TWI697789B - Public opinion inquiry system and method - Google Patents

Abstract

A public opinion inquiry system and a public opinion inquiry method are provided. In some embodiments, an internet module for video capture settings sets the rules for searching primary articles about public opinion related to a keyword, and sets the rules for searching videos about public opinion related to the keyword. Then, a module of text integration and arrangement locates supplementary articles about public opinion based on the text in subtitle regions of the videos about public opinion, and a module for optimizing the index of public opinion combines the primary articles and supplementary articles about public opinion as a result of public opinion inquiry.

Description

Public opinion query system and method

The present invention relates to a public opinion query technology, in particular to a public opinion query system and method for searching public opinion articles through image comparison.

In the traditional public opinion query system, usually users will first enter the keywords of the articles to be searched for public opinion (such as tax reform system, American cattle imports, etc.), and then the query system will find whether there is public opinion containing the keyword in the database Articles, and then provide the users with the found public opinion articles.

However, such public opinion query system has many problems. First of all, the number of public opinion articles found only by whether the content of the article contains the keyword is usually not enough, which is far from enough for users who want to deeply understand the public opinion content of a specific topic (keyword).

Secondly, because users usually only use a single keyword to query in the public opinion query system, and cannot enter all keywords related to a specific topic in the public opinion query system in an exhaustive manner, the system cannot provide complete information when querying. , All-round public opinion articles.

For example, when a user is interested in a public opinion article whose topic is the tax reform system, he enters "tax reform system" in the search field of the public opinion query system. However, in fact, the keywords related to the "tax reform system" may include "tax reform", "tax system change", "tax reform alliance", "tax revision law", etc. At this time, if you only enter the "tax reform system" query, it is basically impossible to find complete public opinion articles related to the "tax reform system" in the general public opinion query system.

In view of the aforementioned problems, it is really necessary to provide an effective public opinion query system and method to improve the difficulties that users face when using the public opinion query system.

Based on the problems in the prior art, the present invention discloses a public opinion query system and method that can provide relatively complete public opinion articles. Compared with the prior art, an embodiment of the present invention discloses a combination of a text recognition module, a merged subtitle module, a text integration and sorting module, and a public opinion index optimization module to search for and create a complete public opinion article. For users to query and understand more complete public opinion content.

An embodiment of the present invention provides a public opinion query system, which includes: a network video capture setting module for setting a first rule and a second rule, wherein the first rule is in plural public opinion articles The rules for searching for the first group of public opinion articles related to a keyword, and the second rule is the rule for searching public opinion images in plural images, the public opinion images are images related to the keyword; a text recognition module , Is to perform text recognition on the public opinion images searched according to the second rule; a combined subtitle module is based on the results of text recognition of the public opinion images to merge the text parts of the public opinion images , To serve as the subtitle area in the public opinion images; a text integration and sorting module is used to perform correlation comparison between the text in the subtitle area and the plural public opinion articles to find the second group of public opinion articles; And a public opinion index optimization module is used to combine the first group of public opinion articles with the second group of public opinion articles to generate a third group of public opinion articles.

In another embodiment, the plural public opinion articles are public opinion articles downloaded from the Internet by a public opinion capturing module and/or stored in a public opinion article database.

In another embodiment, the public opinion images are public opinion images downloaded from the Internet by a video download module and/or stored in an image database.

In another embodiment, the first rule is to use the article name or article content of the plural public opinion articles containing the keyword as the first group of public opinion articles.

In another embodiment, the second rule is to use the file names of the plural images containing the keyword as the public opinion images.

In another embodiment, the text recognition module uses optical text recognition (OCR) to perform text recognition on the public opinion images.

In another embodiment, the combined subtitle module uses a marquee discriminating method to divide the plural text areas in the public opinion images into marquee regions and main subtitle regions, so that the main subtitle regions and/or the The marquee area is used as the subtitle area in the public opinion images.

In another embodiment, the marquee determination method is that the combined subtitle module determines whether the content and position of the text in the plural text areas in the public opinion images will change over time.

