TWI602430B - Multimedia content classification system and method - Google Patents

Description

Multimedia content classification system and method

The present invention relates to a multimedia content classification system and method, and more particularly to a system and method for collecting information of an external social media content interface platform and a multimedia content interface platform and dynamically adjusting the content provided to the user.

In order to cope with the current trend of multimedia users accessing content through multiple screens, future video platforms should extend to different terminals at the same time.

In order to adjust the content of the appropriate audio-visual area for the user according to the user's operating behavior and usage habits, the prior art has a way of connecting the user records generated by the back end of each audio-visual platform, however, this method will lead to The proposed audio and video service has platform limitations or limitations of the classification type. In addition, today's on-board boxes are often shared by multiple people or family members. Therefore, this method is not suitable for analyzing the user's operation behavior of a single individual and then adjusting the content provided.

In addition, some prior art methods achieve flexibility in providing service content through a method in which a user sets information such as the theme of a home page menu through a web page. However, although this method can barely achieve the effect of providing content personalization, the premise is that the user needs to set the party through the terminal device. It can be achieved, and this method will not change the theme area of audio and video over time, which will result in higher user entry threshold and lower willingness to use.

Therefore, if we can provide a method for dynamically classifying and adjusting the sorted multimedia content with the changes of the community's concerns, it can solve the problem of analyzing distortion caused by the popularity of clicks in the past, and can effectively increase the exposure of the unpopular video. Dynamic methods are indispensable technologies for this field.

The present invention provides a multimedia content classification system and method, the purpose of which is to enable a dynamic classification and adjustment of the subject of sorting multimedia content as the focus of the community changes, the main technical content of which will be described in the following paragraphs.

The multimedia content classification method of the present invention includes a dynamic method for calculating attention, because the degree of attention of the event is mainly determined by the number of clicks in the prior art, and the multimedia content with a high total number of clicks is always in the Hot status, even as time goes by, its status may not be shaken. According to the current network media ecology and the speed of replacement, this method has the disadvantage of not being able to change quickly and more with the trend; while the invention is to grow the time relationship value and the event within the unit time. Compared with the consideration, when the overall attention of an event in the region and time interval of the multimedia user's use of the service can be calculated, the information such as the selected topic can be dynamically adjusted according to the degree of attention in the interval. Provided.

The multimedia content classification method of the present invention includes a data collection method. The prior art uses the number of clicks or the personal operation behavior of the multimedia user to collect data, and the present invention is caused by the locking clustering effect. The marginal impact is to capture the focus of attention in the community, aiming to solve the problem of less clicks on multimedia content that needs to be paid, or devices such as STB (Set-Top Box) are often shared. The problem of analytical accuracy caused by the situation, as well as the vicious circle that causes the exposure of less popular multimedia content to be too low.

The multimedia content classification method of the present invention further includes an optimized ranking method, which selects a theme with high fitness through a multimedia content platform and sorts according to the degree of attention. The so-called theme refers to an individual with an obvious impression, and may be a character. Individuals of events, places, organizations, etc., the present invention can further sort the topics according to the degree of attention and then provide them to the multimedia content platform.

Furthermore, the method for classifying the multimedia content of the present invention mainly uses an analysis server to obtain the regional location of the user terminal device through the mediation server, and then through the social media content interface platform and multimedia. The content interface platform can obtain multimedia content and community event content, and can additionally store social media content and multimedia content into the community and multimedia content library. The community and multimedia content library can also be set. Equipment outside the system.

In the present invention, the number of the subject, the area, the kanban type, and the like transmitted by the user terminal device are obtained, and the community event content related to the multimedia user use area and the love kanban type is selected, and the benefit event can be additionally excluded. Then analyze the most interesting events that are suitable for the user and then sort them.

In detail, the present invention provides a multimedia content classification system, and the system architecture mainly includes the following parts.

