Implementation of Recommender Systems based on Context Operating Tensor (COT) Model

This paper propose a novel context modeling method Contextual Operating Tensor model, named COT, which is motivated by the recent work of semantic compositionality in Natural Language Processing (NLP). It provide an efficient implementation inspired by the powerful ability in describing latent properties of words, in recommender systems, using a vector representation of each context value seems a good solution to examine the effect of contexts on user-item interactions. This distributed representation inferred from all contexts has more powerful ability in illustrating the operation properties of contexts. Moreover, in the research direction of sentence sentiment detection, a noun has semantic information as a latent vector, and an adjective has semantic operation on nouns as an operating matrix .This paper assume that contexts in recommendation systems have a similar property of adjectives and can operate latent characteristics of users and items. Then, new latent representations of entities can show not only characteristics of original entities but also new proprieties under a specific contextual situation.