A video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis

Abstract

The invention discloses a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis. The invention extracts keywords from comments of each viewing user in a video live broadcast room to obtain keywords of each viewing user's comments, matches the keywords of each viewing user's comments with set uncivilized language keywords and theme-related keywords respectively, obtains the uncivilized coefficient and the non-theme coefficient of each viewing user's comments through processing, and obtains the malicious attack behavior coefficient of each viewing user's comments in a comprehensive manner, compares the malicious attack behavior coefficient of each viewing user's comments with the range of malicious attack behavior coefficients corresponding to different levels of early warnings respectively, performs corresponding processing on each viewing user according to the comparison result, realizes comprehensive, detailed and differentiated control over the speech of live broadcast viewing users, improves the rigor and reliability of video live broadcast supervision, and enhances the flexibility of video live broadcast supervision.

Description

Technical Field

[0001] The present invention relates to the field of live video monitoring, and in particular to a semantic analysis-based live video malicious attack behavior monitoring and early warning system. Background Art

[0002] The world today is an Internet age, and even more so an era of mobile Internet. The Internet has penetrated into all walks of life in daily life and has deeply affected people's lives. As a product of the Internet age, live streaming has changed the original media ecology and driven a large number of emerging industries with its unparalleled advantages in information dissemination. However, while live streaming has brought prosperity to new industries, it has also encountered many new problems due to irregular management. Uncivilized behaviors such as vulgarity, violence, and insults have seriously affected the healthy development of the live streaming industry. Therefore, it is urgent to strengthen the management of live streaming platforms, promote the healthy development of the live streaming industry, and purify the cyberspace environment. Malicious attacks on live video broadcasts are one of the uncivilized behaviors in live streaming, and monitoring of malicious attacks on live video broadcasts is of great significance.

[0003] At present, although the supervision of live video streaming has achieved certain initial results, there are still some drawbacks:

[0004] The existing supervision of live video streaming mainly regulates the speech and behavior of network anchors, but lacks control over the speech and behavior of users who watch live streaming, which makes the supervision of live video streaming one-sided and has supervision loopholes;

[0005] The current supervision of live video broadcasts focuses on the uncivilized speech during the broadcast, without considering whether the speech in the broadcast room is related to the live broadcast theme. This may cause confusion and affect the live broadcast effect, thus reducing the rigor and reliability of the supervision of live video broadcasts.

[0006] The existing video live broadcast supervision will take corresponding measures once uncivilized comments are detected in the live broadcast room. There is no further differentiation of the malicious attack degree of the uncivilized comments through data analysis, making it impossible to carry out detailed and differentiated processing according to the different malicious attack degrees of the uncivilized comments, which makes the flexibility of video live broadcast supervision low. Summary of the invention

[0007] In view of the above problems, the present invention proposes a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis, which realizes the function of monitoring and early warning of malicious attack behavior in video live broadcast.

[0008] The technical solution adopted by the present invention to solve the technical problem is:

[0009] The present invention provides a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis, comprising:

[0010] The live broadcast information database includes an uncivilized language text library, a theme-related text library, an uncivilized language picture library, and a theme-related picture library;

[0011] The live broadcast viewing user information acquisition module is used to obtain the ID and viewing comments of each viewing user in the target video live broadcast room, where the viewing comments include text comments, voice comments and picture comments;

[0012] The uncivilized language analysis module is used to analyze the viewing comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the viewing comments of each viewing user in the target video live broadcast room;

[0013] The non-topic related analysis module is used to analyze the viewing comments of each viewing user in the target video live broadcast room to obtain the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room;

[0014] The malicious attack behavior evaluation module is used to evaluate the uncivilized coefficient and the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room, and obtain the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room;

[0015] The early warning processing module is used to perform corresponding processing on each viewing user in the target video live broadcast room according to the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room.

[0016] On the basis of the above embodiment, the uncivilized language text library is used to store each set uncivilized language keyword, the theme-related text library is used to store each set theme-related keyword, the uncivilized language picture library is used to store each set uncivilized language picture, and the theme-related picture library is used to store each set theme-related picture.

