the preliminary analysis results of the intelligent public opinion analysis and the bullet screen analysis information, a comprehensive intelligent public opinion analysis scheme is formed, which includes hot event identification, refined profile of key users, and early warning of emergencies. Furthermore, for events that have reached the priority threshold, GPS-level precise location information is combined to realize the scenario-based geographic presentation of public opinion.

[0214] For videos whose public opinion processing priority reaches a preset threshold, based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information, comprehensive intelligent public opinion analysis is achieved, including hot keyword identification, automatic generation of public opinion summaries, detailed profiling of users who post excessive comments / malicious negative content, propagation path prediction (e.g., propagation paths can be predicted through user follow lists and friend lists), and event trend analysis (e.g., analysis of sentiment trends, spatiotemporal development trends, etc.), and corresponding solutions are generated.

[0215] Furthermore, based on a user scenario information database containing GPS-level precise location information, it is possible to realize the GIS-based presentation of public opinion, including industry distribution, attention, keyword geographic distribution, and sentiment geographic distribution, and to achieve functions such as precise positioning of the users of interest.

[0216] Step 503: Output and push the conclusion.

[0217] The intelligent public opinion analysis method based on bullet screen collection can achieve accurate, comprehensive, and dynamic intelligent automatic public opinion analysis, timely monitoring and early warning of major hot events, realize multi-dimensional and accurate profiling of hot videos and key users, provide automatic and effective public opinion analysis basis and solutions, and push them to governments, enterprises, media, etc., providing strong support for government supervision, brand image maintenance, media fair supervision, and social harmony and stability.

[0218] The present invention proposes an intelligent public opinion analysis method based on bullet screen data collection by automatically collecting massive amounts of user video bullet screen data and associating it with scene border data, user XDR data, MDT data, MR data, and engineering parameter data. Compared with traditional methods, this method has the following advantages:

[0219] Lower cost: Data collection is completed automatically through platform integration, and intelligent public opinion analysis solutions can be automatically generated, saving a lot of manpower and resources compared to traditional manual monitoring and analysis of public opinion;

[0220] More comprehensive coverage: This solution includes massive amounts of user behavior, scenario, and location information data, with abundant samples, avoiding the limitations and biases of traditional solutions;

[0221] More efficient: It can realize real-time data collection, analysis, mining and solution output, quickly identify hot events, keywords, etc., timely grasp the development trend of public opinion and public sentiment, and realize intelligent early warning, analysis and tracking of public opinion;

[0222] Higher accuracy: Due to the real-time interactivity of video applications and bullet comments, online public opinion can be represented quickly, extensively, deeply and realistically. In this embodiment, the proposed video profile model and negative bullet comment user profile model can accurately identify the characteristics of hot events compared with traditional solutions, and realize functions such as online public opinion (such as attention and sentiment distribution), negative user tracking, precise positioning and presentation. Compared with traditional solutions, it is more accurate and the precision is significantly improved.

[0223] Furthermore, this invention also proposes an intelligent public opinion analysis device, which includes:

[0224] The data acquisition module is used to collect user video comment data;

[0225] The list generation module is used to generate a profile of the videos of interest based on the collected user video bullet screen data, and to generate preliminary intelligent public opinion analysis results based on the generated profile of the videos of interest.

[0226] The information generation module is used to generate bullet screen analysis information based on the collected user video bullet screen data and the preliminary intelligent public opinion analysis results;

[0227] The analysis module is used to perform video public opinion analysis based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information.

[0228] The principle and implementation process of intelligent public opinion analysis in this embodiment are explained in the above embodiments and will not be repeated here.

[0229] Furthermore, this embodiment of the invention also proposes a terminal device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the intelligent public opinion analysis method as described in the above embodiments.

[0230] Since this intelligent public opinion analysis program adopts all the technical solutions of all the aforementioned embodiments when it is executed by the processor, it has at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be repeated here.

[0231] Furthermore, this embodiment of the invention also proposes a computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent public opinion analysis method as described in the above embodiments.

[0232] Since this intelligent public opinion analysis program adopts all the technical solutions of all the aforementioned embodiments when it is executed by the processor, it has at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be repeated here.