In another embodiment, the text integration and sorting module performs the following operations to compare the text in the subtitle area with the plural public opinion articles to find the second group of public opinion articles: the text integration The text integration and ranking module converts the plural public opinion articles and the text in the subtitle areas into numerical data in a numerical manner; the text integration and ranking module uses the cosine angle algorithm to target the numerical data of the plural public opinion articles And the numerical data of the text in the subtitle areas to obtain the similarity score between the plural public opinion articles and the text in the subtitle areas; and the text integration and ranking module finds the plural public opinion articles The public opinion articles whose similarity score is greater than a preset value are selected as the second group of public opinion articles.

In another embodiment, the public opinion index optimization module combines the first group of public opinion articles with the second group of public opinion articles to generate the third group of public opinion articles.

Another embodiment of the present invention provides a public opinion query method, including the following steps: (1) searching for a first group of public opinion articles related to a keyword among plural public opinion articles; (2) searching for public opinion images among plural images, Among them, the public opinion images are images related to the keyword; (3) perform text recognition on the public opinion images to find the subtitle areas in the public opinion images; (4) the text in the subtitle areas Perform correlation comparison with the plural public opinion articles to find the second group of public opinion articles; and (5) combine the first group of public opinion articles with the second group of public opinion articles to generate a third group of public opinion articles.

In another embodiment, the step (1) is to find a public opinion article containing the keyword among the plural public opinion articles, as the first group of public opinion articles.

In another embodiment, the step (2) is to find images containing the file name of the keyword in the plural images as the public opinion images.

In another embodiment, the step (3) uses optical character recognition (OCR) to perform text recognition on the public opinion images, find the text areas in the public opinion images, and use them as the public opinion images. Subtitle area.

In another embodiment, in this step (3), a marquee discrimination method is used to divide the plural text areas in the public opinion images into marquee areas and main subtitle areas, so that the main subtitle areas and / Or the marquee area as the subtitle area in the public opinion images.

In another embodiment, the marquee determination method is to determine whether the content and position of the text in the plural text areas in the public opinion images will change over time.

In another embodiment, the step (4) includes the following steps: (4-1) convert the plural public opinion articles and the text in the subtitle areas into numerical data in a numerical manner; (4-2) Use the cosine angle algorithm to calculate the numerical data of the plural public opinion articles and the numerical data of the text in the subtitle areas to obtain the similarity scores between the plural public opinion articles and the text in the subtitle areas; and (4-3) In the plural public opinion articles, find public opinion articles whose similarity score is greater than a preset value, and use them as the second group of public opinion articles.

In another embodiment, the step (5) is to combine the first group of public opinion articles with the second group of public opinion articles to generate the third group of public opinion articles.

It should be understood that the subject matter described above can be implemented as a computer-controlled device, computer program, computing system, or as a product, such as a computer-readable storage medium.

In order to make the above-mentioned features and advantages of the present invention more comprehensible, embodiments are specifically described below in conjunction with the accompanying drawings. In the following description, the additional features and advantages of the present invention will be partially explained, and these features and advantages will be partly obvious from the description, or can be learned by practicing the present invention. The features and advantages of the present invention are realized and achieved by means of the elements and combinations specifically pointed out in the scope of the patent application. It should be understood that the foregoing general description and the following detailed description are both illustrative and explanatory, and are not intended to limit the claimed scope of the present invention.

100‧‧‧Video Database

101‧‧‧Network Video Capture Setting Module

102‧‧‧Video download module

103‧‧‧Sampling setting module

104‧‧‧Sampling and slice module

105‧‧‧Video and text query module

106‧‧‧Subtitle block setting module

107‧‧‧Text recognition module

108‧‧‧Combined subtitle module

109‧‧‧Public opinion capture module

110‧‧‧Text Integration and Sorting Module

111‧‧‧Public Opinion Index Optimization Module

112‧‧‧Public opinion article database

113‧‧‧Public Opinion Inquiry System

B1, B2, R1, R2‧‧‧Subtitle area

Figure 1 is a schematic architecture diagram of a public opinion query system according to an embodiment of the present invention; and Figures 2 to 4 are schematic diagrams of public opinion images after slicing according to an embodiment of the present invention.