[0017] On the basis of the above embodiment, the specific method for obtaining the ID and viewing comments of each viewing user in the target video live room in the live viewing user information acquisition module is:

[0018] Obtain the ID of each viewing user in the target video live room through the target video live room background and record it as ID i , i represents the number of the i-th viewing user in the target video live broadcast room, i=1,2,...,n;

[0019] Get the text comments of each viewing user in the target video live room, and number them in order of the time they were posted as W. i j, j represents the number of the jth text comment, j = 1, 2, ..., m, and further obtain the total number of text comments of each viewing user in the target video live broadcast room, the number of words in each text comment, and the release time of each text comment, which are recorded as NW i , ZW ij and TW ij ;

[0020] Get the voice comments of each user watching the target video live room, and number them in order of the time they were posted as Y i k , k represents the number of the kth voice comment, k = 1, 2, ..., f, through the voice-to-text technology, the text corresponding to each voice comment of each viewing user in the target video live broadcast room is obtained, and the total number of voice comments of each viewing user in the target video live broadcast room, the number of words of each voice comment, and the release time of each voice comment are further obtained, which are recorded as NY i , ZY ik TY ik ;

[0021] Get each picture comment of each viewing user in the target video live broadcast room, and number them in order of the comment release time as P i q , q represents the number of the qth picture comment, q = 1, 2, ..., r, and further obtain the total number of picture comments of each viewing user in the target video live broadcast room and the release time of each picture comment, which are recorded as NP i TP iq .

[0022] On the basis of the above embodiment, the uncivilized language analysis module includes an uncivilized language text analysis unit, an uncivilized language voice analysis unit and an uncivilized language picture analysis unit, wherein the uncivilized language text analysis unit is used to analyze the text comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the text comments of each viewing user in the target video live broadcast room, and the specific steps are:

[0023] Extract keywords from each text comment of each viewing user in the target video live broadcast room to obtain each keyword corresponding to each text comment of each viewing user in the target video live broadcast room;

[0024] Match each keyword corresponding to each text comment of each viewing user in the target video live broadcast room with each set non-civilized language keyword stored in the non-civilized language text library. If a text comment corresponding to a certain keyword of a viewing user in the target video live broadcast room is the same as a set non-civilized language keyword, then the text comment of the viewing user in the target video live broadcast room is recorded as a non-civilized text comment. The total number of non-civilized text comments of each viewing user in the target video live broadcast room is counted and recorded as NW i1 , further obtain the number of words and the release time of each non-civilized text comment of each viewing user in the target video live broadcast room, count the number of words and the release time of each non-civilized text comment of each viewing user in the target video live broadcast room, and number them as ZW iu and TW iu , u=1,2,...,v, where u represents the number of the u-th uncivilized text comment;

[0025] By formula Get the uncivilized coefficient of the text comments of each viewing user in the target video live room Among them, δ1, δ2, and δ3 are the influencing factors of the preset total number of text comments, the total number of words in the text comments, and the frequency of posting of text comments.

[0026] On the basis of the above embodiment, the uncivilized language voice analysis unit is used to analyze the voice comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the voice comments of each viewing user in the target video live broadcast room, and the specific steps are:

[0027] The text corresponding to each voice comment of each viewing user in the target video live broadcast room is obtained through speech-to-text technology, and the uncivilized coefficient of the voice comments of each viewing user in the target video live broadcast room is obtained according to the analysis method of the uncivilized coefficient of the text comments of each viewing user in the target video live broadcast room.

[0028] On the basis of the above embodiment, the uncivilized language picture analysis unit is used to analyze the picture comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the picture comments of each viewing user in the target video live broadcast room, and the specific steps are:

[0029] Compare the corresponding pictures of each picture comment of each viewing user in the target video live broadcast room with each set uncivilized language picture stored in the uncivilized language picture library to obtain the similarity between the corresponding pictures of each picture comment of each viewing user in the target video live broadcast room and each set uncivilized language picture. If the similarity between the corresponding picture of a certain picture comment of a certain viewing user in the target video live broadcast room and a set uncivilized language picture is greater than the set similarity, then the picture comment of the viewing user in the target video live broadcast room is recorded as an uncivilized picture comment. The total number of uncivilized picture comments of each viewing user in the target video live broadcast room is counted and recorded as NP i1 , further obtain the release time of each uncivilized picture comment of each viewing user in the target video live broadcast room, count the release time of each uncivilized picture comment of each viewing user in the target video live broadcast room, and number them as TP ih , h=1,2,...,g, where h represents the number of the hth uncivilized picture comment;

[0030] By formula Get the uncivilized coefficient of the picture comments of each viewing user in the target video live room Among them, γ1 and γ2 are the weight factors of the preset total number of picture comments and the frequency of picture comment publishing respectively.

[0031] On the basis of the above-mentioned embodiment, the specific steps of obtaining the uncivilized coefficient of the viewing comments of each viewing user in the target video live broadcast room in the uncivilized language analysis module are as follows:

[0032] The uncivilized coefficient of the text comments of each viewing user in the target video live room Uncivilized coefficient of voice comments Uncivilized coefficient of picture comments Substitute into the formula Get the uncivilized coefficient A of the viewing comments of each viewing user in the target video live broadcast room i ′, where ξ represents the correction factor of the uncivilized coefficient of the viewing user comments.

