CN1588879A - Internet content filtering system and method - Google Patents

Abstract

The invention relates to an Internet content filtering system and filtering method. The system framework includes three parts: Content Filtering Agent (CFA), Query Server (QS), and Content Analysis and Management Server (CAMS). The filtering process of the network content filtering system is: when a user sends a request to visit a certain URL, CFA allows or prohibits the access request according to the black and white lists set by the user. If the URL is not in CFA's black and white list, CFA will send a query request to QS. QS will query the rating information of the URL in its own URL database and return the result to CFA. CFA reacts accordingly. At the same time, QS will regularly download updated URL rating information from CAMS. The invention can accurately identify bad information existing in the network, and actively prevent Internet users from accessing these bad websites. While filtering bad information, the system will maximize the efficiency of network speed.

Description

An Internet content filtering system and filtering method

technical field

The invention belongs to the technical field of the Internet, and in particular relates to an Internet content filtering system and a filtering method, which can be used to prevent users from accessing various media data on the Internet, including text, image, video, audio, graphics and animation.

Background technique

The Internet has become an indispensable part of everyday life. People live on the Internet and enjoy all kinds of services provided by the Internet: online shopping, online banking, sending and receiving emails, information inquiries, etc. However, while people are enjoying the benefits of the Internet, there are also increasing negative effects on the Internet, such as teenagers indulging in adult websites, spreading bad information, or committing crimes through the Internet.

According to the statistics of the N2H2 company in the United States, roughly 8% of the web pages in the world are pornographic web pages. A quarter of the requests submitted to search engines every day are related to pornographic content; in addition, pornographic spam has become one of the biggest headaches for people. Generally, mainstream free mailboxes will receive 3-10 such emails every day, but the sender does not care whether the owner of the mailbox is an adult or not.

Compared with Internet pornography, there are also a plethora of website pages with anti-government and anti-society content. The words "Falun Gong" can be seen everywhere, and the so-called "government secrets" are spreading everywhere. The public's hearing and hearing are confused, and people's lives are disrupted. The negative effects of the Internet and the wide range of bad information are beyond people's expectations.

How to guarantee the operation security and information security of the Internet has aroused the general concern of the whole society. In order to promote the healthy development of my country's Internet, the Standing Committee of the National People's Congress passed a decision on maintaining Internet security in December 2000. The decision expressly stipulates: "In order to maintain national security and social stability, anyone who commits any of the following acts, which constitutes a crime, shall be investigated for criminal responsibility in accordance with the relevant provisions of the Criminal Law:

(1) Using the Internet to spread rumors, slander, or publish or disseminate other harmful information, inciting subversion of state power, overthrowing the socialist system, or inciting to split the country and undermine national unity;

(2) Stealing or leaking state secrets, intelligence or military secrets through the Internet;

(3) Using the Internet to incite ethnic hatred and discrimination and undermine ethnic unity;

(4) Using the Internet to organize cult organizations, contact members of cult organizations, and undermine the implementation of national laws and administrative regulations. "

At present, the Central Committee of the Communist Party of China and the State Council are emphasizing further strengthening and improving the ideological and moral construction of minors. The Ministry of Education also requested in May 2004 to include civilized surfing the Internet and network security knowledge as important contents of school moral education, so as to improve the ability of minors to resist harmful information.

In order to prevent the intrusion of illegal and harmful information, three methods are mainly adopted technically. One is to delete files from the server. Once the host server realizes that there is illegal information on the server, such information must be deleted from the server. The second is to block the transmission of information. If the owner or country of the server where the illegal information is located does not recognize it as illegal information or adopts an uncooperative attitude, other countries can only adopt blocking means to prohibit the retrieval of such information. The third is to develop effective filtering software. At present, three generations of filtering software have been developed. The first generation was called "blacklist" software, the second generation was "whitelist" software, and the third generation was the PICS system.

