In[6, Mika presents two approaches to retrieve relationships between tags concept-mining based on graph clustering algorithms(A-set analysis)and set heory assumptions, which are similar to our work, but does not take into account contextual conditions nor popularity of the tags. It seems that it was performing acceptably on a small chosen domain(such as keywords related to semantic web) Schmitz et al. [10 opted for association rules mining to build-up conceptual structure. Resulting rules have the form Users assigning the tags from a to some resources often also assign the tags from B to them. Even if authors did not provide the way how to derive a taxonomy from the mined rules, we can just look on them as on the subsumption relations, which we are deriving from the overlapping sets of tagged resources Heymann and Garcia-Molina 1l] proposed another approach based on com- paring tag vectors and connecting similar tags together. Then, the taxonomy is created according to tags' centrality measure in the created similarity graph Schwarzkopf et al. 9 extend both algorithms [10, 1l by taking into account a context of a tag, defined similarly to our work. They did not try to filter-out non-popular tags as we do, in order to obtain only a"crowd-agreement"tags, nor they do any further processing in order to enrich furthermore the taxonomy Shepitsen et al. [12 propose context-dependent hierarchical agglomerative clustering technique to organize tags into clusters subsequently used for recom- mendation of resources. As with any other clustering technique, a crucial part is the definition of similarity between items being clustered. Shepitsen et al. used cosine similarity of vectors over the set of tags. We see several differences of our approach compared to the aforementioned ones. Apart from popularity, configurable bidirectional context checking(to an- cestor or to children) and siblings detection, we proposed also an incorporation of other techniques and approaches into one corpus such as spreading activation, which greatly improves the resulting hierarchy and broaden its possible usage 5 Conclusions In this paper, we have shown a method how wisdom of the masses in the form of folksonomy can be used to create a"tagsonomy"(a taxon- omy of tags). We also proposed other techniques with different background such as graph activation search coming from the graph theory and Wordnet's concer tual semantic relationships coming from the cognitive science area, which can contribute and enhance the final taxonomy of tags by adding new " shortcuts between hierarchically ordered tag licious folksonomies, which proved the viability of the approach and pointed out some interesting differences in the two mentioned tagging systems. We have shown that our algorithm for deriving hierarchy from the folksonomy can handle such differences when properly configured. More, the results proved that web 2.0 generated folksonomies can be used, when taking into account tags with a cerIn [6], Mika presents two approaches to retrieve relationships between tags: concept-mining based on graph clustering algorithms (λ-set analysis) and set theory assumptions, which are similar to our work, but does not take into account contextual conditions nor popularity of the tags. It seems that it was performing acceptably on a small chosen domain (such as keywords related to semantic web). Schmitz et al. [10] opted for association rules mining to build-up conceptual structure. Resulting rules have the form Users assigning the tags from A to some resources often also assign the tags from B to them. Even if authors did not provide the way how to derive a taxonomy from the mined rules, we can just look on them as on the subsumption relations, which we are deriving from the overlapping sets of tagged resources. Heymann and Garcia-Molina [11] proposed another approach based on com￾paring tag vectors and connecting similar tags together. Then, the taxonomy is created according to tags’ centrality measure in the created similarity graph. Schwarzkopf et al. [9] extend both algorithms [10, 11] by taking into account a context of a tag, defined similarly to our work. They did not try to filter-out non-popular tags as we do, in order to obtain only a “crowd-agreement” tags, nor they do any further processing in order to enrich furthermore the taxonomy. Shepitsen et al. [12] propose context-dependent hierarchical agglomerative clustering technique to organize tags into clusters subsequently used for recom￾mendation of resources. As with any other clustering technique, a crucial part is the definition of similarity between items being clustered. Shepitsen et al. used cosine similarity of vectors over the set of tags. We see several differences of our approach compared to the aforementioned ones. Apart from popularity, configurable bidirectional context checking (to an￾cestor or to children) and siblings detection, we proposed also an incorporation of other techniques and approaches into one corpus such as spreading activation, which greatly improves the resulting hierarchy and broaden its possible usage. 5 Conclusions In this paper, we have shown a method how wisdom of the masses in the form of social bookmarking folksonomy can be used to create a “tagsonomy” (a taxon￾omy of tags). We also proposed other techniques with different background such as graph activation search coming from the graph theory and Wordnet’s concep￾tual semantic relationships coming from the cognitive science area, which can contribute and enhance the final taxonomy of tags by adding new “shortcuts” between hierarchically ordered tags. We performed several experiments with the algorithm on CiteULike and de￾licious folksonomies, which proved the viability of the approach and pointed out some interesting differences in the two mentioned tagging systems. We have shown that our algorithm for deriving hierarchy from the folksonomy can handle such differences when properly configured. More, the results proved that web 2.0 generated folksonomies can be used, when taking into account tags with a cer-