[0033] On the basis of the above-mentioned embodiment, the specific steps of obtaining the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room in the non-topic correlation analysis module are as follows:

[0034] Match each keyword corresponding to each text comment of each viewing user in the target video live broadcast room with each set theme-related keyword stored in the theme-related text library, and analyze to obtain the non-theme coefficient of the text comment of each viewing user in the target video live broadcast room

[0035] Match each keyword corresponding to each voice comment of each viewing user in the target video live broadcast room with each set theme-related keyword stored in the theme-related text library, and analyze to obtain the non-theme coefficient of the voice comment of each viewing user in the target video live broadcast room

[0036] Compare the corresponding pictures of each picture comment of each viewing user in the target video live room with the pictures related to each set theme stored in the theme-related picture library, and analyze to obtain the non-theme coefficient of the picture comment of each viewing user in the target video live room.

[0037] The non-topic coefficient of the text comments of each viewing user in the target video live room is Non-topic coefficient of voice comment Off-topic coefficient of picture comments Substitute into the formula Get the non-topic coefficient B of the viewing comments of each viewing user in the target video live broadcast room i ′, where ζ represents the correction index of the non-topic coefficient of the viewing user's comments.

[0038] On the basis of the above embodiment, the specific method of obtaining the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room in the malicious attack behavior analysis module is:

[0039] The uncivilized coefficient A of the comments of each viewing user in the target video live broadcast room is i ′ and the non-topic coefficient of viewing comments B i Substitute ′ into the formula φ i =ψ1*A i ′+ψ2*B i ′Get the malicious attack behavior coefficient φ of each viewing user’s comments in the target video live room i , where ψ1 and ψ2 represent the weight factors of the non-civility coefficient and non-topic coefficient of viewing user comments respectively.

[0040] On the basis of the above-mentioned embodiment, the specific steps of the early warning processing module for performing corresponding processing on each viewing user in the target video live broadcast room according to the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room are as follows:

[0041] Compare the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room with the preset malicious attack behavior coefficient range corresponding to the first, second, and third level warnings:

[0042] If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the first-level warning, a warning will be issued to the viewing user in the target video live broadcast room;

[0043] If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the second-level warning, the viewing user in the target video live broadcast room will be muted;

[0044] If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the third-level warning, the ID of the viewing user in the target video live broadcast room will be fed back to the network administrator.

[0045] Compared with the prior art, the video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis described in the present invention has the following beneficial effects:

[0046] The present invention provides a monitoring and early warning system for malicious attack behaviors in live video broadcasts based on semantic analysis. The system extracts keywords from the viewing comments of each viewing user in a live video broadcast room to obtain keywords of the viewing comments of each viewing user. The keywords of the viewing comments of each viewing user are matched with set uncivilized language keywords and set theme-related keywords respectively, and the uncivilized coefficient and the non-theme coefficient of the viewing comments of each viewing user are obtained through processing, so as to realize the control over the speech behaviors of the live broadcast viewing users, thereby making the supervision over the live video broadcast more comprehensive. At the same time, the question of whether the speech in the live broadcast room is related to the live broadcast theme is included in the supervision scope, thereby improving the rigor and reliability of the supervision over the live video broadcast.

[0047] The present invention provides a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis. By obtaining the uncivilized coefficient and the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room, the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room is comprehensively obtained. The malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room is compared with the malicious attack behavior coefficients corresponding to the first, second and third level warnings respectively. According to the comparison result, corresponding processing is performed on each viewing user in the target video live broadcast room, thereby realizing detailed and differentiated supervision of the viewing comments of each viewing user in the target video live broadcast room, and enhancing the flexibility of video live broadcast supervision. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for describing the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative work.

[0049] Figure 1 This is a system module connection diagram of the present invention. DETAILED DESCRIPTION

[0050] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0051] See also Figure 1 As shown, the present invention provides a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis, including a live broadcast information database, a live broadcast viewing user information acquisition module, an uncivilized language evaluation module, a non-topic related evaluation module, a malicious attack behavior evaluation module and an early warning processing module.

[0052] The live viewing user information acquisition module is connected to the uncivilized language evaluation module and the non-topic related evaluation module respectively, the malicious attack behavior evaluation module is connected to the uncivilized language evaluation module and the non-topic related evaluation module respectively, the early warning processing module is connected to the malicious attack behavior evaluation module, and the live broadcast information database is connected to the uncivilized language evaluation module and the non-topic related evaluation module respectively.

[0053] The live broadcast information database includes an uncivilized language text library, a theme-related text library, an uncivilized language picture library and a theme-related picture library.