The multimedia content classification system of the present invention comprises an intermediary server connected to the social media content interface platform outside the system and the external multimedia content interface platform through the network, and the foregoing The external social media content interface refers to an application programming platform that collects social media data from multiple social media websites (Ptt, Sina, Weibo, Facebook, etc.) and processes data structured programs. And the external multimedia content interface platform refers to an application interface platform for collecting metadata data of multimedia content files in a plurality of multimedia content websites (such as YouTube, MOD, kkBox, Spotify, etc.), the intermediary The server system acts as a mediator in tandem with the overall system program and serves as a bridge between the various platforms.

The multimedia content classification system of the present invention comprises an analysis server, which is connected to the mediation server, and is connected to the external social media content platform and multimedia content through the serial connection of the mediation server. The social media content data and the multimedia content selectively acquired by the platform, and the social media content data collected according to an interest event analysis algorithm for the external multimedia user usage interval and a theme optimization algorithm And the multimedia content is calculated, and finally an analysis content is generated.

The multimedia content classification system of the present invention further includes a web application server, which is coupled to the terminal device of the external multimedia content user through the multimedia content platform. The analysis content generated by the analysis server is processed according to a specific format to provide a customized service content corresponding to an individual external multimedia content user, so that the classification service provided by the overall system can more conform to the user's needs. At the same time, the exposure rate of the subject category content with less interest is added.

The foregoing attention event analysis algorithm is an algorithm for selecting a attention event with a high degree of interest from a large amount of content, and the main reference parameters of the attention event analysis algorithm are two, respectively, for individual external multimedia content. The area in which the user is located and the time period in which they are used. , The performing the event analysis algorithm program includes the following steps: 1. The analysis server of the present invention obtains the use area of the terminal device of the individual external multimedia content user through the mediation server; 2. the analysis servo The media server collects the social media content data and the multimedia content data from the external social media content interface platform and the external multimedia content interface platform through the mediation server; 3. The analysis server uses the social media content data and the multimedia content. The content of the film and the content of the social event included in the data are stored in the theme data table of the film and the event content table; 4. The analysis server obtains the number of content topics and the area where the content of the external multimedia content user is frequently used. And the kanban type information; 5. The analytics server finds an event that matches the area in which the external multimedia content user frequently uses and the type of the viewing version from the event content table in which the content of the community event is stored; 6. The analysis servo Excluding individual external multimedia content users only for events such as category events And after the exclusion, the step of obtaining the content of the topic number, the area and the kanban type of the content of the content of the external external multimedia content is performed again; 7. The analysis server performs the analysis of the event of interest, and the event analysis is popular. The value growth ratio and the time relationship value are both reference parameters, which are used to calculate a attention weight value for the complex attention event, which is the calculation result of the popularity value growth ratio divided by the time relationship value. . The popular value growth ratio in the reference parameter is calculated by the TF-IDF (term frequency-inverse document frequency) algorithm to calculate the popularity value of the attention event, and the time relationship value is used to calculate the attention event and the attention event analysis. The program is executed The degree of relationship caused by the degree of time difference of the timing points; 8. The analysis server sorts each of the attention events according to the attention weight value to select a selectable number of attention events with high attention weights.

In the multimedia content classification system of the present invention, the theme optimization algorithm is configured to calculate a multimedia theme set suitable for a user by using variability between platforms for individual external multimedia content users, wherein The execution of the theme optimization algorithm program includes the following steps: 1. The analysis server obtains the lower limit of the video content content of the individual external multimedia content user; 2. The analysis server selects the attention weight value from the attention event. Filtering out the attention events of the top ranked rankings; 3. The analysis server classifies the filtered attention events according to the movie theme and the event theme; 4. The analysis server determines whether the number of collection topics is not empty, and the video and audio Whether the lower limit of the content is greater than or equal to a certain value, and if not, filtering out the attention events of the most prior rankings from the attention attention weights in the attention event; and 5. the analysis server will select and pass the judgment The set of topics of interest is sorted by the weight of interest.