[0233] Furthermore, this invention also proposes a computer program product, characterized in that the computer program product includes a computer program, which, when executed by a processor, implements the intelligent public opinion analysis method as described in the above embodiments.

[0234] Since this intelligent public opinion analysis program adopts all the technical solutions of all the aforementioned embodiments when it is executed by the processor, it has at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be repeated here.

[0235] The intelligent public opinion analysis method, device, terminal equipment, and storage medium proposed in this invention collect user video bullet screen data; based on the collected user video bullet screen data, generate a profile of the video of interest; based on the generated profile of the video of interest, generate preliminary intelligent public opinion analysis results; generate bullet screen analysis information based on the collected user video bullet screen data and the preliminary intelligent public opinion analysis results; and perform video public opinion analysis based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information. Therefore, it can automatically collect massive amounts of user video bullet screen data, generate profiles of the video of interest based on the collected user video bullet screen data, and generate preliminary intelligent public opinion analysis results. It can realize analytical functions such as detailed profiling of key users, identification of key hot information, prediction of dissemination and evaluation trends, and early warning of emergencies, as well as GIS presentation of the results. This enables efficient and accurate automatic monitoring and precise intelligent analysis of public opinion, achieving intelligent mining, analysis, and positioning of social public opinion, improving the accuracy of public opinion information, and providing an efficient and accurate public opinion solution for monitoring, tracking, and preventative maintenance of social hot events. The intelligent public opinion analysis method based on bullet screen collection proposed in this invention is lower in cost, more comprehensive in coverage, more efficient, and more accurate than traditional methods.

[0236] Furthermore, this invention also proposes a dynamic networking device for a streaming media gateway, the dynamic networking device for a streaming media gateway comprising:

[0237] The join request receiving module is used to receive session join request messages sent by clients to join the target session;

[0238] The recording module is used to select a target gateway to join the target session based on the session join request message, and record the current session gateway list that has joined the target session;

[0239] The sending module is used to notify the target gateway that the target session has been created, notify other gateways in the current session gateway list of the address of the target gateway so that the other gateways can establish a transmission channel with the target gateway, and return the address of the target gateway to the client.

[0240] The principle and implementation process of dynamic networking of streaming media gateways in this embodiment are described in the above embodiments and will not be repeated here.

[0241] Furthermore, this invention also proposes a dynamic networking device for a streaming media gateway, the dynamic networking device for a streaming media gateway comprising:

[0242] The leave request receiving module is used to receive session leave request messages sent by clients to leave the target session;

[0243] The lookup module is used to find the target gateway associated with the client based on the session departure request message;

[0244] The deletion module is used to delete the client from the list of clients assigned by the target gateway associated with the client;

[0245] The determination and notification module is used to determine that the target gateway has left the target session if the client list is empty, and to notify other gateways in the current session gateway list of the session network to which the target session belongs of the message that the target gateway has left the target session, so that the other gateways can disconnect the transmission channel with the target gateway, thereby causing the target gateway to leave the session network.

[0246] The principle and implementation process of dynamic networking of streaming media gateways in this embodiment are described in the above embodiments and will not be repeated here.

[0247] Furthermore, this embodiment of the invention also proposes a terminal device, which includes a memory, a processor, and a streaming media gateway dynamic networking program stored in the memory and executable on the processor. When the streaming media gateway dynamic networking program is executed by the processor, it implements the steps of the streaming media gateway dynamic networking method as described in the above embodiments.

[0248] Since the dynamic networking program of this streaming media gateway adopts all the technical solutions of all the aforementioned embodiments when it is executed by the processor, it has at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be repeated here.

[0249] Furthermore, this embodiment of the invention also proposes a computer-readable storage medium storing a dynamic networking program for a streaming media gateway. When the dynamic networking program for a streaming media gateway is executed by a processor, it implements the steps of the dynamic networking method for a streaming media gateway as described in the above embodiments.