The following describes the implementation of the present invention with specific specific embodiments. Those familiar with this technology can easily understand the other advantages and effects of the present invention from the contents disclosed in this specification, and can also be implemented by other different specific embodiments. Or apply.

The present invention discloses a public opinion query system and method. Now referring to FIG. 1, which illustrates a schematic architecture diagram of a public opinion query system according to an embodiment of the present invention. The public opinion query system 113 mainly includes a video database 100, a network video capture setting module 101, a video download module 102, a sampling setting module 103, a sampling slicing module 104, a video and audio text query module 105, and a subtitle block setting Module 106, text recognition module 107, combined subtitle module 108, public opinion capture module 109, text integration and sorting module 110, public opinion index optimization module 111 and/or public opinion article database 112.

As shown in the figure, the video database 100 is a database that can store image data. The image data includes, but is not limited to, visual or multimedia data such as images, movies, and audio-visual data. In contrast, the public opinion article database 112 is a database for storing public opinion articles, where the content of the public opinion articles includes, but is not limited to, public opinion and speech on social events, political elections, and educational system reforms.

The network video capture setting module 101 allows users to set rules for searching public opinion articles related to specific keywords (such as "telecommunications", etc.) and images related to keywords (referred to as public opinion images).

For example, users can set the specifications for searching public opinion articles, which means searching for articles whose article name or article content contains keywords. In addition, users can also set rules for searching public opinion images, which means searching for images whose file names contain keywords.

It should be noted that the user's rule setting in the web video capture setting module 101 is only illustrative. In fact, the rule setting of the user in the network video capture setting module 101 can be increased, decreased or modified according to user needs.

In addition, after the network video capture setting module 101 has set the search rules of the public opinion image, it can directly search the public opinion image from the video database 100. Alternatively, the network video capture setting module 101 can also send the keywords (search rules) to the video download module 102, and then the video download module 102 searches and downloads public opinion images on the Internet according to the keywords (search rules) .

For example, after receiving a keyword, the video download module 102 can directly search for and download public opinion images using the keyword (such as "telecommunications") as the file name in an image search engine such as youtube.

On the other hand, after the network video capture setting module 101 has set the search rules of the public opinion articles, it can directly search the public opinion articles from the public opinion article database 112. Alternatively, the web video capture setting module 101 can send the keywords (search rules) to the public opinion capture module 109, and then the public opinion capture module 109 searches and downloads on the Internet according to the keywords (search rules) Public opinion articles.

For example, after receiving keywords, the public opinion capture module 109 can directly publish articles on news websites, forums, current affairs discussion platforms, online message boards, and other online platforms to find the name of the article or the content of the article. Keyword articles are used as search results of public opinion articles (hereinafter referred to as main public opinion articles).

After searching and downloading the public opinion image, the sampling setting module 103 can set the frequency of video slicing (sampling) for the public opinion image. For example, the sampling setting module 103 can be configured to only take samples within a specific time period (e.g., from 6 am to 3 pm on April 20, 2007, 6:30 am to 6:40 am every Monday, The first 5 minutes of each public opinion video). It should be noted that the aforementioned frequency setting is only illustrative, and not intended to limit the protection scope of the present invention.

For example, the sampling setting module 103 sets the frequency of video slices (sampling) of public opinion images about the keyword "telecommunications" downloaded by the video download module 102 from youtube to a certain period of time from November 24, 2017 Public opinion images are sliced 15 times per second.

Thereafter, the sampling and slicing module 104 can slice the public opinion image according to the set frequency of slicing (sampling) the public opinion image and record the time stamp of each sliced video segment. The time stamp format of the sliced video may be a date. + Time + random number + slice order .

For example, the format of the time stamp of the sliced video may be " 20171124080808-824791419114957289-01, 20171124080808-824791419114957289-02, ... 20171124080808-10071954288746-15 ".