[0054] Furthermore, the uncivilized language text library is used to store each set uncivilized language keyword, the theme-related text library is used to store each set theme-related keyword, the uncivilized language picture library is used to store each set uncivilized language picture, and the theme-related picture library is used to store each set theme-related picture.

[0055] The live broadcast viewing user information acquisition module is used to obtain the ID and viewing comments of each viewing user in the target video live broadcast room, wherein the viewing comments include text comments, voice comments and picture comments.

[0056] Furthermore, the specific method for obtaining the ID and viewing comments of each viewing user in the target video live room in the live viewing user information acquisition module is:

[0057] Obtain the ID of each viewing user in the target video live room through the target video live room background and record it as ID i , i represents the number of the i-th viewing user in the target video live broadcast room, i=1,2,...,n;

[0058] Get the text comments of each viewing user in the target video live room, and number them in order of the time they were posted as W i j , j represents the number of the jth text comment, j = 1, 2, ..., m, and further obtain the total number of text comments of each viewing user in the target video live broadcast room, the number of words in each text comment, and the release time of each text comment, which are recorded as NW i , ZW ij and TW ij ;

[0059] Get the voice comments of each user watching the target video live room, and number them in order of the time they were posted as Y i k , k represents the number of the kth voice comment, k = 1, 2, ..., f, through the voice-to-text technology, the text corresponding to each voice comment of each viewing user in the target video live broadcast room is obtained, and the total number of voice comments of each viewing user in the target video live broadcast room, the number of words of each voice comment, and the release time of each voice comment are further obtained, which are recorded as NY i , ZY ik TY ik ;

[0060] Get each picture comment of each viewing user in the target video live broadcast room, and number them in order of the comment release time as P i q , q represents the number of the qth picture comment, q = 1, 2, ..., r, and further obtain the total number of picture comments of each viewing user in the target video live broadcast room and the release time of each picture comment, which are recorded as NP i TP iq .

[0061] The uncivilized language analysis module is used to analyze the viewing comments of each viewing user in the target video live broadcast room to obtain the uncivilized language coefficient of the viewing comments of each viewing user in the target video live broadcast room.

[0062] Furthermore, the uncivilized language analysis module includes an uncivilized language text analysis unit, an uncivilized language voice analysis unit and an uncivilized language picture analysis unit, wherein the uncivilized language text analysis unit is used to analyze the text comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the text comments of each viewing user in the target video live broadcast room, and the specific steps are:

[0063] Extract keywords from each text comment of each viewing user in the target video live broadcast room to obtain each keyword corresponding to each text comment of each viewing user in the target video live broadcast room;

[0064] Match each keyword corresponding to each text comment of each viewing user in the target video live broadcast room with each set non-civilized language keyword stored in the non-civilized language text library. If a text comment corresponding to a certain keyword of a viewing user in the target video live broadcast room is the same as a set non-civilized language keyword, then the text comment of the viewing user in the target video live broadcast room is recorded as a non-civilized text comment. The total number of non-civilized text comments of each viewing user in the target video live broadcast room is counted and recorded as NW i1 , further obtain the number of words and the release time of each non-civilized text comment of each viewing user in the target video live broadcast room, count the number of words and the release time of each non-civilized text comment of each viewing user in the target video live broadcast room, and number them as ZW iu and TW iu , u=1,2,...,v, where u represents the number of the u-th uncivilized text comment;

[0065] By formula Get the uncivilized coefficient of the text comments of each viewing user in the target video live room Among them, δ1, δ2, and δ3 are the influencing factors of the preset total number of text comments, the total number of words in the text comments, and the frequency of posting of text comments.

[0066] Further, the uncivilized language voice analysis unit is used to analyze the voice comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the voice comments of each viewing user in the target video live broadcast room, and the specific steps are:

[0067] The text corresponding to each voice comment of each viewing user in the target video live broadcast room is obtained through speech-to-text technology, and the uncivilized coefficient of the voice comments of each viewing user in the target video live broadcast room is obtained according to the analysis method of the uncivilized coefficient of the text comments of each viewing user in the target video live broadcast room.

[0068] As a preferred solution, the specific steps of the above process to obtain the uncivilized coefficient of the voice comments of each viewing user in the target video live broadcast room are:

[0069] Match each keyword corresponding to each voice comment of each viewing user in the target video live broadcast room with each set non-civilized language keyword stored in the non-civilized language text library. If a certain keyword corresponding to a voice comment of a viewing user in the target video live broadcast room is the same as a certain set non-civilized language keyword stored in the non-civilized language text library, the voice comment of the viewing user in the target video live broadcast room is recorded as an uncivilized voice comment. The total number of uncivilized voice comments of each viewing user in the target video live broadcast room is counted and recorded as NY i1 , further obtain the number of words and release time of each uncivilized voice comment of each viewing user in the target video live broadcast room, count the number of words and release time of each uncivilized voice comment of each viewing user in the target video live broadcast room, and number them as ZY ix and TY ix , x=1,2,...,a, where x represents the number of the xth uncivilized voice comment;

[0070] By formula Get the uncivilized coefficient of the voice comments of each viewing user in the target video live room Among them, ε1, ε2, and ε3 are the influencing factors of the preset total number of voice comments, total number of words in voice comments, and frequency of voice comment publishing, respectively.

