From Word Embeddings to Item Recommendation

Social network platforms can archive data produced by their users. Then, the archived data is used to provide better services to the users. One of the services that these platforms provide is the recommendation service. Recommendation systems can predict the future preferences of users using various different techniques. One of the most popular technique for recommendation is matrix-factorization, which uses lowrank approximation of input data. Similarly, word embedding methods from natural language processing literature learn lowdimensional vector space representation of input elements. Noticing the similarities among word embedding and matrix factorization techniques and based on the previous works that apply techniques from text processing to recommendation, Word2Vec’s skip-gram technique is employed to make recommendations. The aim of this work is to make recommendation on next check-in venues. Unlike previous works that use Word2Vec for recommendation, in this work non-textual features are used. For the experiments, a Foursquare check-in dataset is used. The results show that use of vector space representations of items modeled by skip-gram technique is promising for making recommendations. Keywords—Recommendation systems, Location based social networks, Word embedding, Word2Vec, Skip-gram technique