Twitter Sentiment Analysis of Juvenile Behaviour Deviations using LSA (Latent Semantic Analysis)

Abstract

Sentiment analysis is an effort to mine data or can be called text mining. One of the uses from sentiment analysis is to analyze public opinion on a particular topic or product. Research related to sentiment analysis has been carried out with various methods and different accuracy results. The approach method in analyzing sentiment is divided into 2, namely with statistical approach and semantic approach. Research with a statistical approach has been developed previously. While the semantic approach is still quite rare. Based on previous study, The LSA (Latent Semantic Analysis) approach seems effective in analyzing data and further more for semantic understanding of sentiment orientation. In this study, we build two model, the first is statistic approach using TF-IDF for feature extraction then classified into SVM (Support Vector Machine) algorithm. Then compare it with semantic approach using TF-IDF and SVD (Singular Value Decomposition) for feature extraction then put into LSA (Latent Semantic Analysis) model. The LSA (Latent Semantic Analysis) model will convert the original feature space to a new lower-dimension space by calculate cosine similarity between sentence vectors with vector topics to produce a score for each sentence by measuring how closely the semantic meaning between the sentence and all the words on the topic. Then divided into three collection of words that sentiment oriented including its vector similarity value. The results of the study prove that the research with statistical approach using SVM (Support Vector Machine) algorithm gives results of precission average about 0.56 and recall about 0.54. While with semantic approach using LSA (Latent Semantic Analysis) gives a slight increase in result of precission about 0.57 and recall about 0.56.