SEMANTICALLY ENRICHED DOCUMENT – LEVEL SENTIMENT ANALYSIS: A COMPREHENSIVE STUDY

Abstract

In the era of digital communication, vast amounts of textual data are generated daily through social media, reviews, blogs, and forums. Extracting meaningful insights from this unstructured data presents both opportunities and challenges. Sentiment analysis sometimes referred to as opinion mining, it is a branch of Natural Language Processing (NLP) that focuses on recognizing and classifying opinions in written material in order to ascertain if the author has a positive, negative or neutral attitude toward a certain subject. Document-level sentiment analysis is a vital task within Natural Language Processing (NLP) that aims to determine the overall sentiment expressed in an entire document, rather than individual sentences or phrases. Unlike sentence-level or aspect-level sentiment analysis, the document-level approach evaluates cumulative sentiment to understand the general opinion conveyed by the author. This method is crucial in domains such as product reviews, movie critiques, customer feedback and news articles, where the holistic sentiment of the text is more informative than isolated fragments. This study explores various methodologies for document-level sentiment classification, starting from traditional machine learning models such as Naïve Bayes, Support Vector Machines (SVM) and Random Forests, which rely on bag-of-words, TF-IDF and n-gram features. While these models offer baseline performance, they often fail to capture the deeper context and inter-sentence dependencies present in longer texts.