For the content of the sliced movie (image), please refer to FIGS. 2 to 4, which illustrate a schematic diagram of a public opinion image after sliced according to an embodiment of the present invention.

Then, the subtitle block setting module 106 can set and mark the subtitle block of the sliced public opinion image.

For example, the user can directly use the subtitle block setting module 106 to set and mark the subtitle block (region) of interest in the sliced public opinion image. Alternatively, the text recognition module 107 can also recognize the text in the public opinion image, and then the subtitle block setting module 106 automatically sets and marks the subtitle area of interest in the public opinion image according to the result of the text recognition.

For example, the subtitle block setting module 106 can combine the recognized text in the sliced video (image) shown in Figures 2 to 4 according to the time stamp, and perform subtitle area for the sliced video , For example: the subtitle area at the bottom (B1, B2) and the subtitle area at the right (R1, R2)

In one embodiment, the text recognition module 107 can use optical text recognition (OCR) to perform text recognition on the public opinion images.

For example, for the sliced video in Figures 2 to 4, the text recognition module 107 can recognize the text in the subtitle area R1, R2, B1, B2 in the figure, and the recognition result is as follows: " R1 ₁ : Go to the streets! Anti-nuclear anger! Labor grievances

R1 ₂ : Go to the streets! Nuclear anger again! Hard work

R1 ₃ : Soil street! Anti-nuclear anger! Labor grievance

R2 ₁ : The government's next big landmine prediction

R2 ₂ : The government's prediction of the next landmine

R2 ₃ : Prophecy of the next big landmine in Zhengyoufu

B1 ₁ : Subsidiary established by myself

B1 ₂ : Earned a lot of money and became the parent company Golden Rooster

B1 ₃ : But the salary adjustment of employees is not proportional to the money earned

B2 ₁ : The ring regiment took to the streets, the anti-nuclear mobilization

B2 ₂ : Choking low salary! Counter dispatch! 20,000 people encircle the Ministry of Labor on Labor Day

B2 ₃ : choking low salary! Counter dispatch! 20,000 people encircle the Ministry of Labor on Labor Day "

Then, the subtitle merging module 108 can merge the text part (text area) in the public opinion image according to the result of the text recognition of the video after slicing to generate the subtitle area in the public opinion image.

來代表合併字幕模組108所產生的輿情影像中之字幕區域，其中，x代表特定時間的切割點，C_x代表特定時間的切割點下所產生輿情影像中的字幕區域，而符號U代表聯集。 In one embodiment, the formula can be used

To represent the subtitle area in the public opinion image generated by the combined subtitle module 108, where x represents the cutting point at a specific time, C _x represents the subtitle area in the public opinion image generated at the cutting point at a specific time, and the symbol U represents the joint set.

Take the sliced video in Figures 2 to 4 as an example, the subtitle area C _x in the public opinion image is equal to {The subsidiary I established by myself has made a lot of money to become the parent company's Golden Rooster, but the salary adjustment of the employees is not the same. Money disproportionate|The ring group took to the streets, and the anti-nuclear mobilization choke low salary! Counter dispatch! On Labor Day, 20,000 people angered the Ministry of Labor to choke low wages! Counter dispatch! Twenty thousand people encircle the Ministry of Labor on Labor Day|Take the streets! Anti-nuclear anger! Labor complaints go to the streets! Nuclear anger again! Hard work and resent the streets! Anti-nuclear anger! Labor grievances|The government's next big landmine is predicted.

In one embodiment, the combined subtitle module 108 can use the main subtitle analysis method or the marquee discrimination method to divide the text area in the public opinion image into a marquee area and a main subtitle area, so as to divide the main subtitle area and/or the marquee area. The area is used as the subtitle area in the public opinion image.

In one embodiment, the main subtitle analysis method may be: in a text area in a public opinion image, if a character string stays in the text area for a shorter time and the frequency of subtitles is changed, the text area is a public opinion The greater the possibility of the main subtitle area of the video.

Taking the aforementioned text area (R1, R2, B1, B2) as an example, because the content of the string that appears three times in R1, R2 is the same, it is less likely that R1, R2 may be the main subtitle area of the public opinion image.

