ASPECT IDENTIFICATION USING TOPIC MODELING FOR DRUG REVIEWS

Abstract

Now a days in the world of internet, many studies have observed that patients are getting important and supporting information on persistent diseases and related drugs from resources such as online reviews, web blogs, and discussion forums. Mining of information from such bulk of text is a challenging task but this information is very useful. Such information can be processed using the model called Aspect Identification using Topic Modeling (AITM) on Drug reviews which is the proposed system. This model recognizes the aspects or topics associated with the class label of the mixture of data. The important characteristic of Aspect Identification using Topic Modeling on Drug Reviews is that it determines aspect relevant to one class only. So, the chances of getting aspects of different concepts from different classes are suppressed and resulting aspects can be easily interpreted. The resulting aspects are class distinguishing means they are helpful in distinguishing a class from other. The coherent Latent Dirichlet Allocation (LDA) topic modeling algorithm is used for result extraction. Four different drugs (i.e. Citalopram, Escitalopram, Lisinopril and Simvastatin) are used for analyzing experimental results of the proposed system. The analysis shows that AITM is giving better performance than other existing approaches. AITM has improved accuracy up to 98.3% compared to existing system, also the numbers of aspects identified by AITM are more than the existing system.