Automating Citation Intent Classification on SciCite Dataset

Scientific papers often contain citations that provide important information on the background, methods used, and results obtained. Identifying the intent of these citations is a critical machine learning problem, and it is also essential for the automated analysis of scientific literature. The SciCite dataset, consisting of 11,020 rows and 15 features, was used for this purpose. Preprocessed data was then fed to machine learning models such as KNN, Random Forest Classifier, Light Gradient Boost Classifier, Gradient Boost Classifier, and XGBoost. A comparison was drawn among these models to identify the one that yields the highest weighted F1 score, which was found to be 87% for the Light Gradient Boost Classifier. The findings of this work demonstrate the effectiveness of using Machine Learning models and word embedding techniques, such as Doc2Vec and GloVe, to accurately classify the intent of citations in scientific papers.