[0071] Further, the uncivilized language picture analysis unit is used to analyze the picture comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the picture comments of each viewing user in the target video live broadcast room, and the specific steps are:

[0072] Compare the corresponding pictures of each picture comment of each viewing user in the target video live broadcast room with each set uncivilized language picture stored in the uncivilized language picture library to obtain the similarity between the corresponding pictures of each picture comment of each viewing user in the target video live broadcast room and each set uncivilized language picture. If the similarity between the corresponding picture of a certain picture comment of a certain viewing user in the target video live broadcast room and a set uncivilized language picture is greater than the set similarity, then the picture comment of the viewing user in the target video live broadcast room is recorded as an uncivilized picture comment. The total number of uncivilized picture comments of each viewing user in the target video live broadcast room is counted and recorded as NP i1 , further obtain the release time of each uncivilized picture comment of each viewing user in the target video live broadcast room, count the release time of each uncivilized picture comment of each viewing user in the target video live broadcast room, and number them as TP ih, h=1,2,...,g, where h represents the number of the hth uncivilized picture comment;

[0073] By formula Get the uncivilized coefficient of the picture comments of each viewing user in the target video live room Among them, γ1 and γ2 are the weight factors of the preset total number of picture comments and the frequency of picture comment publishing respectively.

[0074] As a preferred solution, the above-mentioned method of obtaining the similarity between the corresponding picture of each picture comment of each viewing user in the target video live room and each set uncivilized language picture stored in the uncivilized language picture library is as follows:

[0075] Extract the pattern outline in the picture corresponding to each picture comment of each viewing user in the target video live broadcast room, overlap the pattern outline in the picture corresponding to each picture comment of each viewing user in the target video live broadcast room with the pattern outline in each set uncivilized language picture stored in the uncivilized language picture library, and obtain the overlap degree of the pattern outline in the picture corresponding to each picture comment of each viewing user in the target video live broadcast room and the pattern outline in each set uncivilized language picture stored in the uncivilized language picture library, which is recorded as the similarity between the picture corresponding to each picture comment of each viewing user in the target video live broadcast room and each set uncivilized language picture stored in the uncivilized language picture library.

[0076] Furthermore, the specific steps of obtaining the uncivilized coefficient of the viewing comments of each viewing user in the target video live broadcast room in the uncivilized language analysis module are:

[0077] The uncivilized coefficient of the text comments of each viewing user in the target video live room Uncivilized coefficient of voice comments Uncivilized coefficient of picture comments Substitute into the formula Get the uncivilized coefficient A of the viewing comments of each viewing user in the target video live broadcast room i ′, where ξ represents the correction factor of the uncivilized coefficient of the viewing user comments.

[0078] The non-topic related analysis module is used to analyze the viewing comments of each viewing user in the target video live broadcast room to obtain the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room.

[0079] Furthermore, the specific steps of obtaining the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room in the non-topic correlation analysis module are:

[0080] Match each keyword corresponding to each text comment of each viewing user in the target video live broadcast room with each set theme-related keyword stored in the theme-related text library, and analyze to obtain the non-theme coefficient of the text comment of each viewing user in the target video live broadcast room

[0081] Match each keyword corresponding to each voice comment of each viewing user in the target video live broadcast room with each set theme-related keyword stored in the theme-related text library, and analyze to obtain the non-theme coefficient of the voice comment of each viewing user in the target video live broadcast room

[0082] Compare the corresponding pictures of each picture comment of each viewing user in the target video live room with the pictures related to each set theme stored in the theme-related picture library, and analyze to obtain the non-theme coefficient of the picture comment of each viewing user in the target video live room.

[0083] The non-topic coefficient of the text comments of each viewing user in the target video live room is Non-topic coefficient of voice comment Off-topic coefficient of picture comments Substitute into the formula Get the non-topic coefficient B of the viewing comments of each viewing user in the target video live broadcast room i ′, where ζ represents the correction index of the non-topic coefficient of the viewing user's comments.