In contrast, the content of the string that appears three times in B1 and B2 is not the same, so the possibility that B1 and B2 may be the main subtitle area of the public opinion image is relatively high.

After the combined subtitle module 108 determines that the B1 and B2 blocks are more likely to be the main subtitle area, the formula M _x can be used to find the most likely main subtitle area in the B1 and B2 blocks, and the symbol BM is used to indicate the main subtitle area. Subtitle area. Formula M _{x M x = Max (Main_Text (B} i) | i = 1 to 2) ( i.e., the B1, B2 block, whichever is the highest likelihood based subtitle region, as a main subtitle area) .

At this time, it can be calculated that BM=B1, that is, in the aforementioned text area, the area most likely to be the main subtitle is B1={The subsidiary that I established by myself has made a lot of money to become the parent company’s Golden Rooster but employees The salary adjustment is not proportional to the money earned}.

Then, in addition to the B1 area as the main subtitle area, the merged subtitle module 108 can also treat the remaining areas (ie, B2, R1, R2) as non-main subtitle areas (such as the marquee area), and target non-main subtitle areas. In the subtitle area, typos are corrected and repeated words are combined to generate another set. After that, the text collection in the main subtitle area and the text in the non-main subtitle area after correction are combined in a united manner.

={自己出去成立的子公司 賺了非常多錢 成為母公司金雞母 但員工薪水調幅卻跟賺的錢不成比例；環團上街頭，反核總動員 嗆低薪！反派遣！勞動節兩萬人怒圍勞動部|上街頭！反核怒！勞工怨|政府下個大地雷預言}。 For example, the Cx of the aforementioned B1, B2, R1, and R2 regions can be updated to

={The subsidiary I set up by myself made a lot of money to become the parent company Golden Rooster, but the salary adjustment of the employees was not proportional to the money earned; the ring group took to the streets, and the anti-nuclear general mobilization choked low salary! Counter dispatch! Twenty thousand people encircle the Ministry of Labor on Labor Day|Take the streets! Anti-nuclear anger! Labor grievances|The government's prediction of a big landmine}.

In addition, in one embodiment, the marquee determination method refers to whether the content and position of the text in the text area of the public opinion image of the combined subtitle module 108 will change over time.

For example, in a text area in a public opinion image, if a character string is constantly shifting in the text area (that is, the position changes continuously), but the content of the string does not change, then the text area may be a marquee area The probability is very high.

After determining the marquee area and the main subtitle area in the public opinion image, the combined subtitle module 108 can use the main subtitle area and/or the marquee area as the subtitle area in the public opinion image.

The text integration and sorting module 110 can combine the text in the subtitle area in the public opinion image with (i) public opinion articles downloaded from the public opinion capturing module 109, (ii) public opinion articles stored in the public opinion article database 112, or ( iii) Perform correlation comparison of the public opinion articles downloaded from the public opinion capture module 109 and stored in the public opinion article database 112 (the public opinion articles mentioned in (i)~(iii) above are referred to as candidate public opinion articles), Find more public opinion articles that meet user needs.

In one embodiment, the text integration and sorting module 110 may first use a numerical method to convert the text in the subtitle area of the candidate public opinion article and public opinion image into numerical data.

For example, take the text "China Power Corporation" in the public opinion article and the text "Company China Power" in the subtitle area as an example. If you use (中，华，电,公司,司, China, Huadian, Diangong, Company, Division) is used as the standard for the digitization of words and sentences (here, 10-dimensional vectorization), and it is assumed that if the words shown in a certain dimension appear in the words and sentences (such as: "China", "Huadian", "Sizhong", etc. ), the value of the phrase in that dimension is 1; otherwise, it is 0.

In this way, the result of the digitization of the phrase "Chonghua Power Company" is (1,1,1,1,1,1,1,1,1,0), and the result of the digitization of the phrase "China Power Company" is (1 ,1,1,1,1,1,1,1,1,1).

