基于短语级情感分析的不良信息检测方法

Sensitive Information Detection Based on Phrase-level Sentiment Analysis

摘要:

针对基于关键词字符匹配和粗粒度情感分析方法的传统不良信息检测方法准确率低的问题，提出一种基于短语级情感分析的不良信息检测方法，该方法制定语法规则来提取敏感词所在短语，结合二次分类的情感词典，通过分析短语的情感倾向来判断表达者对敏感关键词的情感倾向，从而判定内容的敏感性。该方法克服了字符串匹配方法忽视敏感词上下文，从而导致大量误报的缺点，及粗粒度情感分析方法由于无法准确定位情感对象，只能以文本整体情感倾向代表表达者主观情感，导致不能精准分析与敏感词相关的情感倾向及敏感性的缺点。实验表明，与传统方法相比，该方法的准确率有较大的提升。

Abstract:

String matching method and coarsegrained sentiment analysis method are often used in sensitive information detection，the accuracy, however, is rather low. To mitigate this issue, this paper proposed a method based on phraselevel sentiment analysis. This method takes advantage of a rule set which is used to extract phrases concerning the sensitive word, by analyzing the sentiment orientation of relative phrases, it is possible to determine the sentiment orientation of the expresser towards the sensitive word, thus determine the sensitivity of the information. This method takes the context of the sensitive word into consideration and is able to extract sentiment orientation towards the sensitive word instead of the whole text, which is missing from string matching method and coarsegrained method respectively. Experimental results suggest that the accuracy is considerably increased compared to string matching and coarsegrained sentiment analysis method.