Data Mining Techniques to Categorize Single Paragraph-Formed Self-narrated Stories

Abstract

In this age of natural language processing, most of the sentiment analysis tasks are done by polarization, for example, 0 for negative or 1 for positive of the given context/text. In some work, the tasks are done using fine-grained polarization, such as very negative or very positive. The proposed system of this paper includes the categorization of the paragraphs using its nature. All the paragraphs are self-narrated, and the number of words in those self-narrated paragraphs contains 50–4200 words. The paragraphs are categorized using three categorizations: “work stress,” “bullying,” and “sexual harassment” in both real and cyber worlds. Artificial neural network paragraph vectors, a distributed bag-of-words and distributed memory, are used to get the embedding of each paragraph and later for classification by data mining techniques. The accuracy of each algorithm lies between 70 and 94%. The best model gives a 77.46% F1 score in the test set.