[0084] As a preferred solution, the non-topic coefficient of the text comments of each viewing user in the target video live room is The method to obtain is:

[0085] Match each keyword corresponding to each text comment of each viewing user in the target video live broadcast room with each set theme-related keyword stored in the theme-related text library. If each keyword corresponding to a text comment of a viewing user in the target video live broadcast room is different from each set theme-related keyword, then the text comment of the viewing user in the target video live broadcast room is recorded as a non-theme text comment. The total number of non-theme text comments of each viewing user in the target video live broadcast room is counted and recorded as NW. i1 ′, further obtain the number of words and the release time of each non-topic text comment of each viewing user in the target video live broadcast room, count the number of words and the release time of each non-topic text comment of each viewing user in the target video live broadcast room, and number them as ZW is ′ and TW is ′, s = 1, 2, ..., t, where s represents the number of the sth non-topic text comment;

[0086] By formula Get the non-topic coefficient of the text comments of each viewing user in the target video live room Among them, β1, β2, and β3 are the influencing factors of the preset total number of text comments, the total number of words in the text comments, and the frequency of posting of text comments.

[0087] As a preferred solution, the non-topic coefficient of the picture comments of each viewing user in the target video live room is The method to obtain is:

[0088] Compare the corresponding pictures of each picture comment of each viewing user in the target video live broadcast room with the pictures related to each set theme stored in the theme-related picture library to obtain the similarity between the corresponding pictures of each picture comment of each viewing user in the target video live broadcast room and the pictures related to each set theme. If the similarity between the corresponding picture of a certain picture comment of a certain viewing user in the target video live broadcast room and the pictures related to a certain set theme is less than the set similarity, then the picture comment of the viewing user in the target video live broadcast room is recorded as a non-theme picture comment. The total number of non-theme picture comments of each viewing user in the target video live broadcast room is counted and recorded as NP i1 ′, further obtain the release time of each non-topic picture comment of each viewing user in the target video live broadcast room, count the release time of each non-topic picture comment of each viewing user in the target video live broadcast room, and number them as TP id ′, d = 1, 2, ..., z, where d represents the number of the d-th non-topic picture comment;

[0089] By formula Get the non-topic coefficient of the picture comments of each viewing user in the target video live room Among them, κ1 and κ2 are the weight factors of the preset total number of picture comments and the frequency of picture comment publishing, respectively.

[0090] It should be noted that the present invention extracts keywords from the viewing comments of each viewing user in the video live broadcast room to obtain the keywords of each viewing comment of each viewing user, matches the keywords of each viewing comment of each viewing user with the set uncivilized language keywords and the set theme-related keywords respectively, and obtains the uncivilized coefficient and the non-theme coefficient of the viewing comment of each viewing user, thereby realizing the control over the speech and behavior of the live broadcast viewing users, thereby making the supervision of live video broadcast more comprehensive, and at the same time, bringing the issue of whether the speech in the live broadcast room is related to the live broadcast theme into the scope of supervision, thereby improving the rigor and reliability of the supervision of live video broadcast.

[0091] The malicious attack behavior evaluation module is used to evaluate the uncivilized coefficient and the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room, and obtain the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room.

[0092] Furthermore, the specific method for obtaining the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room in the malicious attack behavior analysis module is:

[0093] The uncivilized coefficient A of the comments of each viewing user in the target video live broadcast room is i ′ and the non-topic coefficient of viewing comments B i Substitute ′ into the formula φ i =ψ1*A i ′+ψ2*B i ′Get the malicious attack behavior coefficient φ of each viewing user’s comments in the target video live room i , where ψ1 and ψ2 represent the weight factors of the non-civility coefficient and non-topic coefficient of viewing user comments respectively.

[0094] The early warning processing module is used to perform corresponding processing on each viewing user in the target video live broadcast room according to the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room.

[0095] Furthermore, the specific steps of performing corresponding processing on each viewing user in the target video live broadcast room according to the viewing comment malicious attack behavior coefficient of each viewing user in the target video live broadcast room in the early warning processing module are:

[0096] Compare the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room with the preset malicious attack behavior coefficient range corresponding to the first, second, and third level warnings:

[0097] If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the first-level warning, a warning will be issued to the viewing user in the target video live broadcast room;

[0098] If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the second-level warning, the viewing user in the target video live broadcast room will be muted;

[0099] If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the third-level warning, the ID of the viewing user in the target video live broadcast room will be fed back to the network administrator.

[0100] It should be noted that the present invention obtains the uncivilized coefficient and the non-topic coefficient of the viewing comments of each viewing user in the target video live room, and comprehensively obtains the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live room. The malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live room is compared with the malicious attack behavior coefficients corresponding to the first, second and third level warnings, respectively, and each viewing user in the target video live room is processed accordingly according to the comparison result, thereby realizing detailed and differentiated supervision of the viewing comments of each viewing user in the target video live room, thereby enhancing the flexibility of video live supervision.

