Words Embedding in Form of Symmetric and Skewsymmetric Operator

Words embedding is a representation of words with real-valued vectors. They were originally created to solve natural language processing tasks. Examples of these tasks are document indexing, semantic analysis, and question answering. There is another approach to most of these tasks — language modeling. However, all language models use pre-trained word embeddings, which means constructing new word embedding models is still a relevant task. There are many word embedding models known: Word2Vec, GloVe, FastText, etc. Each one of these models due to words relations asymmetric nature represents each word with two real-valued vectors: central and context. This leads to problems like final embedding construction ambiguity and model training time extension. I introduce a new approach based on asymmetric information extraction that uses the advantages of Global Vectors model. Due to the reduction of asymmetric information impact on resulting words representations, the model converges faster and outperforms existing models on words analogies tasks. Index Terms — word embeddings, GloVe, word analogies, language modeling, natural language processing, matrix decomposition.