In the multimedia content classification system of the present invention, the calculation method of the popularity value growth ratio in the reference event analysis reference parameter is a single key for the analysis server to extract the social media content data and the multimedia content data. The word vocabulary appears in the subject content of each social media or multimedia classification kanban, in the message content, in the kanban classification name, and in the content collection of the searched keywords, TF-IDF (term frequency-inverse document frequency) of each individual vocabulary Value. Among them, the calculation formula of the value of TF (term frequency) is as follows: , d _j D ; In addition, the calculation formula of the IDF (inverse document frequency) value is as follows: In the above formula, D represents the total number of all social media content materials and multimedia content data files, and i represents the serial number of individual keyword vocabulary, D represents the total number of all social media content materials and multimedia content data files, and d _j represents D. In the individual files, t _i represents an individual vocabulary, Σ _j count ( t _i , d _j ) represents the number of occurrences of t _i in d _j , and Σ _k Σ j count ( t _k , d _j ) represents d _j The number of occurrences of all words; and | t _i { j : t _i d _j }| represents the number of files that appear in the total number of all social media content materials and multimedia content data for the i- th individual keyword vocabulary. Among them, the final TF-IDF value is calculated as follows: ( tf - idf ) _i = tf _i × idf _i .

Multimedia content classification system. Among them, the time relationship value in the attention event analysis reference parameter refers to the current time point of the year, month, day, minute, and second of the individual attention event being issued, and the current time point of the year, month, day, minute, and second of the event analysis program being executed. Take the absolute value again. Wherein, the expression of the time relationship value is as follows: f ( a , b , c , d , e , f )= a _i λ _year + b _i λ _month + c _i λ _day + d _i λ _hour + e _i λ _min + f _i λ _sec + C ; where f (a, b, c, d, e, f) 1, C is a non-zero constant, i represents the number of individual attention event numbers, a _i , b _i , c _i , d _i , e _i , f _i respectively represent the absolute value of the difference between the year, month, day, hour, minute and second; , λ _year = 10 ¹⁰ , λ _month = 10 ⁸ , λ _day = 10 ⁶ , λ _hour = 10 ⁴ , λ _min = 10 ² , λ _sec = 10 ⁰ .

With respect to the system of the present invention, the present invention also provides a multimedia content classification method, the steps of which include: 1. An analysis server is connected through an intermediary server to collect external social media content interface platform and multimedia. The content mediation of the social media content data and the multimedia content of the platform, and the social media content data and the multimedia content are calculated according to an event analysis algorithm and a theme optimization algorithm to generate an analysis content; and 2 A web application server provides the service content generated by the analysis server to a multimedia content user corresponding to an individual external content according to a specific format.

In summary, the present invention provides a multimedia content classification system and method, the main purpose of which is to develop a technology that can dynamically classify and adjust the content of sorted multimedia content as the focus of the community changes.

A‧‧‧Community media content interface platform

B‧‧‧Multimedia Content Interface Platform

1‧‧‧Intermediary server

2‧‧‧Analysis server

3‧‧‧Web application server

4‧‧‧Community and multimedia content library

5‧‧‧Multimedia content platform

S201~S207‧‧‧ method steps

S301~S309‧‧‧ method steps

S401~S406‧‧‧ method steps

1 is a system architecture diagram of a multimedia content classification system of the present invention.

2 is a schematic diagram of steps of a method for classifying a multimedia content according to the present invention.

FIG. 3 is a schematic diagram of method steps of an event analysis algorithm of interest in a multimedia content classification method according to the present invention.

4 is a method diagram of a method for optimizing a theme algorithm in a multimedia content classification method according to the present invention;

The present invention will be further described by way of embodiments with reference to the drawings. First, please refer to FIG. 1, which is a system architecture diagram of the multimedia content classification system of the present invention. There is an intermediary server 1 in the system, and the intermediary server 1 is connected to the external social media content interface platform A and the external multimedia content interface platform B through the network. The external social media content interface platform A and the plural social media website link ( ) to collect community multimedia materials, and the external multimedia content interface platform B is used to collect plural multimedia content websites ( ) Metadata of the multimedia content files within.

