The following figure (Fig. 3) shows how many WordNet entries have been associated to each of the 930 annotated hashtags. It may be seen that E-64-c almost 600 hashtags, including the 409 hashtags that have been found directly in WordNet, are linked to a single WordNet entry (that may contain a list of synsets, if the hashtag has different meanings). The rest of the hashtags have been annotated with the support of Wikipedia. Almost 100 of them have 2 entries, and very few hashtags have more than five associated WordNet entries.
Finally, the most relevant clusters were selected using the mechanism explained in Section 3.3. The tree was analysed in a bottom-up fashion, from the horizontal cut pith divides the set of hashtags in 200 clusters up to the one that partitions the set in 5 clusters. In order to analyse the influence of the values of the filtering parameters, the selection procedure was applied 50 times, considering the following values for them:•Threshold thHomogeneity (minimum inter-class homogeneity): values ranging from 0.5 to 0.9 in 0.1 intervals (5 values).•Threshold thElems (minimum number of elements): even values ranging from 2 to 20 (10 values).