A Neural Network Approach for Text Classification Using Low Dimensional Joint Embeddings of Words and Knowledge.

The continuous expansion of textual data collection and dissemination in electronic means has made text classification a crucial task to help exploit, in a variety of applications, massive amounts of digital texts available nowadays. Knowledge Graphs (KGs) or their embeddings can provide additional semantics to improve text classification. However, most proposals from the literature rely solely on words found in the texts to classify them. A few text classification approaches employ knowledge embeddings besides word embeddings, but which are produced separately and not integrated into the same vector space. Different from previous proposals, this work applies an existing solution for generating text and knowledge embeddings in an integrated way to feed neural classifiers. Experiments using these joint embeddings with 50 dimensions yielded results comparable to those of state-of-the-art approaches on the AG News dataset and slightly superior to the BBC news dataset.