"Blacklist" software works by blocking URLs that should not be retrieved, and "whitelist" software is used to retrieve URLs that are only allowed to be accessed. "Blacklist" software is widely used in the first generation of filtering software, the most famous being Cyber Patro, which was put into use in the early 1990s and can cooperate with the retrieval software of Internet search providers and online service providers. The software recorded approximately 7,000 URLs with 12 categories of illegal and harmful information (violence/blasphemy, racism/inappropriate comments about minorities, devil worship, drugs, militant speech/extremism, gambling, etc.). "White list" is software that works just opposite to "black list". It first blocks all Internet addresses, and then selects accessible addresses. Since this software is logically opposite to the Internet, its scope of application is very limited. .

Another effective technical means to filter illegal and harmful information is to use the "Internet Content Selection Platform" (PICS-Platform for Internet Content Selection) and the "neutral labeling" system. The system, developed by Professor Jim Miller of MIT's Computer Science Laboratory, is similar to a V-chip TV program selector that filters out sex and violence from TV programs. It was officially promulgated by the World Wide Web Consortium (W3C-Wold Wide WebConsortium) in May 1996. It has been widely used at present. PICS has been widely supported by 39 international computer companies, computer software and hardware manufacturers, search service providers, online service providers, publishers, and content providers. It is installed in Internet browsers for users to choose to use. The main work of PICS is to classify the content of each webpage and add labels according to the content characteristics. At the same time, computer software monitors the labels of webpages to limit the retrieval of specific content webpages. Labels on web pages can be either numeric characters or passwords. The tags are embedded in the RFC-822 transmission format and HTML text format, and can be transmitted with the file through the HTTP protocol.

Today, many software companies are aware of the business opportunities brought by network content filtering, and various filtering software are constantly coming out. "Internet Dad", "Meiping Anti-Pornography Expert", "e Anti-Pornography Software", and "Justice Warrior" are all anti-pornography software that emerged in the early days of our country. Among the few web filtering software in China, there are some more characteristic software, such as: "MyIE2", "Don't Touch NoPorn!", "Filter Net" and "Huahuashishi" and so on. Looking at domestic filtering software, most of them use simple URL matching and keyword judgment technology to filter web pages, and there are basically no products that really use content-based analysis and processing methods to filter network media files.

In contrast, the development of similar foreign products is faster than domestic ones, and the filtration technology is relatively mature. ZyXEL, WebSense, FilterLogix, and SurfControl are widely used network content filtering software, and they all have a huge, classified URL database. Commonly used techniques are also black and white lists and keyword matching queries.

ISS's Proventia Web Filter has the world's largest and latest content filtering database. It not only relies on keyword query and manual website classification, but also uses a text image analysis system to process media content.

FortiGuard's URL database includes more than 5 million URLs and includes classification information. Whenever there is a request, the system will first query the FortiGuard database for the classification of the web page, and allow or deny the request for the web page according to the policy set in advance by the customer.

South Korea's WebWacher is a pretty good network image filtering software. The software is aimed at home users, and provides two functions of controlling online time and filtering harmful content on the Internet, so as to protect children from using the Internet reasonably.

In addition, most US companies that develop filtering software are engaged in the development of anti-spam and anti-virus software. Therefore, on the one hand, major software companies can quickly build a huge URL database for query based on their own foundation; Often only for enterprise-level users.

Contents of the invention

The purpose of the present invention is to propose a new Internet content filtering system and filtering method, so that the system has self-learning ability, and can improve the classification accuracy of the system and reduce labor costs; There are various types of media data, including: text, image, video, audio, graphics and animation, etc.

Let's first introduce the concept of URL.

URL is the abbreviation of Uniform Resource Locator (Uniform Resource Locator), and its data structure is: protocol://host name: port number/directory path/file name.

A URL corresponds to a specific data object on a website or server. For example, a URL corresponds to a portal site or a BBS server, and may also correspond to a specific picture in a directory of a site. Therefore, if you want to prevent users from accessing a certain website, server or specific data object, you only need to prevent the URL request from being sent to network users.

The protocol section describes the resource type of the Internet, such as: http means hypertext transfer protocol or WWW. Other protocols include: ftp (represents file transfer protocol), telnet (represents remote login), news (represents newsgroups), mailto (represents email), mms (represents streaming media), etc.

