利用复杂网络为自由评论鉴定词汇情感倾向性

摘要

Identifying word sentiment orientation (WSO) is usually the foundation of mining coarse-grained emotion information. In free comments, there exist many grammatical errors which disable previous grammatical text-based methods in identifying WSO for free comments, and there exist some context-sensitive words which disable offline opinion words in mining coarse-grained emotion information. In view of the above questions, a new method which identifies WSO for free comments via complex network is proposed. This method consists of two parts. The first part makes use of context information in free comments to build a sentiment orientation relationship network (SORN) for effectively solving the context sensitive and noise problems. For this purpose, two algorithms are brought forward. One is the algorithm for building the pyramid anti-noise information model and the other is the algorithm for optimizing the sentiment orientation relationship network by anti-noise information. The second part identifies WSO for free comments via SORN. For this purpose, the SORN-based WSO algorithm is put forward. Experimental results show that our method far exceeds HM in identifying WSO for free comments and has good timeliness.%词汇情感倾向性(Word sentiment orientation,WSO)的鉴定通常是对文本进行粗粒度意见挖掘的基础.自由评论中存在许多语法噪声,这使得以往基于规范文本提出的WSO鉴定方法不再适合自由评论.自由评论中的情感词汇往往是上下文敏感的,这使得非当前鉴定的情感词汇难以适用于当前自由评论的粗粒度意见挖掘.针对上述问题,提出一种新的利用复杂网络为自由评论鉴定WSO的方法.该方法主要有两个部分:1)为了利用自由评论中词汇之间的上下文信息建模一个能够有效解决上下文敏感问题且具有良好抗噪声能力的情感倾向性关系网络(Sentiment orientation relationship network,SORN),提出了两个算法:金字塔抗噪声信息模型算法和利用抗噪声信息优化调整SORN的算法；2)为了有效利用SORN为自由评论鉴定WSO,提出了基于SORN的WSO鉴定算法.实验表明:对于在线为自由评论鉴定WSO,本文方法不仅在精确度方面远高于Hatzivassiloglou提出的方法,且具有良好的时间效率.