Enhanced Topic Representation by Ambiguity Handling

Most of the existing semantic-based topic models and topic
generation approaches use external knowledgebases or ontology to interpret
the meanings of the words. However, general ontologies do not cover
many ambiguous or specific domain-related words in a text collection.
Hence those ambiguous or domain-specific words are neglected in capturing
the meanings in topic generation. In this paper, we introduce an
approach to disambiguate the unmatched words in a text collection based
on related and similar meaning words. Word embeddings are applied
to discover similar or related words. We evaluated the topic generation
approach with our ambiguity handling technique with a set of state-of-theart
systems which uses an external ontology. Our approach outperformed,
and the generated topics were more meaningful. Our ambiguity handling
approach interpreted all the important words and included them in the
topic generation process.