The host name section indicates the server name of the Internet, for example: www.fudan.edu.cn . A directory path segment indicates the location of a file or part of a file on an internet server. Each level of directory is separated by a forward slash (/) symbol.

The filename field is the actual name of the document, image or script that will be accessed, for example: index.html, logo.gif, script.cgi. Port numbers, directory paths, and file names are optional components of URLs.

Some examples of URLs are given below:

http://www.w3.org/index.html : This URL corresponds to a website

http://10.64.130.4/images/advice.gif : This URL corresponds to an image

ftp://10.11.3.8 : This URL corresponds to an FTP server

mms://10.11.4.6/abc.avi : This URL is used to order an audio-visual program

telnet: //bbs.fudan.edu.cn : This URL corresponds to a BBS server

The network content filtering system that the present invention proposes comprises following several parts (referring to Fig. 1): the content filtering proxy under the support of Internet, query server, content analysis and management server, they are located between client equipment and target site. in

1. User terminal equipment (UT: User Terminal), which can be a computer or other equipment that can access the Internet. Users access network resources through UT, such as browsing web pages, retrieving documents, downloading files, etc.

2. Content Filtering Agent (CFA: Content Filtering Agent), which stores blacklists (sites or files that are prohibited from being accessed) and whitelists (sites or files that are allowed to be accessed), which are actually a set of URL lists. The module will run on different types of platforms in many forms.

3. The query server (QS: Query Server) has a massive URL library with classification and classification information. When QS receives the URL submitted by UT, it will query in the classification and grading library and tell UT the result. The basic reason for adopting QS is that the resources of CFA are limited, so too much classification and grading information cannot be stored, and only a small amount of black/white list can be stored, while a large amount of classification and grading information can be stored on QS, and one QS can support a large number of CFA concurrent access. At the same time, multiple QSs can be deployed on the Internet, and QSs can also be deployed on the Intranet of a unit to deal with a large number of concurrent query requests.

4. Content Analysis and Management Server (CAMS: Content Analysis and Management Server), its main task is to classify and classify resources in the Internet. For example, record the list of "websites or bad URLs that store pornographic pictures or audio and video". Authorized QS can download the URL library with classification and classification information from here. Usually, different enterprises or departments focus on different types of CAMS, and there may be multiple different types of CAMS. CAMS must also have management and publishing functions, and can also exist as a web portal.

5. The target site (TWS: Target Website or Server) can be any website or server that stores resources, and UT can access its public resources through the Internet.

The specific steps of the network content filtering system work are summarized as follows:

1. When a user sends a request to access a certain URL, CFA allows or prohibits the access request according to the blacklist or whitelist;

2. If the URL is not in the blacklist and whitelist of CFA, CFA will send a query request to QS;

3. QS will query the classification information of the URL in the local URL library and return the result to CFA, and CFA will respond accordingly;

4. QS will regularly download updated URL rating information from CAMS;

5. CAMS automatically searches, downloads, analyzes and processes multimedia data on the Internet, uses manual interactive labeling methods and machine automatic classification methods to classify and grade network content, and forms a classified and graded URL information base.

The Internet content filtering system proposed by the present invention can be applied to various application occasions, such as:

1. Used to prevent access to sites that are politically reactionary or endanger national security.

2. It is used to prevent access to yellow sites that affect the physical and mental health of young people.

3. Sites used to block access to e-sports games.

4. Used to prevent access to specific types of sites or resources, determined by specific application requirements.

Filter agent CFA can run on various types of software and hardware platforms in various ways, for example:

1. CFA can run on a proxy server.

2. CFA can run on the firewall.

3. CFA can run on the browser as a browser plug-in.

4. CFA can run in ADSL modem, Cable Modem, telephone line modem, ISDN PC adapter and other network access devices.

Description of drawings

Fig. 1 is a schematic diagram of the overall framework structure of the content filtering system on the Internet.