Then, the text integration and ranking module 110 uses a correlation comparison algorithm (such as the cosine angle algorithm) to calculate the numerical data of the candidate public opinion articles and the text in the public opinion image caption area to calculate the candidate public opinion articles and The similarity score of the text in the subtitle area of the public opinion image.

Take the aforementioned terms "Chonghua Power Company" and "Company Chunghwa Power" as an example, the cosine angle of the numerical result of the two words is (1 * 1+1 * 1+1 * 1+1 * 1+1 * 1+1 * 1+1 * 1+1 * 1+1 * 1+0 * 1)/10=0.9, so the calculated similarity score is 0.9.

Then, the text integration and sorting module 110 finds articles whose similarity scores with the text in the subtitle area of the public opinion image are greater than a preset value (such as 0.8, 0.7, etc.) among the candidate public opinion articles, as supplementary public opinion articles article.

The public opinion index optimization module 111 can combine the aforementioned main public opinion articles directly obtained from the keyword search and the supplementary public opinion articles generated by the text integration and ranking module 110 to generate the public opinion articles finally presented to the user.

In one embodiment, the public opinion index optimization module 111 combines the main public opinion articles and supplementary public opinion articles to generate the public opinion articles that are finally presented to the user.

In addition, a cumulative algorithm or an average algorithm can be used to evaluate the number and quality of the articles generated by the public opinion index optimization module 111.

，其中，主要輿情文章得分數為M×P_(hits)；補充式輿情文章的分數為

；M代表單一篇輿情文章的分數(如：前述計算出的相似度分數)；P代表利用關鍵字搜尋到的輿情文章數；r|s代表利用關鍵字所搜尋到的輿情影像文字組；因其位置s不同而有不同權重r，一影片字幕總和值為

，Q _(hits) 代表利用關鍵字搜尋到的輿情影像文字組。此外，平均型演算法可如下所示：主要輿情文章和補充式輿情文章結合後呈現給使用者的輿情文章之分數可設定為

，其中，主要輿情文章得分數為

，M代表單一篇輿情文章的分數；P代表利用關鍵字搜尋到的輿情文章數；base代表要除之數(如篇數)；其中r|s代表利用關鍵字所搜尋到的輿情影像文字組；因其位置s不同而有不同權重r，一影片字幕總和值為

。將此總和值乘以代表利用關鍵字搜尋到的輿情影像文字組Q_(hits)，與M×P_(hits)相加後，除以輿情篇數base|P與影 片篇數base|Q。 For example, the cumulative algorithm can be as follows: The score of the public opinion article presented to the user after the combination of the main public opinion article and the supplementary public opinion article can be set as

, Where the score of the main public opinion article is M×P _{( hits )} ; the score of the supplementary public opinion article is

; M represents the score of a single public opinion article (such as the similarity score calculated above); P represents the number of public opinion articles searched by keywords; r | s represents the public opinion image text group searched by keywords; Its position s has different weight r, the sum of subtitles of a movie is

, Q _(hits) represents the public opinion image text group searched by keywords. In addition, the average algorithm can be as follows: The score of the public opinion article presented to the user after the combination of the main public opinion article and the supplementary public opinion article can be set as

, Where the scores of main public opinion articles are

, M represents the score of a single public opinion article; P represents the number of public opinion articles searched by keywords; base represents the number to be divided (such as the number of articles); where r | s represents the public opinion image text group searched by keywords ; Because of its different position s, there are different weights r, the sum of subtitles of a movie is

. Multiply this total value by the Q _{( hits )} of the public opinion image text group searched by keywords, add M×P _{( hits )} , and divide by the number of public opinion articles base|P and the number of videos base|Q.

Take the cumulative algorithm as an example, but not limited to this. Assuming that the score of each main public opinion article is 1, and 100 articles are searched by keywords, the total score of the main public opinion articles is 1 x 100=100. In addition, in the supplementary public opinion article, suppose that the keyword hits B1=1 (that is, the similarity score of the B1 area is 1 point), there are 20 articles, and it hits L1=0.5 (that is, the similarity score of the L1 area is 0.5 If there are 10 articles, 10 articles with R1=0.2, and 10 articles with T=0.1, the score of the final public opinion article presented to the user is 100+20+5+2+1=128 points.

