A constrained optimization algorithm for learning GloVe embeddings with semantic lexicons

摘要: Abstract GloVe representations of words as vector embeddings in continuous spaces are learned from matrix factorization the words’ co-occurrences constructed large corpora. Due to their high quality textual features, have been extensively utilized for many text mining and natural language processing tasks with considerable success. Further improvements these word can be obtained by also taking into account valuable information semantic properties complex relationships between them provided lexicons. In this paper we adopt optimization techniques domain machine learning constrained order leverage relational knowledge words, propose an efficient algorithm that produces enhanced information. The proposed outperforms other related approaches utilize either during training or a post-processing step. Our claims validated experiments on popular tasks, including similarities, analogies, sentiment analysis, which demonstrate our model significantly improve representations.