Figure 2 is a schematic diagram of the basic composition and workflow of the Content Analysis and Management Server (CAMS).

Numbers in the figure: 1 is the user terminal UT, 2 is the content filtering agent CFA, 3 is the query server QS, 4 is the content analysis and management server CAMS, and 5 is the target site TWS.

Detailed ways

The content of the present invention is further described below by giving examples.

About Content Analysis and Management Server (CAMS)

As we all know, there are various and ever-changing content on the Internet, such as text, image, video, audio, graphics, animation, dynamic web pages, Flash, etc.; from the perspective of the world, the data in the Internet network is truly massive.

CAMS should always pay attention to various ever-changing and massive multimedia data content in the Internet, and can make objective classification and classification of network content in time. This is a difficult and challenging task that requires large-scale computing and storage equipment, as well as a lot of human assistance.

The table below gives examples of ratings for the categories "Violence", "Nudity", etc. Category: violence level 0 no violence No aggressive violence, no natural or accidental violence level 1 fighting Injury or massacre to living things, injury to living objects level 2 to kill retaliatory injury in which a person or living being is injured or killed without endangering the living being level 3 bloody killing scene people killed or injured level 4 wanton, highly senseless violence Malicious and gratuitous acts of violence Category: Nude level 0 none no nudity level 1 revealing clothing revealing clothing level 2 half naked half naked level 3 frontal nudity frontal nudity level 4 provocative frontal nudity Provocative frontal nudity Category: sex level 0 none Non-sexual depictions/romance stories level 1 passionate kiss passionate kiss level 2 Clothed Sexual Touching Clothed Sexual Touching level 3 non-revealing sexual touching non-revealing sexual touching level 4 exposed sex exposed sex Category: Politics level 0 none no reactionary content level 1 Taboo content cryptic references to level 2 Generally speaking about politically sensitive content Generally speaking to politically sensitive reactionary content level 3 reactionary remarks Straight-talking reactionary remarks level 4 extremely reactionary remarks extremely reactionary remarks

It is a very important task of CAMS to automatically or semi-automatically classify and classify various data on websites or servers. What must be pointed out here is that the standards for classification and grading should be formulated, promulgated and implemented by relevant state departments.

With the classification and grading standards for network content, it is possible for different companies, units or portals to conduct grading evaluations for a certain type of data. For example, one CAMS only focuses on political content, another CAMS may only focus on pornographic content, which can generate many business opportunities.

Obviously, whether a certain type of CAMS can fully and accurately classify network data content will directly affect the accuracy of network content filtering. It is very difficult to fully rely on computer processing and analysis to automatically evaluate network content. In this invention, a method combining manual guidance and machine learning is used to guide the computer to complete the evaluation task of massive time-varying network data.

Figure 2 shows a content-based multimedia data analysis, processing and evaluation method (for a specific class, the class can be manually determined in advance), it can classify various media contents such as images, videos, audios, texts, etc., and its working steps are:

1. Extract features from various media objects. For example, extract color and color histogram from pictures, analyze image area color and texture structure, etc.; extract camera or object motion information, color information, texture information, etc. from video data; extract keywords from text, etc.

2. Manually mark some small number of objects. These human-labeled objects will be used as samples for machine learning.

3. The system learns according to the manual labeling results, obtains higher-level semantic information, and forms a knowledge base for classification.

4. Finally, the system automatically grades the vast majority of data objects that are not manually labeled, thereby greatly reducing labor costs.

In order to ensure that the machine has sufficient classification accuracy, it is necessary to conduct spot checks and manual evaluation of the machine classification results, that is, to further improve the classification performance of the machine through manual re-evaluation, that is, relevant feedback.

The main features of the above method are: using content-based analysis and processing methods, the understanding of various media objects enters the semantic level; introducing manual interaction and labeling, allowing machine learning to enhance the classification accuracy of the system; using a feedback mechanism, the system has Self-learning ability. Through appropriate manual guidance and machine learning methods, the accuracy of machine classification can be better improved and labor costs can be greatly reduced.

