Tag Recommendation

Tagging is an important feature of the Web 2.0. It allows the user to annotate items/resources like songs, pictures, bookmarks, etc. with keywords. Tagging helps the user to organize his items and facilitate e.g. browsing and searching. Tag recommenders assist the tagging process of a user by suggesting him a set of tags that he is likely to use for an item. Personalized tag recommenders take the user’s tagging behaviour in the past into account when they recommend tags. That means each user is recommended a personalized list of tags – i.e. the suggested list of tags depends both on the user and the item. Personalization makes sense as people tend to use different tags for tagging the same item. This can be seen in systems like Last.fm that have a non-personalized tag recommender but still the people use different tags. For this, we will provide an empirical evaluation on a subset of Last.fm that shows, that our proposed personalized tag recommender outperform even the theoretical upper-bound for any non-personalized tag recommender.