CUFE at SemEval-2016 Task 4: A Gated Recurrent Model for Sentiment Classification

摘要

In this paper we describe a deep learningudsystem that has been built for SemEvalud2016 Task4 (Subtask A and B). In this workudwe trained a Gated Recurrent Unit (GRU)udneural network model on top of two sets ofudword embeddings: (a) general word embeddingsudgenerated from unsupervised neural languageudmodel; and (b) task specific word embeddingsudgenerated from supervised neuraludlanguage model that was trained to classifyudtweets into positive and negative categories.udWe also added a method for analyzing andudsplitting multi-words hashtags and appendingudthem to the tweet body before feeding it to ourudmodel. Our models achieved 0.58 F1-measureudfor Subtask A (ranked 12/34) and 0.679 Recalludfor Subtask B (ranked 12/19).