Feature Generation and Aspect Identification for Sentiment Analysis

Abstract

Sentiment analysis is a technique for analyzing a piece of text to determine the sentiment contained within it.This is a powerful Artificial Intelligence system with major business ramifications. Sentiment Analysis is performedby merging natural language processing(NLP) and machine learning (ML).Using basic sentiment analysis, the software can identify whether the emotion behind a piece of text is positive, negative, or neutral.We proposed a syntax-based aspect identification algorithm to identify the sentiments of reviews. The goal of this paper is to generate a feature for aspect-based sentiment analysis using the term frequency-inverse document and bag of a word, as well as to develop a model using a statistical learning approach.The dataset includes trip advisor reviews of various hotels. There are around 20,000 reviews in this dataset. Before utilizing Bow and TF-IDF to extract features, the data was cleaned and pre-processed. Training and evaluation were performed out after the classifiers were implemented. The accuracy of a classifier is measured using evaluation metrics. Out of the four classifiers used to assess accuracy, Logistic Regression has the highest accuracy in the TF-IDF. With logistic regression,the accuracy in TF-IDF was 83percent and the classification rate in Bag of Words was 80%.