With continued reference to FIG. 1, the system of the present invention also includes an analysis server 2, which is coupled to the mediation server 1 to enable collection of external social media content interface platform A and multimedia content media. Based on the data of the platform B, the analysis server 2 performs an operation on the collected data according to the focused event analysis algorithm and the topic optimization algorithm for analyzing the interval and time interval of the user to generate an analysis content.

The system also includes a web application server 3 for providing the analysis content generated by the analysis server 2 to the user according to a set format, and the web application server 3 connects the multimedia through the network. The content platform 5, the multimedia user can connect to the multimedia content platform 5 through its terminal device to obtain service content suitable for their respective preferences.

Finally, the system of the present invention includes a community and multimedia content library 4, and the community and multimedia content library 4 is coupled to the mediation server 1 for storing the collected social media content data, multimedia content data, analysis Content or content of the service.

Referring to FIG. 2, it is a method step diagram of the multimedia content classification method of the present invention, which can be roughly divided into the following steps: Step S201: Acquire a use area of the terminal device; Step S202: Interface platform content and community event content Step S203: storing the platform content and the community event content; step S204: analyzing the theme of the content and the community event content; step S205: focusing on the event analysis; step S206: optimizing the theme algorithm; step S207: generating the theme Gather to the user.

Among them, the detailed procedures of each step will be described in the following paragraphs.

Please refer to FIG. 3 , which is a method step diagram of an event analysis algorithm for interest in a multimedia content classification method according to the present invention. The steps are as follows: Step S301: Obtain a use area of the terminal device, that is, the analysis server passes the intermediary. a server to obtain a use area where the terminal device of the external multimedia content user is located; step S302: interface between the platform piece content and the community event content, that is, the analysis server sends the mediation server to the external community through the mediation server The media content interface platform and the external multimedia content interface platform obtain the social media content data and the multimedia content data; step S303: storing the content of the piece and the content of the community event, the analysis server stores the social media content data and the multimedia The content of the content in the content material and the content of the community event to the theme information table of the movie and the event content table; Step S304: obtaining the number of the subject, the area, and the kanban type of the interface, and obtaining the external external for the analysis server Multimedia content users use frequently used content topics, location and kanban type information; Step S305: Find an event that matches the area and the kanban type from the event data table, and the analysis server finds, from the event content table, an event that matches the area and the type of the version in which the external multimedia content user frequently uses. Step S306: Whether it is a request-type event, which is a judgment step, the analysis server will judge and exclude the event of the individual external multimedia content user to perform the benefit category, and after the exclusion, proceed to step S305 to obtain the individual external The multimedia content user uses the frequent content subject number, the area and the kanban type information step; step S307: focus on the event analysis, and the analysis server performs the attention event analysis by using the popular value growth ratio and the time relationship value as reference parameters. The plural attention event calculates a attention weight value, which is a calculation result of the value of the popular value growth ratio divided by the time relationship value; step S308: sorting the events according to the weight value, the analysis server will focus on each of the concerns The events are sorted according to the attention weight value to select each of the attention events with high attention weight; And step S309: retrieving Top (m) event from the event information table, that is, the analysis server will be selected m sort of attention before the event most of the attention, according to the weight value ordering of the event of interest.