[0250] Since the dynamic networking program of this streaming media gateway adopts all the technical solutions of all the aforementioned embodiments when it is executed by the processor, it has at least all the beneficial effects brought about by all the technical solutions of all the aforementioned embodiments, which will not be repeated here.

[0251] Compared to existing technologies, the dynamic networking method, apparatus, system, terminal device, and medium for streaming media gateways proposed in this invention receive a session join request message from a client to join a target session; select a target gateway to join the target session based on the session join request message, and record a list of current session gateways that have joined the target session; notify the target gateway that the target session has been created, and notify other gateways in the current session gateway list of the address of the target gateway so that the other gateways can establish a transmission channel with the target gateway; and return the address of the target gateway to the client. This solution can dynamically adjust the network topology of streaming media gateways (such as WebRTC gateways) based on whether clients join or leave sessions. The streaming media gateway is responsible for forwarding client traffic to other streaming media gateway nodes in the session network related to the session, and then these other streaming media gateway nodes push the traffic to other clients in the session network, thereby reducing the forwarding of invalid traffic within the gateway cluster.

[0252] Compared with existing technologies, the present invention achieves dynamic joining and / or leaving of the streaming media gateway in a session-based manner through signaling services. This allows for dynamic adjustment of the streaming media gateway's network topology, enabling fine-grained scheduling and management of network traffic and efficient utilization of network resources.

[0253] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0254] The sequence numbers of the above embodiments of the present invention are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0255] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) as described above, and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, controlled terminal, or network device, etc.) to execute the methods of each embodiment of the present invention.

[0256] The above are merely preferred embodiments of the present invention and do not limit the scope of the patent. Any equivalent structural or procedural transformations made based on the description and drawings of the present invention, or direct or indirect applications in other related technical fields, are similarly included within the scope of patent protection of the present invention.

Claims

1. An intelligent public opinion analysis method, characterized in that, The method includes the following steps: Collect user video comment data; Based on the collected user video bullet comment data, a profile of the videos of interest is generated, and based on the generated profile of the videos of interest, preliminary intelligent public opinion analysis results are generated. This step includes: training a categorized video description model based on the collected user video bullet comment data to obtain video description models for different categories; establishing a video profile model based on the video description model; generating a profile of the videos of interest based on the video profile model; and generating preliminary intelligent public opinion analysis results based on the generated profile of the videos of interest. The video profile model includes profile dimensions, video profile tags, and tag weights. Based on the collected user video bullet comment data and the preliminary intelligent public opinion analysis results, bullet comment analysis information is generated. This step includes: reading the corresponding user video bullet comment data for the videos in the public opinion monitoring video list; generating basic attribute analysis results for bullet comments and generating bullet comment sentiment analysis information based on the read user video bullet comment data; selecting bullet comments with sentiment scores lower than a preset threshold based on the bullet comment sentiment analysis information to generate negative bullet comment user profile results; and integrating the generated basic attribute analysis results for bullet comments, bullet comment sentiment analysis information, and negative bullet comment user profile results to generate bullet comment analysis information. Based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information, video public opinion analysis was conducted. The steps of generating a video profile of interest based on the video profile model, and generating preliminary intelligent public opinion analysis results based on the generated video profile of interest, include: Based on the tag weights of the video profiling model, the public opinion monitoring coefficient of the video of interest is obtained. Select a predetermined number of videos with the highest public opinion monitoring coefficients as video profiles of interest and generate corresponding video description information by category; A list of public opinion monitoring videos is generated based on the corresponding video description information, serving as the preliminary result of intelligent public opinion analysis. The step of selecting bullet comments with sentiment scores below a preset threshold based on the bullet comment sentiment analysis information and generating negative bullet comment user profile results includes: Based on the aforementioned bullet screen sentiment analysis information, read video bullet screen data of users with negative bullet screens whose sentiment scores are below a preset threshold; The negative bullet comment user video bullet comment data is read and associated with a pre-established user scene information database to obtain a bullet comment user positioning database; the user scene information database is established based on automatically collected user video bullet comment data and associated with XDR data, MDT data, MR data, engineering parameter data, and geographic scene border data. Based on the aforementioned danmaku user location database, a detailed profile is created for the negative danmaku users of interest identified through analysis and location, generating negative danmaku user profile results.

