Séminaire Donnees et APprentissage Artificiel
Personalized Interaction with Multimedia Data Collections
Thursday, March 20, 2008Andreas NUERNBERGER (Magdeburg University, Germany)
Results of multimedia queries are usually sorted simply by the similarity score of an object to the query. These scores are obtained using different similarity computations and with it different ranking methods depending, e.g., on the type of the query. However, a major issue of the usually employed ranking methods is that they neglect user interests. Instead, the ranking is purely based on information extracted from the collection itself and the similarity of the documents to the query. For the predominant case of a user searching on the web for currently popular topics or highly linked pages, this provides good results. However, these methods could even inhibit the retrieval of documents on special topics. In this case, the user must specify the correct additional keywords that distinct his topic from the rest. Otherwise, the relevant documents will get a rather low rank. The research results presented in this talk are two-fold. First, methods to create user profiles based on the user interaction with the system are discussed. Second, methods that are capable of integrating this information in the presentation of search results or in navigational support for accessing multimedia collections are presented.
More details here …
Javier.Diaz (at) nulllip6.fr