[0101] The above contents are merely examples and explanations of the concept of the present invention. The technicians in this technical field may make various modifications or additions to the specific embodiments described or replace them in a similar manner. As long as they do not deviate from the concept of the invention or exceed the scope defined by the claims, they should all fall within the protection scope of the present invention.

Claims

1. A video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis, characterized in that: include: The live broadcast information database includes an uncivilized language text library, a theme-related text library, an uncivilized language picture library, and a theme-related picture library; The live broadcast viewing user information acquisition module is used to obtain the ID and viewing comments of each viewing user in the target video live broadcast room, where the viewing comments include text comments, voice comments and picture comments; The uncivilized language analysis module is used to analyze the viewing comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the viewing comments of each viewing user in the target video live broadcast room; The non-topic related analysis module is used to analyze the viewing comments of each viewing user in the target video live broadcast room to obtain the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room; The malicious attack behavior evaluation module is used to evaluate the uncivilized coefficient and the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room, and obtain the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room; The early warning processing module is used to perform corresponding processing on each viewing user in the target video live broadcast room according to the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room.

2. According to the semantic analysis-based video live broadcast malicious attack behavior monitoring and early warning system according to claim 1, it is characterized by: The uncivilized language text library is used to store each set uncivilized language keyword, the theme-related text library is used to store each set theme-related keyword, the uncivilized language picture library is used to store each set uncivilized language picture, and the theme-related picture library is used to store each set theme-related picture.

3. According to claim 1, a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis is characterized by: The specific method for obtaining the ID and viewing comments of each viewing user in the target video live broadcast room in the live viewing user information acquisition module is: Obtain the ID of each viewing user in the target video live room through the target video live room background and record it as ID i , i represents the number of the i-th viewing user in the target video live broadcast room, i=1,2,...,n; Get the text comments of each viewing user in the target video live room, and number them in order of the time they were posted as W. i j , j represents the number of the jth text comment, j = 1, 2, ..., m, and further obtain the total number of text comments of each viewing user in the target video live broadcast room, the number of words in each text comment, and the release time of each text comment, which are recorded as NW i , ZW ij and TW ij ; Get the voice comments of each user watching the target video live room, and number them in order of the time they were posted as Y i k , k represents the number of the kth voice comment, k = 1, 2, ..., f, through the voice-to-text technology, the text corresponding to each voice comment of each viewing user in the target video live broadcast room is obtained, and the total number of voice comments of each viewing user in the target video live broadcast room, the number of words of each voice comment, and the release time of each voice comment are further obtained, which are recorded as NY i , ZY ik TY ik ; Get each picture comment of each viewing user in the target video live broadcast room, and number them in order of the comment release time as P i q , q represents the number of the qth picture comment, q = 1, 2, ..., r, and further obtain the total number of picture comments of each viewing user in the target video live broadcast room and the release time of each picture comment, which are recorded as NP i TP iq .

4. According to claim 1, a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis is characterized by: The uncivilized language analysis module includes an uncivilized language text analysis unit, an uncivilized language voice analysis unit and an uncivilized language picture analysis unit, wherein the uncivilized language text analysis unit is used to analyze the text comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the text comments of each viewing user in the target video live broadcast room, and the specific steps are: Extract keywords from each text comment of each viewing user in the target video live broadcast room to obtain each keyword corresponding to each text comment of each viewing user in the target video live broadcast room; Match each keyword corresponding to each text comment of each viewing user in the target video live broadcast room with each set non-civilized language keyword stored in the non-civilized language text library. If a text comment corresponding to a certain keyword of a viewing user in the target video live broadcast room is the same as a set non-civilized language keyword, then the text comment of the viewing user in the target video live broadcast room is recorded as a non-civilized text comment. The total number of non-civilized text comments of each viewing user in the target video live broadcast room is counted and recorded as NW i1 , further obtain the number of words and the release time of each non-civilized text comment of each viewing user in the target video live broadcast room, count the number of words and the release time of each non-civilized text comment of each viewing user in the target video live broadcast room, and number them as ZW iu and TW iu , u=1,2,...,v, where u represents the number of the u-th uncivilized text comment; By formula Get the uncivilized coefficient of the text comments of each viewing user in the target video live room Among them, δ1, δ2, and δ3 are the influencing factors of the preset total number of text comments, the total number of words in the text comments, and the frequency of posting of text comments.

5. According to claim 4, a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis is characterized by: The uncivilized language speech analysis unit is used to analyze the voice comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the voice comments of each viewing user in the target video live broadcast room, and the specific steps are: The text corresponding to each voice comment of each viewing user in the target video live broadcast room is obtained through speech-to-text technology, and the uncivilized coefficient of the voice comments of each viewing user in the target video live broadcast room is obtained according to the analysis method of the uncivilized coefficient of the text comments of each viewing user in the target video live broadcast room.