Among them, the popularity value growth ratio in the reference parameter is calculated by the TF-IDF algorithm to calculate the popularity value of the event of interest into a length. The example is expressed as: =TF-IDF( a , b , c , d ), the meaning of which is the high word frequency and then multiply the low file frequency of the word in the total number of files, then the TF-IDF weight value can be generated, where a represents the electronic forum. The content of the kanban; b represents the message content of each classification kanban; c represents the kanban classification name; d represents the search keyword of each classification kanban. The time relationship value is the time point difference between the timing of the execution of the attention event and the attention event analysis program, which can be expressed as: = f (|Today (Y)-release date (Y)|,|Today (M)-release date (M)|,|Today (D)-release date (D)|,|Today (H)-release date (H)|,|Today (M)-release date (M)|,|Today (S)-release date(S)|), the difference is accurate to the number of seconds, the actual calculation method is f ( a , b , c , d , e , f ) = a _i λ _year + b _i λ _month + c _i λ _day + λ _hour + e _i λ _min + f _i λ _sec + C ; where λ _year =10 ¹⁰ , λ _{month ^{= 10 8, λ day = 10}} 6, λ hour = 10 4, λ min = 10 2, λ sec = 10 0.

The weight of attention is based on / To calculate.

In the above, the steps of the event analysis algorithm of interest in the multimedia content classification method of the present invention are completed.

Next, referring to FIG. 4, which is a method step diagram of optimizing a theme algorithm in the multimedia content classification method of the present invention, the steps of the method include: Step S401: Obtaining the lower and lower video and audio number (n) of the interface, which is the analysis servo Obtaining the lower limit of the content of the audio and video content of the individual external multimedia content user; Step S402: Obtaining the Top(m) event in the event data set, and the analysis server filters out the highest ranking according to the set attention weight value from the attention event The m-th order attention event; step S403: classifying the film-single topic and the event topic, wherein the analysis server classifies the filtered attention event according to the piece-single theme and the event topic; step S404: the number of the assembly topics is not zero, a judging step, that is, the analysis server determines whether the number of collection topics is not empty, and if the result is no, returns to step S402, which again filters out the attention of the top m from the attention event weights according to the set attention degree weight value. Event; step S405: the number of lower and lower audio and video is greater than or equal to the fixed value, and is also a determining step, whether the lower limit of the content of the video or audio is greater than or equal to a certain value, and if the result is no Returns to step S402, it is again filtered out of the events of interest in accordance with the set value of the heavy weights in attention Sorting the attention event of the top m; Step S406: Sorting the topic set according to the weight value, and sorting the theme set selected by the analysis server and determining the attention event according to the attention weight value.

In the above, the steps of optimizing the theme algorithm in the multimedia content classification method of the present invention are completed. Among them, the three main components of the main reference are: combined attention weight value (considering the use of time interval), the number of topics required to meet the platform (m), and the number of lower audio and video (n).

The present invention does provide a system and method for dynamically classifying and adjusting the sorted multimedia content with changes in the focus of the community, which can solve the problem that the popularity of the clicks is not used in the usage interval, or the paid video and audio causes fewer clicks. The number of times, or the analysis distortion caused by the sharing of the set-top box.

In summary, the present invention is innovative in terms of technical ideas, and also has various functions that are not in the prior art, and has fully complied with the statutory invention patent requirements of novelty and progressiveness, and has filed a patent application according to law, and invites you to approve the invention. The patent application was inspired to invent, and it was a matter of feeling.

A‧‧‧Community media content interface platform

B‧‧‧Multimedia Content Interface Platform

1‧‧‧Intermediary server

2‧‧‧Analysis server

3‧‧‧Web application server

4‧‧‧Community and multimedia content library

5‧‧‧Multimedia content platform

Claims (8)