In addition, other functional modules of CAMS include: managing URL classification and grading information database; publishing URL classification and grading information database; there must also be an important module called "web crawler", which is used to automatically explore the Internet, access websites or servers, and crawl Various media files. Existing crawler software of a lot of similar functions now, this is not the focus of the present invention.

The detailed working steps of CAMS are given below (see Figure 2):

(1) Web crawler group: independently search and download various types of data from the Internet, such as web pages, pictures, videos, music, etc. [corresponding process ①]; download data objects according to suspicious URL information database requirements [corresponding process ⑦] . Note that the "suspicious URL information base" here is mainly a list of URLs sent by the query server (QS) that cannot be processed by the QS yet.

(2) Feature extraction: analyze and process all kinds of downloaded multimedia data objects, and extract features. For example, extract features such as color, texture, and shape of images; extract features of videos, such as object motion, camera motion, etc.; organize and store the URL of each downloaded data object and its features [corresponding process ②].

(3) Manual labeling: From the manually downloaded multimedia data objects, select some data objects for classification and grading labeling; manually check the results of automatic classification and grading, which can reduce errors, and can also pass this method of related feedback Improve classification performance [corresponding process ③].

(4) Training classifiers: Automatically classify and classify data objects corresponding to URLs. Machine learning methods can be used to train classifiers with manual guidance and related feedback information to obtain high-precision classification and classification results [corresponding to Process ④].

(5) Automatic classification and classification: The trained classifier can automatically classify and classify each downloaded data object, and obtain the URL information base after classification and classification [corresponding process ⑤]; the URL information base can be Regularly update and release, because the content of the Internet is changing all the time, so the update and release cycle is required to be as short as possible [corresponding process ⑥].

About Query Server (QS)

Massive URL classification and grading information databases are stored on QS, which may come from one or more CAMS. The general data structure of the URL classification and grading information base in QS is as follows (example): serial number URL (string type) "Violence" level (integer) "naked" level (integer) "political" level (integer) ... 1 URL-1 1 0 4 … 2 URL-2 2 2 0 … … … … … … … L URL-L 3 2 1 …

The main job of QS is to make a judgment on the URL submitted by the Content Filtering Agent (CFA), which is a simple look-up process. If the URL exists in the classification and grading table, QS will feed back the table lookup result (ie level) to CFA; otherwise, QS will do two things: (1) Feed back "undecidable" (NAN) information to CFA; ( 2) Submit the URL to CAMS for analysis and processing by CAMS. Since the content on the Internet changes all the time, undecidable situations cannot be avoided. If CAMS can analyze, process, and track network content changes in time, the probability of "undecidable" will be very small.

When implementing QS, you must consider supporting concurrent access. The invention adopts the URL index structure based on the Trie tree, and utilizes the main memory caching strategy to store frequently accessed URL items in the main memory of the server, and store infrequently used URL items on the disk. This strategy of using index structure and cache greatly improves the verification speed of QS and supports large concurrent access.

QS can be deployed in large numbers on the Internet or Intranet to serve various users, including home users or business users. QS will download classification and classification information bases from various authorized CAMS. CAMS should promptly process the data corresponding to URLs that QS cannot judge the results, and periodically publish classification and grading information for QS to download.

About the Content Filtering Agent (CFA)

CFA is a very simple software module, which runs on various hardware and software system platforms in various forms. A white list (WNList) and a black list (BNList) are stored in the CFA. Essentially, a black/white list is a list of URLs.

The data structure of CFA's black/white list is as follows: serial number URL (string type) property (boolean) 1 http://www.fudan.edu.cn/news/ 0 (represents belonging to the white list) 2 http://www.private.com 1 (represents belonging to the blacklist) 3 ... ...

The basic working process of CFA:

1. When the URL belongs to WNList, CFA allows the URL to pass through and forwards the URL to TWS, and TWS will return the result to UT according to the URL request.

2. When the URL belongs to the BNList, the CFA prohibits the URL from passing through, and the CFA directly sends the "Access Prohibited or Warning" message to the UT, which actually cuts off the request information of the UT.

