一种利用语义相似特征提升细粒度情感分析方法

A FINE-GRAINED SENTIMENT ANALYSIS METHOD USING SEMANTIC SIMILARITY FEATURE

情感分析主要研究人们正面或负面情感的表达。随着网页文本的爆炸式增长,情感分析在学术研究和实际应用中都成了热门话题。细粒度的情感分析方法通常采用两步策略,从而极易产生自底向上的层叠错误问题。为了解决这个问题,研究者们提出了一种基于马尔科夫逻辑的细粒度的情感分析联合框架。其中最常用的传统全局特征是自底向上和自顶向下特征。为了更好地提升细粒度情感分析的联合学习能力,一种新的语义相似特征被提了出来,中文情感分析数据集上的实验证明,此特征能对情感分析联合框架带来极大的改进。

Sentiment analysis mainly focuses on the study of people＇ s emotional expressions including positive and negative sentiment. With the explosive growth of web texts, sentiment analysis has become a hot topic in both academic researches and practical applications. The method of fine-grained sentiment analysis traditionally adopts a 2-step strategy, which is extremely easy to result in stack-up bottom-up errors. A joint fine-grained sentiment analysis framework based on Markov logic is proposed to solve this problem. ＂Bottom-up＂ and ＂Top-down＂ are the two most commonly used traditional overall semantic semantic similarity similarity features. In order to improve the joint learning ability of fine-grained sentiment analysis, a new feature has been proposed. Experiments on the data set of Chinese sentiment analysis prove that the feature can bring a significant improvement to the joint fine-grained sentiment analysis framework.