Scientific text citation analysis using CNN features and ensemble learning model.

Area of Science:

• Bibliometrics
• Natural Language Processing
• Machine Learning

Background:

• Citations are crucial for evaluating academic achievements but are often treated equally, ignoring qualitative nuances.
• Current methods predominantly rely on quantitative measures, neglecting the sentiment and importance conveyed by citations.
• Existing qualitative citation analysis often uses a binary classification (important/unimportant), limiting its scope.

Purpose of the Study:

• To develop a novel framework for multi-class sentiment analysis of in-text citations in research articles.
• To address the challenge of imbalanced data in citation sentiment analysis.
• To incorporate qualitative aspects into citation evaluation alongside quantitative metrics.

Main Methods:

• Feature extraction using a convolutional neural network (CNN).
• Classification using a voting classifier combining Logistic Regression (LR) and Stochastic Gradient Descent (SGD).
• Handling class imbalance with the Synthetic Minority Oversampling Technique (SMOTE).

Main Results:

• The proposed framework achieved perfect scores (accuracy, precision, recall, F1) of 0.99.
• Demonstrated superior performance compared to traditional methods using Term Frequency (TF) and TF-Inverse Document Frequency (TF-IDF).
• Effectively handled multi-class sentiment classification on imbalanced datasets.

Conclusions:

• The study advances sentiment analysis in academic citations by incorporating qualitative evaluation.
• The proposed framework offers a more nuanced and accurate method for assessing citation impact.
• Highlights the necessity of integrating qualitative sentiment analysis for comprehensive citation evaluation.