A multimedia content classification system, comprising: an intermediary server, wherein the intermediary server is connected to an external social media content interface platform and an external multimedia content interface platform through the network, wherein the external social media interface The content interface platform refers to an application programming platform that collects community multimedia data in a plurality of social media websites and processes the data structure program, and the external multimedia content interface platform refers to collecting multimedia content in the website. An application programming platform for the metadata of the multimedia content file; an analysis server, the analysis server is connected through the intermediary server to collect the external social media content interface platform and the multimedia content interface platform Group media content data and multimedia content data, and according to one of the usage intervals, an event analysis algorithm and a theme optimization algorithm are used to calculate social media content data and multimedia content to generate an analysis content; a community and multimedia content library The community and multimedia content library are used to store external social media. The social media content material of the content interface platform, the multimedia content material of the multimedia content interface platform, the analysis content or a service content; and a web application server, the web application server generates the analysis server The analysis content provides the service content according to a specific format corresponding to an individual external multimedia content user. The multimedia content classification system according to claim 1, wherein the attention event analysis algorithm is an algorithm for selecting a attention event with high attention, and the main reference parameters of the attention event analysis algorithm are individual The use area of the external multimedia content user and the use time interval, wherein the program of the event analysis algorithm includes the following steps: the analysis server obtains the use of the terminal device of the individual external multimedia content user through the mediation server The analysis server obtains the social media content data and the multimedia content data from the external community media content interface platform and the external multimedia content interface platform through the mediation server; the analysis server stores the social media content The content of the film and the contents of the social event data and the content of the social event to the topic data table of the film and the event content table; the analysis server obtains the number of content topics, the area and the billboard of the content of the external external multimedia content user Type information; the analysis server finds events from the event content table that match the area in which the external external multimedia content users frequently use and the type of the viewing version; the analysis server will exclude individual external multimedia content users from the benefit Category events and after exclusion The steps of obtaining the content topic number, the area and the kanban type information frequently used by the individual external multimedia content users are performed; the analysis server performs the attention event analysis, and the attention event analysis system uses the popular value growth ratio and the time relationship value as reference. The parameter calculates a attention weight value for the complex attention event, and the attention weight value is a calculation result of the value of the popular value growth ratio divided by the time relationship value, wherein the growth ratio of the popular value in the reference parameter is transmitted through the TF- The IDF algorithm operation focuses on the popularity value of the event into a length, and the time relationship value is a time difference between the time point at which the attention event and the attention event analysis program are executed; and the analysis server according to the attention level of each attention event Weight value ordering Select each of these attention events with a high degree of attention. The multimedia content classification system according to claim 2, wherein the theme optimization algorithm is to calculate a theme set suitable for the user according to the variability of the platform between the external multimedia content users. The program of the theme optimization algorithm includes the following steps: the analysis server obtains the lower limit number of the audio and video content of the individual external multimedia content user; the analysis server filters out the sorting priority from the attention event weight value in the attention event The number of ranking events of interest; the analysis server classifies the filtered attention events according to the topic of the sheet and the subject of the event; the analysis server determines whether the number of the collection topics is not empty, and whether the lower limit of the content of the audio and video content is greater than or equal to a certain value, If not, the attention event is filtered out from the attention attention weight value again by the other ranking priority events; and the theme set selected by the analysis server and determined by the attention weight value according to the attention weight value Sort. For example, in the multimedia content classification system described in claim 3, wherein the calculation of the popularity value growth ratio in the event analysis reference parameter is performed by the analysis server for extracting the social media content data and the multimedia content data. The individual vocabulary appears in the content of each social media or multimedia classification kanban, in the message content, in the kanban classification name, and in the TF-IDF value of each individual vocabulary in the collection of search keywords, wherein the TF value is calculated as follows: Where