SIFT - Solution for Internet Combined Filtering

The SIFT Project will be carried out within the framework of the multiannual European Community Action Plan on Promoting Safer Use of the Internet by combating illegal and harmful content on global networks (IAP). For more information go to www.saferinternet.org

The SIFT Project began on January 2002 and will last for 18 months. Its partners are:

· OPTENET - OPTENET is a Spanish company founded in 1997 that has developed an efficient technology leader in the filtering of accesses to the Internet (OPTENET). The company came about out of the need for filling in the gap that existed in the market of tools for the optimization of the Internet use in companies, schools and universities, public administration, as well as the homes (www.optenet.com).

· ICRA (Internet Content Rating Association) - The Internet Content Rating Association (ICRA) is an independent non-profit organization with offices in Europe (UK) and the USA. ICRA's mission is to protect children from potentially harmful material on the Internet, whilst protecting the content providers' freedom of expression. ICRA has developed a global system with which providers can label their content objectively. A filtering tool for use by parents according to what they feel is appropriate for their children to view has also been developed (www.icra.org).

· NCSR - The National Center for Scientific Research "Demokritos" is a self-governing research organization, under the supervision of the Greek Government. The laboratory has participated (and is currently participating) in numerous national and EU projects and its expertise focuses on the following areas: information extraction, content categorization, machine learning and personalization. Recently, it has developed a filtering software (FILTERIX) in an attempt to illustrate the application of innovative and efficient methods for multimedia document categorization (www.iit.demokritos.gr/skel).

The INCORE project recognised the need to make self-labelling and various commercially available filtering solutions inter-operable. This will empower European Internet users to select a combination of filtering choices which best meets their needs and culture: "User choice will be promoted by widening the existing choice between competing third-party rating systems and self-labelling and by allowing users to draw on the content information available in different systems which are inter-operable." (INCORE Executive Summary, para 1.4.5). The project concludes that better filtering solutions for various European languages are necessary: "There is a need for the promotion of labelling and filtering (and the development of profiles) on a national or language group basis."

The SIFT Project will define and implement an Integration Platform to answer this call. Existing technologies will be used to minimise the amount of software development needed. An Application Program Interface will be made public so that software companies will be able to develop new filtering solutions or adapt existing ones to work with the Integration Platform. The interception of a user's requests for web pages and the delivery to the browser either of the requested page or a block message will all be carried out by the Integration Platform. This is the aspect of any filter which is most deeply embedded in the PC or network server. A user will be able to buy a filter, such as OPTENET's solution which will have been designed to work with the Integration Platform. The user will choose the order of precedence given to block/allow results from the third party filter, self-labels and pseudo labels. Logically, any number of third party filters which have been designed to work with the Integration Platform may be used.

A user will thus have a complete choice of filtering solutions which may work either singly or in combination.

NCSR's learning-based classifiers will be used to generate a pseudo label covering nudity and sexual content as defined in the ICRA model and will cover (at least) English, French and German language sites. The Integration Platform, including the reading of any labels present on sites and the NCSR filters, will be tested in combination with OPTENET's text-analysis based filtering solution which is available in Spanish, Portuguese, English, Italian, French and German.