19.4.3 Content-Based Social Tagging RS
All the algorithms described so far do not exploit the content of resources, and hence can be applied to any folksonomy regardless the type of resource supported. Nonetheless, the content of resources is a valuable source of information, specially in cold-start scenarios where there is scarcity of explicit user feedback. In the following, we shortly discuss recommenders that make explicit use of resources’ content.
19.4.3.1 Text-Based
Song et al. [32] proposed an approach based on graph clustering for tagging textual resources, like web pages or other kinds of documents. It does not perform personalized recommendations, as it does not examine users individually. In particular, it considers the relationship among documents, tags, and words contained in resources. These relationships are represented in two bipartite graphs. The approach is divided in two stages: