Abstract: Emotional polarity identification of new words is very difficult because of the lack of priori knowledge in part of speech and semantic. This paper thinks of the polarity identification as calculating polar distribution of new words, and proposes an emotional polarity recognition method of new words based on label propagation algorithm (LPA). Firstly, Hownet is used to calculate the polar distribution of words that appear together with new words and select the words with a strong emotional tendency. Then, label propagation algorithm is used to estimate the polar distribution of new words based on the correlations of new word and other new words and the selected words. Finally, the emotional polarity of new word is identified by constructing linear classifier based on the polar distribution of  new word. In the evaluation task of COAE2014, this method achieves a better effect, and the precise rate and recall rate respectively are 16.167% and 13.775%.

Key words: new word, emotional polarity, label propagation, Hownet, graph model

摘要： 由于缺乏词性、语义方面的先验知识，新词的情感极性识别更加困难。将新词的极性识别看作计算新词的极性分布问题，提出了基于标鉴传播算法的新词情感极性识别方法。首先根据知网计算与新词共现的其他词汇的情感极性分布，从中挑选出具有强烈情感倾向的词汇；然后根据新词与它们以及其他新词的相关度，利用标签传播算法对新词的极性分布进行估计；最后根据新词的极性分布，通过构建线性分类器对新词的情感极性进行识别。该方法在COAE2014评测任务中，准确率达到16.167%，召回率达到13.775%，取得了相对较好的效果。

关键词: 新词, 情感极性, 标签传播, 知网, 图模型