Improving Search by Extending Tags According to Recommendation Level and Combinations of Types

As the collaborative tagging systems such as Delicious, Flickr and Last. fm become more and more popular, a large amount of resources produced by publishers, together with rich semantic metadata, become available. There are lots of ways to make use of tag information and so far the most discussed usage is for searching. The distribution of tag types differs greatly across different systems. Also the distribution shows large difference between publishers and searchers. In order to expand tags of resources for publishers and keywords for searchers reasonable, this paper shows a comparison of the distributions of both kinds of users and proposes an approach which could calculate the recommendation level of types. The level of types could be used to describe whether the type is valuable to the corresponding users. The distribution of different combination of types has also been investigated, and with such information we analysed the most popular combination of types in query log across different collections. We compare the searching accuracy on original datasets against the datasets with resources after being expanded tags by our proposed methods. Experimental results show that our method could improve searching accuracy.