Build Emotion Lexicon from the Mood of Crowd viarnTopic-Assisted Joint Non-negative Matrix Factorization

摘要

In the research of building emotion lexicons, we witness thernexploitation of crowd-sourced affective annotation given byrnreaders of online news articles. Such approach ignores thernrelationship between topics and emotion expressions whichrnare often closely correlated. We build an emotion lexicon byrndeveloping a novel joint non-negative matrix factorizationrnmodel which not only incorporates crowd-annotated emotionrnlabels of articles but also generates the lexicon usingrnthe topic-specific matrices obtained from the factorizationrnprocess. We evaluate our lexicon via emotion classificationrnon both benchmark and built-in-house datasets. Resultsrndemonstrate the high-quality of our lexicon.