In this example, originally only relying on keywords to search for the main public opinion article score is 100 points; but using this public opinion query system 113 to search, the final public opinion article presented by the score is 128 points, obviously using this public opinion query system The search of public opinion articles can find more complete and diverse public opinion articles than general public opinion article search systems.

Finally, the audiovisual text query module 105 allows the user to query public opinion images and public opinion articles related to keywords.

The above embodiments are only illustrative of the principles, features and effects of the present invention, and are not intended to limit the scope of implementation of the present invention. Anyone who is familiar with the art can comment on the above without departing from the spirit and scope of the present invention. Modifications and changes to the implementation form. Any equivalent changes and modifications made by using the contents disclosed in the present invention should still be covered by the scope of the patent application. Therefore, the protection scope of the present invention should be as listed in the scope of patent application.

100‧‧‧Video Database

101‧‧‧Network Video Capture Setting Module

102‧‧‧Video download module

103‧‧‧Sampling setting module

104‧‧‧Sampling and slice module

105‧‧‧Video and text query module

106‧‧‧Subtitle block setting module

107‧‧‧Text recognition module

108‧‧‧Combined subtitle module

109‧‧‧Public opinion capture module

110‧‧‧Text Integration and Sorting Module

111‧‧‧Public Opinion Index Optimization Module

112‧‧‧Public opinion article database

113‧‧‧Public Opinion Inquiry System

Claims (18)

