The words representation, as basic elements of documents representation, plays a crucial role in natural language processing. Topic models and Word embedding models have made great progress on words representation. There are some researches that combine the two models with each other, most of them assume that the semantics of context depends on the semantics of the current word and topic of the current word. This paper proposes a topic enhanced word vectors model (TEWV), which enhances the representation capability of word vectors by integrating topic information and semantics of context. Different from previous works, TEWV assumes that the semantics of the current word depends on the semantics of context and the topic, which is more consistent with common sense in dependency relationship. The experimental results on the 20NewsGroup dataset show that our approach achieves better performance than state-of-the-art methods.
CITATION STYLE
Li, D., Li, Y., & Wang, S. (2017). Topic enhanced word vectors for documents representation. In Communications in Computer and Information Science (Vol. 774, pp. 166–177). Springer Verlag. https://doi.org/10.1007/978-981-10-6805-8_14
Mendeley helps you to discover research relevant for your work.