NATURAL LANGUAGE PROCESSING USING SOFT COMPUTING

Abstract

Natural Language Processing (NLP) is a field of study that focuses on the interaction between computers and human language. It is an area related to computational linguistics, computer science and artificial intelligence. NLP has many applications including text summarization, sentiment analysis, text classification, keyword extraction, information retrieval etc. This thesis proposes different models to solve various NLP problems such as text normalization of code-mixed social media text, word sense disambiguation, interval valued fuzzy Hindi WordNet lexicon generation, domain-specific sentiment lexicon generations, text summarization and keyword extraction. The proposed models used fuzzy graphs, interval-valued fuzzy graphs, game theory and various word embeddings to solve the above-mentioned problems. A brief description of our contribution to NLP areas is discussed as follows:  This thesis proposed a novel text correction framework for code-mixed Hindi English social media text. The proposed method for the first time handles language identification along with non-word and real-word errors in code-mixed text.  This study proposed the first research for word sense disambiguation that uses fuzzy WordNet and BERT embeddings to automatically assign the membership values to fuzzy graphs.  This research work proposed a novel method to improvise the famous and widely used NLP lexicon i.e. Hindi WordNet. This work applied the concept of Interval Valued Fuzzy graphs and proposed Interval- Valued Fuzzy Hindi WordNet (IVFHWN).  This thesis proposed methods for domain-specific sentiment lexicons, especially for sentiment analysis. One method for the first time proposed DoSLex unsupervised, language-independent framework with contextual semantics based for low resource languages. Two methods are proposed for domain-specific lexicon generation using a random walk algorithm and label propagation on conceptnet graph.  This thesis presents novel methods for text summarization and keyphrase extraction using game theory and various word embeddings such as m-bert, fastext, glove and word2vec. v The proposed research works in this thesis are compared with existing state-of-the-art methods of relevant areas. The study exhibits improved performance compared to the state-of-the-art methods. It is observed that soft computing approaches viz. fuzzy graphs, fuzzy graph centrality measures, interval-valued fuzzy graphs, word embeddings, and embeddings-based game theory are able to solve complex, uncertain and ambiguous NLP issues.