Users need to be able to browse and retrieve Web content. One way of doing this is through the use of tags, which are keywords that users can freely choose to describe and organize Web resources. The use of a tag recommender system--software that suggests recommended tags--can aid users in the tag selection process.
This paper discusses STaR, a tag recommender system that combines two general approaches to recommending tags: content based and collaboration. A content-based technique uses heuristics to extract tags directly from the textual content of resources. A collaborative approach uses a community of users who are familiar with a particular resource to annotate that resource.
The paper presents the architecture of the STaR system. The system generates a set of candidate tags, which are made up of both collaborative and content-based candidate tags. These tags are merged to produce a set of recommendations that adhere most closely with a linear combination of partial relevance scores that the recommendation algorithm returns.
The authors conducted experiments to evaluate the performance of the STaR system against alternative systems. The system achieved good predictive accuracy, although it did not rank at the top in any of the evaluations against competing systems. That should not detract from the value and quality of this work, as being the best is not the only determinant of success. The paper is part of an evolutionary process that provides a stepping-stone for future work.