exFLORe is a query expansion algorithm that aims to improve folksonomy search by making use of the semantically enriched tagspaces produced by FLOR. Currently, folksonomy search is limited to matching search keywords against the tags (of other textual descriptions) of resources. For example, a query for all European lakes phrased as "europe lake" will only return those resources that are tagged with both these keywords.

However, other relevant resources might exist that are not tagged with exactly these keywords but rather with their semantic variations, for example, "Balaton, Hungary" of resource 16668. To solve such situations, exFLORe focuses on concept rather than keyword search. The algorithm uses the semantic structure built with FLOR to semantically interpret the user keywords into possible concepts and then to retrieve the resources whose tags are defned by these concepts, i.e., the resources that describe these concepts. Intuitively, all resources that describe these concepts, their related concepts (currently we explore subClassOf and partOf relations) and instances (typeOf) are retrieved.




In step A of the algorithm each keyword ki is mapped to a number of potential senses sj from the semantic structure of the tagspace. This mapping is obtained by lexically matching ki to the lexical information of the senses. The number of senses can vary from zero, if this keyword does not exist in the structure, to n depending on its level of polysemy. When searching for "europe lake" both keywords are mapped to the respective senses.

The second step, B, disambiguates the retrieved senses leading to either zero or one sense for ki. The disambiguation technique of exFLORe is similar to the one performed by step F of FLOR. Query Q is then transformed into Q' by replacing each keyword with its sense if such a sense exists and keeping the keyword as such otherwise. In addition Q' is expanded with the related senses of the keywords' senses (currently, based on subClassOf and partOf relatiosn).

Step C retrieves all the resources that describe a maximum combination of elements (senses and keywords) from Q'.

Since more than one of such combinations may exist, step D groups the resources if they share the same combination of elements.

Finally step E ranks the groups of resources according to 1. the number of elements (tags and senses) their resources share with Q' and 2. the number of their resources. For example, resource 16668 is retrieved because it contains: a sense for Europe; a related sense for Europe, the sense of Hungary which is partOf Europe; and Balaton Lake which is an instance of Lake.