A Word Embedding Model Learned from Political Tweets

摘要

Distributed word representations have recently contributed to significant improvements in many natural language processing (NLP) tasks. Distributional semantics have become amongst the important trends in machine learning (ML) applications. Word embeddings are distributed representations of words that learn semantic relationships from a large corpus of text. In the social context, the distributed representation of a word is likely to be different from general text word embeddings. This is relatively due to the unique lexical semantic features and morphological structure of social media text such as tweets, which implies different word vector representations. In this paper, we collect and present a political social dataset that consists of over four million English tweets. An artificial neural network (NN) is trained to learn word co-occurrence and generate word vectors from the political corpus of tweets. The model is 136MB and includes word representations for a vocabulary of over 86K unique words and phrases. The learned model shall contribute to the success of many ML and NLP applications in microblogging Social Network Analysis (OSN), such as semantic similarity and cluster analysis tasks.