A public opinion query system, comprising: a network video capture setting module for setting a first rule and a second rule, wherein the first rule is to search for a keyword related to a plurality of public opinion articles The rules of the first group of public opinion articles, and the second rule is the rule of searching public opinion images in plural images. The public opinion images are the images related to the keyword; a text recognition module will be based on the second rule The searched public opinion images perform text recognition; a combined subtitle module is based on the results of the text recognition of the public opinion images to merge the text parts in the public opinion images to be used in the public opinion images The subtitle area of the subtitle area, wherein the combined subtitle module uses a marquee discrimination method to divide the plural text areas in the public opinion image into a marquee area and a main subtitle area, so that the main subtitle area and/or the running lights The marquee area is used as the subtitle area in the public opinion images; a text integration and sorting module is used to perform correlation comparison between the text in the subtitle area and the plural public opinion articles to find the second group of public opinion articles; And a public opinion index optimization module is used to combine the first group of public opinion articles with the second group of public opinion articles to generate a third group of public opinion articles. For example, in the public opinion query system described in the first item of the scope of patent application, the plural public opinion articles are public opinion articles downloaded from the Internet by a public opinion capturing module and/or stored in a public opinion article database. For example, in the public opinion query system described in item 1 of the scope of patent application, the public opinion images are public opinion images downloaded from the Internet by a video download module and/or stored in an image database. For example, in the public opinion query system described in item 1 of the scope of patent application, the first rule is to regard the article name or article content of the plural public opinion articles containing the keyword as the first group of public opinion articles. For example, in the public opinion query system described in item 1 of the scope of patent application, the second rule is to use the file name of the plural images containing the keyword as the public opinion images. For example, in the public opinion query system described in item 1 of the scope of patent application, the text recognition module uses optical text recognition (OCR) to perform text recognition on the public opinion images. For example, in the public opinion query system described in item 1 of the scope of patent application, the public opinion index optimization module combines the first group of public opinion articles with the second group of public opinion articles to generate the third group of public opinion articles. For example, the public opinion query system described in item 1 of the scope of patent application, wherein the marquee judging method is determined by the combined subtitle module to determine whether the content and position of the text in the plural text areas in the public opinion images will change over time And change. A public opinion query system, comprising: a network video capture setting module for setting a first rule and a second rule, wherein the first rule is to search for a keyword related to a plurality of public opinion articles The first set of rules for public opinion articles, and the second rule is the rule for searching public opinion images in plural images. A sentiment image is an image related to the keyword; a text recognition module is used for text recognition of the public opinion images searched according to the second rule; a combined subtitle module is used for text recognition based on the public opinion images The result of the recognition is to merge the text parts in the public opinion images to serve as the subtitle area in the public opinion images; a text integration and sorting module is used to combine the text in the subtitle areas with the plural public opinion The articles are compared to find the second group of public opinion articles; and a public opinion index optimization module is used to combine the first group of public opinion articles with the second group of public opinion articles to generate the third group of public opinion articles ; Among them, the text integration and ranking module performs the following operations to compare the text in the subtitle area with the plural public opinion articles to find the second group of public opinion articles: the text integration and ranking module Convert the plural public opinion articles and the text in the subtitle areas into numerical data in a numerical manner; the text integration and sorting module uses the cosine angle algorithm to target the numerical data of the plural public opinion articles and the subtitles The numerical data of the text in the area is calculated to obtain the similarity score between the plural public opinion article and the text in the subtitle areas; and the text integration and ranking module finds the similarity in the plural public opinion article Public opinion articles with scores greater than a preset value are used as the second group of public opinion articles. Such as the public opinion query system described in item 9 of the scope of patent application, where the The public opinion index optimization module combines the first group of public opinion articles with the second group of public opinion articles to generate the third group of public opinion articles. A public opinion query method includes the following steps: (1) Searching for a first group of public opinion articles related to a keyword in plural public opinion articles; (2) Searching for public opinion images in plural images, wherein the public opinion images are related to Images related to the keyword; (3) Perform text recognition for the public opinion images to find the subtitle areas in the public opinion images, wherein a marquee discrimination method is used to divide the plural text areas in the public opinion images The marquee area and the main subtitle area, using the main subtitle area and/or the marquee area as the subtitle area in the public opinion images; (4) The text in the subtitle area is combined with the plural public opinion articles Correlation comparison to find the second group of public opinion articles; and (5) combine the first group of public opinion articles with the second group of public opinion articles to generate a third group of public opinion articles. For example, the public opinion query method described in item 11 of the scope of patent application, wherein the step (1) is to search for public opinion articles containing the keyword among the plural public opinion articles, as the first group of public opinion articles. For example, the public opinion query method described in item 11 of the scope of the patent application, wherein the step (2) is to find the images containing the file name of the keyword in the plural images as the public opinion images. For example, the public opinion query method described in item 11 of the scope of patent application, wherein the step (3) is to use optical character recognition (OCR) for the And other public opinion images perform text recognition, find out the text area in the public opinion images, and use them as the subtitle area in the public opinion images. For example, in the public opinion query method described in claim 11, the step (5) is to combine the first group of public opinion articles with the second group of public opinion articles to generate the third group of public opinion articles. For example, the public opinion query method described in item 11 of the scope of patent application, wherein the marquee judgment method is to determine whether the content and position of the text in the plural text areas in the public opinion images will change over time. A public opinion query method includes the following steps: (1) Searching for a first group of public opinion articles related to a keyword in plural public opinion articles; (2) Searching for public opinion images in plural images, wherein the public opinion images are related to Images related to the keyword; (3) Perform text recognition for the public opinion images to find the subtitle area in the public opinion images; (4) Compare the text in the subtitle area with the plural public opinion articles Yes, to find the second group of public opinion articles; and (5) combine the first group of public opinion articles with the second group of public opinion articles to generate a third group of public opinion articles; wherein, the step (4) includes the following steps: (4-1) Convert the plural public opinion articles and the text in the subtitle areas into numerical data in a numerical manner; (4-2) Use the cosine angle algorithm to target the numerical data and the plural public opinion articles The numerical data of the text in the subtitle area is entered Perform operations to obtain the similarity scores between the plural public opinion articles and the text in the subtitle areas; and (4-3) in the plural public opinion articles, find public opinion articles whose similarity scores are greater than a preset value, As the second group of public opinion articles. For example, in the public opinion query method described in the scope of patent application, the step (5) is to combine the first group of public opinion articles with the second group of public opinion articles to generate the third group of public opinion articles.

Images