From Word Embedding to Inferring user latent Interests

Social media websites captured web space. The members of these media s increasing daily. With the data shared by people, researchers try to use them in a proper way to help recommender systems. One of the hot research areas is user interest detection. Intelligent web systems try to extract user primitive interest from contents which are shared by users. While most of the works concentrate on extracting user initial interest, less attempt dedicated to understanding latent ones. In this paper, we demonstrate how word embedding methods could help us to enrich user interests profile. We generating state-of-art user interest modeling which deploys word2vec method for enriching user initial interests that extracted from user s twitter account. Our experimental results demonstrate that using semantic similarity measures, especially when using Word embedding methods, outperform traditional methods. Empirical results show that enriching user interest profile leads to better personalized content based recommendation.

کلیدواژه ها:

Latent interest ، twitter ، word embedding ، Word2Vec ، User interest detection