Sentiment Analysis of Citation Sentences using Machine Learning Techniques

Abstract

The scientific citations, sentiment analysis got very much attention in the latest period as there is an increase in the scientific publications availability. Scholarly databases and publications are the valuable sources for citation information where authors will write about other paper ideas and results by contradicting or verify them. Citation sentence can be an essential indicator in finding the relevancy and importance of an article for the author based on various aspects of the cited work such as corpus, technique, task, method, measure, concept, tool, methodology, and model etc., Citation Sentiment Analysis is a technique to classify the sentiment polarity of citations in scientific articles. The Citation sentences corpus bears three polarities viz. positive, negative, and neutral. It is used to find the contribution towards the research field of the author and to find limitations and identifying problems in a specific approach through negative citation sentences. This will help the upcoming researchers who are searching for the research topic or research articles to focus on. We used Word2Vec model to get the word vectors and applied classification algorithms like SVM, Random forest, and linear SVC on the feature vectors and extracted the sentiment. By using the accuracy, recall, precision, and F-measure metrics we analyzed the performance of each algorithm in the classification of sentiment from the scientific citations.