Felipe Almeida

This work addresses the problem of how to predict tags that will be assigned by users in social tagging systems. It is widely known that a tag prediction functionality helps promote system usability and increase the quality of the tag vocabulary in use. With that in mind, we verify the difference in the performance of several label ranking techniques on two datasets, which differ from each other in several key metrics, such as the average number of tags per resource, tag vocabulary length, the total number of resources, etc. We also analyze a specific label ranking technique, namely MIMLSVM. We verify whether it generalizes to dense text representations in addition to traditional sparse ones. Experiments are conducted on the two datasets, and the results are analyzed.

This work lists and describes the main recent strategies for building fixed-length, dense, and distributed representations for words based on the distributional hypothesis. These representations are now commonly called word embeddings. In addition to encoding surprisingly good syntactic and semantic information, they have been proven useful as extra features in many downstream NLP tasks.