D represents the total number of all social media content materials and multimedia content data files, d _j represents individual files in D , t _i represents individual words, Σ _j count ( t _i , d _j ) represents t _i appears in d _j The number, Σ _k Σ j count ( t _k , d _j ) represents the sum of the occurrences of all words in d _j ; where the IDF value is calculated as follows: Where | t _i { j : t _i d _j} | i representative of the individual words present in all social media content information and the number of files that appear in the total number of data of the multimedia content; wherein the TF-IDF value calculation formula as follows _{:( tf - idf) i = tf} i × idf _i . The multimedia content classification system according to any one of claims 3 or 4, wherein the time relationship value in the attention event analysis reference parameter is a time point of the year, month, day, minute, minute, and second when the individual attention event is issued. Pay attention to the difference between the time point of the year, month, day, minute, and second when the event analysis program is executed. The value of the time relationship value is as follows: f ( a , b , c , d , e , f )= a _i λ _year + b _i λ _month + c _i λ _day + di λ _hour + eiλ _min + fiλ _sec + C ; where f (a, b, c, d, e, f) 1, C is a nonzero constant, i representative of the sequence number of individual events of _{interest, a i, b i, c} i, d i, e i, minutes and seconds of the absolute value when the difference f _i each represent sequential date; wherein , λ _year = 10 ¹⁰ , λ _month = 10 ⁸ , λ _day = 10 ⁶ , λ _hour = 10 ⁴ , λ _min = 10 ² , λ _sec = 10 ⁰ . A multimedia content classification method includes the following steps: an analysis server serializes through an intermediary server to collect social media content data and multimedia content of an external social media content interface platform and a multimedia content interface platform, and The social media content data and the multimedia content are calculated according to one of the usage intervals and the theme optimization algorithm to generate an analysis content, wherein the external social media content interface platform refers to collecting plural society An application programming platform for community multimedia materials and data structure processing in a group media website, and the external multimedia content interface platform refers to an application for collecting metadata data of multimedia content files in a plurality of multimedia content websites The program interface platform, wherein the interest rate growth reference ratio in the attention event analysis reference parameter is used by the analysis server to extract the social media content data and the multimedia content data, and the individual words appear in each social media or multimedia category In the content of the board, in the message content, and in the board Names TF-IDF value and a set of search keywords for each word of the individual, wherein the TF value calculation formula as follows: Where D represents the total number of all social media content materials and multimedia content data files, d _j represents individual files in D , t _i represents individual words, Σ _j count ( t _i , d _j ) represents t _i appears in d _j The number, Σ _k Σ _j count ( t _k , d _j ) represents the sum of the occurrences of all words in d _j ; where the IDF value is calculated as follows: Where | t _i { j : t _i d _j} | i representative of the individual words present in all social media content information and the number of files that appear in the total number of data of the multimedia content; wherein the TF-IDF value calculation formula as follows _{:( tf - idf) i = tf} i × idf _And a web application server provides the service content generated by the analysis server to a multimedia content user corresponding to an individual external content according to a specific format. The multimedia content classification method according to claim 6, wherein the theme optimization algorithm is to calculate a theme set suitable for the user according to the variability between the platforms of the individual external multimedia content users, wherein The program of the theme optimization algorithm includes the following steps: the analysis server obtains the lower limit number of the audio and video content of the individual external multimedia content user; the analysis server filters out the sorting priority from the attention event weight value in the attention event The number of ranking events of interest; the analysis server classifies the filtered attention events according to the topic of the sheet and the subject of the event; the analysis server determines whether the number of the collection topics is not empty, and whether the lower limit of the content of the audio and video content is greater than or equal to a certain value, If not, the attention event is filtered out from the attention attention weight value by the other attention rankings; and the theme set of the attention event that the analysis server will select and pass the judgment Sort by the weight of interest. The multimedia content classification method according to any one of claims 6 to 7, wherein the time relationship value in the attention event analysis reference parameter is a time point of the year, month, day, minute, minute, and second when the individual attention event is issued. Pay attention to the difference between the time point of the year, month, day, minute, and second when the event analysis program is executed. The value of the time relationship value is as follows: f ( a , b , c , d , e , f )= a _i λ _year + b _i λ _month + c _i λ _day + d _i λ _hour + e _i λ _min + f _i λ _sec + C ; where f (a, b, c, d, e, f) 1, C is a non-zero constant, i represents the number of individual attention event numbers, a _i , b _i , c _i , d _i , e _i , f _i respectively represent the absolute value of the difference between the year, month, day, hour, minute and second; , λ _year = 10 ¹⁰ , λ _month = 10 ⁸ , λ _day = 10 ⁶ , λ _hour = 10 ⁴ , λ _min = 10 ² , λ _sec = 10 ⁰ .