Sequential Covering Rule Learning for Language Rule-based Aspect Extraction

Aspect extraction in aspect-based sentiment analysis has become a very challenging and important study today. One of the aspect extraction approaches is based on language rules (grammar). Language rules on most methods are manually determined, so they are more vulnerable to error. In this paper, we propose a rule learning method for aspect extraction using the Sequential Covering algorithm. The language features used are part-of-speech, dependency and constituent parse tree from each review sentence. This method generates rules starting from the simplest rule with a length of dependency relationship 1 to a certain length of dependency. Tests are carried out on several datasets with different domains. The test results show that this method succeeded in increasing f-measure aspect extraction compared to the baseline.