Automated Assessment of the Validity of Content Take-Down Notices?

The next WebSci 2016 paper session starts with a presentation by Pei Zhang, which introduces what she calls the Content-Linking-Context model, or CLC. The context for this is legislation such as the Digital Millennium Copyright Act (DMCA) and the European e-Commerce Directive, as well as various national legislation in the EU.

The DMCA requires services providers to take down content on request expeditiously, even without verification of copyright infringement claims, and providers such as Google and Dailymotion are known to act on such requests, but there is little information about the criteria they use to vet requests. Can automated systems be used to help such providers in assessing the validity of take-down notices, then?

The project used Google's Transparency Report – which covers take-down notices – as a benchmark on the status of the content highlighted in take-down notices; from this it emerges that more than half of the take-down notices relate to music and audio content, and that nearly half relate to online streaming services.

Google itself has created a Trusted User Programme for requesters whose take-down notices are processed especially expeditiously, and also uses domain information to assess the likelihood of requests being genuine. In all of this, different types of linking – simple links and embedded links – also need to be taken into account.

The CLC model, then, takes into account a set of criteria to assess the similarity of allegedly infringing content with original copyright work: URL accessibility, content existence, and similarity with copyrighted works; the linking of allegedly infringing work: online playability, downloadability, link types of online-playable resources, and link type of downloadable resources; and the context of allegedly infringing content: title of and information about the content; performer information; and the URLs of the content.

It assesses content and linking factors first, and on this basis makes a first judgment about the likelihood of copyright infringement; to further improve that judgment, the context is then further considered. The idea is that this decision tree can be implemented in an automated system that helps take-down notice receivers improve the accuracy of their responses to such notices.