2. The intelligent public opinion analysis method according to claim 1, characterized in that, Following the step of generating a list of public opinion monitoring videos based on the generated corresponding video description information, the following is also included: For each video profile of interest corresponding to the public opinion monitoring video list, compare the corresponding video description information to obtain the similarity correlation of each video; For multiple videos with a similarity correlation higher than a set threshold, the video with the highest public opinion monitoring coefficient is retained, the remaining similar videos are deleted, and replacement videos are sequentially selected from the preset number of videos according to the public opinion monitoring coefficient until the public opinion monitoring video list no longer changes, thus obtaining the final public opinion monitoring video list as the preliminary result of intelligent public opinion analysis.

3. The intelligent public opinion analysis method according to any one of claims 1-2, characterized in that, The steps for conducting video public opinion analysis based on the preliminary intelligent public opinion analysis results and bullet screen analysis information include: The priority for handling key public opinion events is calculated based on the list of public opinion monitoring videos and the public opinion monitoring coefficients of the corresponding videos. For videos whose public opinion processing priority reaches a preset threshold, a comprehensive intelligent public opinion analysis scheme is formed based on the preliminary intelligent public opinion analysis results and the bullet screen analysis information, and the public opinion analysis results are output and / or pushed to external parties; and / or for videos whose public opinion processing priority reaches a preset threshold, the public opinion is presented in a contextualized geographic manner based on a pre-established user scenario information database.

4. An intelligent public opinion analysis device, characterized in that, The intelligent public opinion analysis device includes: The data acquisition module is used to collect user video comment data; The list generation module is used to generate a profile of the videos of interest based on the collected user video bullet comment data, and to generate preliminary intelligent public opinion analysis results based on the generated profiles of the videos of interest. The list generation module is also used to train a categorized video description model based on the collected user video bullet comment data to obtain video description models for different categories; to establish a video profile model based on the video description model; to generate profiles of the videos of interest based on the video profile model; and to generate preliminary intelligent public opinion analysis results based on the generated profiles of the videos of interest. The video profile model includes profile dimensions, video profile tags, and tag weights. The list generation module is also used to obtain the public opinion monitoring coefficient of the video of interest based on the tag weight of the video profile model; select a preset number of videos with the highest public opinion monitoring coefficient as the video profile of interest and generate corresponding video description information; generate a list of public opinion monitoring videos based on the generated corresponding video description information as the preliminary result of intelligent public opinion analysis. The information generation module is used to generate bullet screen analysis information based on the collected user video bullet screen data and the preliminary intelligent public opinion analysis results. The information generation module is also used to read the corresponding user video bullet screen data from the videos in the public opinion monitoring video list; generate basic attribute analysis results for bullet screens based on the read user video bullet screen data, and generate bullet screen sentiment analysis information; select bullet screens with sentiment scores lower than a preset threshold based on the bullet screen sentiment analysis information to generate negative bullet screen user profile results; and integrate the generated basic attribute analysis results, bullet screen sentiment analysis information, and negative bullet screen user profile results to generate bullet screen analysis information. The information generation module is also used to read negative bullet screen user video bullet screen data with sentiment scores below a preset threshold based on the bullet screen sentiment analysis information; associate the read negative bullet screen user video bullet screen data with a pre-established user scenario information database to obtain a bullet screen user positioning database; and based on the bullet screen user positioning database, create a detailed profile of the negative bullet screen users of interest in the analysis and positioning, and generate negative bullet screen user profile results; the user scenario information database is established based on automatically collected user video bullet screen data and associated with XDR data, MDT data, MR data, engineering parameter data, and geographic scene border data; The analysis module is used to perform video-based public opinion analysis based on the preliminary results of the intelligent public opinion analysis and the bullet screen analysis information.

5. A terminal device, characterized in that, The terminal device includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the computer program is executed by the processor, it implements the intelligent public opinion analysis method as described in any one of claims 1-3.

6. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program, which, when executed by a processor, implements the intelligent public opinion analysis method as described in any one of claims 1-3.

Images