6. According to claim 4, a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis is characterized in that: The uncivilized language picture analysis unit is used to analyze the picture comments of each viewing user in the target video live broadcast room to obtain the uncivilized coefficient of the picture comments of each viewing user in the target video live broadcast room, and the specific steps are: Compare the corresponding pictures of each picture comment of each viewing user in the target video live broadcast room with each set uncivilized language picture stored in the uncivilized language picture library to obtain the similarity between the corresponding pictures of each picture comment of each viewing user in the target video live broadcast room and each set uncivilized language picture. If the similarity between the corresponding picture of a certain picture comment of a certain viewing user in the target video live broadcast room and a set uncivilized language picture is greater than the set similarity, then the picture comment of the viewing user in the target video live broadcast room is recorded as an uncivilized picture comment. The total number of uncivilized picture comments of each viewing user in the target video live broadcast room is counted and recorded as NP i1 , further obtain the release time of each uncivilized picture comment of each viewing user in the target video live broadcast room, count the release time of each uncivilized picture comment of each viewing user in the target video live broadcast room, and number them as TP ih , h=1,2,...,g, where h represents the number of the hth uncivilized picture comment; By formula Get the uncivilized coefficient of the picture comments of each viewing user in the target video live room Among them, γ1 and γ2 are the weight factors of the preset total number of picture comments and the frequency of picture comment publishing respectively.

7. According to claim 1, a video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis is characterized by: The specific steps of obtaining the uncivilized coefficient of the viewing comments of each viewing user in the target video live broadcast room in the uncivilized language analysis module are: The uncivilized coefficient of the text comments of each viewing user in the target video live room Uncivilized coefficient of voice comments Uncivilized coefficient of picture comments Substitute into the formula Get the uncivilized coefficient A of the viewing comments of each viewing user in the target video live broadcast room i ′, where ξ represents the correction factor of the uncivilized coefficient of the viewing user comments.

8. The video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis according to claim 1 is characterized by: The specific steps of obtaining the non-topic coefficient of the viewing comments of each viewing user in the target video live broadcast room in the non-topic correlation analysis module are: Match each keyword corresponding to each text comment of each viewing user in the target video live broadcast room with each set theme-related keyword stored in the theme-related text library, and analyze to obtain the non-theme coefficient of the text comment of each viewing user in the target video live broadcast room Match each keyword corresponding to each voice comment of each viewing user in the target video live broadcast room with each set theme-related keyword stored in the theme-related text library, and analyze to obtain the non-theme coefficient of the voice comment of each viewing user in the target video live broadcast room Compare the corresponding pictures of each picture comment of each viewing user in the target video live room with the pictures related to each set theme stored in the theme-related picture library, and analyze to obtain the non-theme coefficient of the picture comment of each viewing user in the target video live room. The non-topic coefficient of the text comments of each viewing user in the target video live room is Non-topic coefficient of voice comment Off-topic coefficient of picture comments Substitute into the formula Get the non-topic coefficient B of the viewing comments of each viewing user in the target video live broadcast room i ′, where ζ represents the correction index of the non-topic coefficient of the viewing user's comments.

9. The video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis according to claim 1 is characterized by: The specific method of obtaining the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room in the malicious attack behavior analysis module is: The uncivilized coefficient A of the comments of each viewing user in the target video live broadcast room is i ′ and the non-topic coefficient of viewing comments B i Substitute ′ into the formula φ i =ψ1*A i ′+ψ2*B i ′Get the malicious attack behavior coefficient φ of each viewing user’s comments in the target video live room i , where ψ1 and ψ2 represent the weight factors of the non-civility coefficient and non-topic coefficient of viewing user comments respectively.

10. The video live broadcast malicious attack behavior monitoring and early warning system based on semantic analysis according to claim 1 is characterized by: The specific steps of performing corresponding processing on each viewing user in the target video live broadcast room according to the viewing comment malicious attack behavior coefficient of each viewing user in the target video live broadcast room in the early warning processing module are: Compare the malicious attack behavior coefficient of the viewing comments of each viewing user in the target video live broadcast room with the preset malicious attack behavior coefficient range corresponding to the first, second, and third level warnings: If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the first-level warning, a warning will be issued to the viewing user in the target video live broadcast room; If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the second-level warning, the viewing user in the target video live broadcast room will be muted; If the malicious attack behavior coefficient of a viewing comment of a certain viewing user in the target video live broadcast room is within the preset malicious attack behavior coefficient range corresponding to the third-level warning, the ID of the viewing user in the target video live broadcast room will be fed back to the network administrator.

Images