3. When the URL does not belong to WNList or BNList, CFA sends the URL to QS, requests QS to verify the URL, and handles accordingly according to the verification result.

The above operation details are described in more detail in the follow-up workflow.

Each CFA will have an authorized account. Authorized users can set various CFA options through the graphical interface of the client to form their own filtering strategies, including:

1. The setting of the URL classification level to determine that the URL belongs to the black/white list

For example, assume that the user sets "violence" level 1 or higher (including level 1), and "naked" level 2 or higher (including level 2) URLs into the blacklist. When UT requests to access a URL that is not in CFA's black/white list, CFA sends the URL to QS. Assume that in the QS classification and rating database, the rating information of this URL is "violence" level 0, "nude" level 3, when QS returns this rating information to CFA, CFA will determine that the URL is black according to user settings. list to block the URL.

2. When the information returned by QS is "NAN", determine the attribute setting of the URL

Assuming that the user sets this option to "white list", when QS returns "NAN" to CFA, CFA will automatically determine that the URL belongs to the white list; otherwise, it will be considered as black list.

3. Users can manually manage the black/white list in CFA, including browsing, adding and deleting.

4. The user can modify the password of the authorized account in CFA.

When the storage resources of CFA are limited, a certain caching strategy needs to be adopted, such as keeping the latest and most frequently used black and white lists.

Computational power and storage resources of CFA are usually limited. For example, CFA runs in the modem (MODEM) of ADSL, and at this time the computing power is obviously insufficient, and the black/white list that can be stored is also quite limited. For this application, the CFA must be designed to be simple, compact and fast. Apparently, the CFA proposed by the present invention does not require complex programs, but is only a process of table lookup and cache maintenance, and the cache mechanism greatly reduces the demand for storage space.

Finally, it should be pointed out that the communication between CFA, QS and CAMS can be realized through Socket programming, or through other methods. The communication between CFA and QS, and between QS and CAMS must pass identity verification. The concrete steps of Internet content filtering among the present invention are as follows (seeing as shown in Figure 1)

1. When a user wants to visit a target site or server (TWS) for web browsing, video on demand or file download, an http (or ftp, mms, telnet, etc.) request will be sent, and the content filtering agent CFA will immediately intercept the request and compare it with the URLs in the black and white list of CFA [corresponding process ①].

If the URL requested by UT is in the CFA blacklist, the URL request will be intercepted, and an error or warning message will be returned to UT [corresponding process ②].

2. If the URL requested by UT is in the CFA whitelist, the URL request will be forwarded directly to the target site TWS [corresponding process ③]; TWS will reply to UT with a corresponding response [corresponding process ⑥].

3. If the requested URL is neither in the CFA blacklist nor in the CFA whitelist, CFA sends the URL to the query server QS [corresponding process ④], QS queries the URL, obtains classification information or NAN, and sends To CFA [corresponding process ⑤].

(1) If the URL is in the URL library of QS, and according to the user’s setting, its classification level belongs to the blacklist, CFA will determine that the URL belongs to the blacklist, immediately and automatically update its blacklist, prohibit UT from accessing the URL, and return Error or warning information to UT [corresponding process ②].

(2) If the URL is in the URL library of QS, and according to user settings, its classification level belongs to the white list, CFA determines that the URL belongs to the white list, automatically updates its white list immediately, and forwards the request to TWS [corresponding Process ③]; TWS will reply UT with a corresponding response [corresponding process ⑥].

(3) If the URL is not in the URL library of QS, QS will notify CFA that the URL cannot be determined, and CFA will respond according to the policy set by the user in advance: one is automatically treated as a whitelist, and the other is automatically treated as a blacklist list processing. However, in this case, CFA will no longer update its blacklist or whitelist. On the other hand, QS will hand over the URL to CAMS for processing [corresponding process ⑦].

The content analysis management server CAMS will regularly publish updated URL classification database [corresponding process ⑩] to QS, so that it can reflect the changes of the content in the Internet in time. The performance of CAMS directly affects the filtering accuracy, so it needs to pay a large price to maintain and update CAMS.

