Sentiment Hotspots’ Extraction in Large Text Documents Using Neutrosophic Sets

Extraction of sentiments from large text documents has always been a research challenge. Due to the text volume, this process is computationally and temporally intensive. In addition, it has been observed that significant sentiments in lengthy texts are typically condensed into a few sentences. The remaining text is either objective or adds nothing to the overall meaning of the document. This paper seeks to extract several subjective, sentiment-rich sentences from a large document. Thus, it is sufficient to extract the overall sentiment of a lengthy document by analyzing a few sentences. This paper presents a novel aggregation technique to find a cluster of sentences rich in sentiments from a document. This technique utilizes neutrosophic sets’ distance to find the density of opinion words in a document and combine them together in a cluster to form a hotspot. This technique also caters to the uncertain part of the sentiment by using neutrosophic sets. Further, it is experimentally demonstrated that finding sentiments through dense hotspots yields superior results as 0.74 F1 micro and 0.62 F1 macro to finding sentiments from the entire document using a lexicon-based method.