In order to improve the discrimination speed of CFA and reduce the requirements of CFA for storage resources, it is necessary to introduce a caching mechanism in CFA, that is, to store the black and white lists frequently accessed by user UT, and reduce the chance of UT sending verification requests to QS, because a verification request requires a certain amount of time. waiting time.

Authorized users can manage the black and white lists in CFA according to their own needs, browse, add or delete them [corresponding processes ⑧ and ⑨].

Claims (10)

1, a kind of Web content filtration system is characterized in that being made up of information filtering agency (being designated as CFA), querying server (being designated as QS) and content analysis and management server (being designated as CAMS), and wherein, information filtering agency storage has blacklist and white list; Querying server has a URL storehouse with classification and rating information; Content analysis and management server are that the resource among the Internet is classified and classified estimation.

2, Web content filtration system according to claim 1 is characterized in that being provided with among the CFA user individual configuration, comprising: (1) judges that URL belongs to the setting of the URL category level of blacklist or white list; (2) when QS return information during, judge the setting of this URL attribute for these URL clauses and subclauses not; (3) blacklist or the white list among the manual management CFA, comprise browse, increase and delete function.

3, Web content filtration system according to claim 1 is characterized in that CFA operates on the following all kinds of software and hardware system platform in a variety of forms: (1) acting server; (2) fire compartment wall; (3) browser; (4) network access devices such as ADSL Modem, Cable Modem, telephone line modem, ISDN PC adapter.

4, Web content filtration system according to claim 1 is characterized in that QS has the URL classification and the rating information of magnanimity, and the URL that CFA is submitted to carries out quick search and returns corresponding rating information.

5, Web content filtration system according to claim 1 is characterized in that QS can dispose in a large number on Internet or Intranet, support concurrent inquiry, is used to serve all types of user; QS will download classification and rating information storehouse from all kinds of CAMS that obtain the authorization.

6, Web content filtration system according to claim 1 is characterized in that CAMS adopts content-based multimedia analysis and processing method, all kinds of media contents in the Internet analyzed and assessed, and according to different their mark classifications that is categorized as.

7, Web content filtration system according to claim 1 is characterized in that CAMS introduces man-machine interactively and mark, utilizes the classification accuracy of machine learning enhanced system.

8, Web content filtration system according to claim 1 is characterized in that between CFA and the QS, and communicating by letter between QS and the CAMS all needs by authentication.

9, a kind of method of Web content filtration is characterized in that utilizing the described Web content filtration system of claim 1, and concrete steps are as follows:

(1) when the user sends the request that certain URL is conducted interviews, CFA is according to blacklist or white list, forbids or allows this access request;

(2) if this URL not in the blacklist and white list of CFA, CFA then sends query requests to QS;

(3) QS will inquire about the rating information of this URL and the result is returned to CFA in local URL storehouse, and CFA then makes a response in view of the above;

(4) QS understands the URL rating information of down loading updating from CAMS regularly;

(5) CAMS search for automatically, multi-medium data on download and the analyzing and processing the Internet, adopt man-machine interactively mask method and machine automatic classification method, Web content is classified and classified estimation, form the URL information bank of classification and classification.

10, Web content filter method according to claim 9 is characterized in that the job step of CAMS is as follows:

(1) web crawlers group: go up from main search download various types of data, according to suspicious URL information bank requirement, data download object from Internet;

(2) feature extraction: all kinds of multimedia data objects of downloading are carried out analyzing and processing, extract feature;

(3) artificial mark: from the multimedia data downloaded object, select segment data object to classify and the classification mark; Manually the result to automatic classification and classification checks;

(4) training classifier: to classifying automatically and classification, adopt machine learning method,, grader is trained with artificial mark and the related feedback information of instructing with the corresponding data object of URL;

(5) classification and classification automatically: the grader that trains is handled each data downloaded object being carried out classify and grading automatically, obtains classifying and classification URL information bank afterwards; To this URL information bank regular update and issue.

Images