Finding the reference text in citation contexts using attention model

Precise reference text extraction from citation contexts (CCs) is important in computational linguistics and information retrieval applications. The extraction of CCs forms the basis of critical tasks like citation network analysis and literature search/recommendation systems, all of which hinge on the fidelity of extracted reference information. However, traditional methods, often relying on full sentences or fixed-window approaches, suffer from unnecessary information inclusion and negatively affecting accuracy. In this study, we aim to bridge this gap by introducing a novel deep learning approach utilizing an Attention Model to directly extract reference text from CCs. This eliminates the need to sift through lengthy surrounding text. The model was trained on a dataset consisting of 100 cited papers from the fields of Natural Language Processing and Computational Linguistics. Each paper was converted into a text file and sequentially loaded for training. Tokenization was applied to convert textual data into numerical values. In the training phase, we used the Adam optimizer and adjusted batch sizes based on the number of citations in each paper. The proposed model obtained promising results, achieving a macro-F1 score of 0.87. This demonstrates its superior performance compared to standard benchmark techniques such as conditional random field and dependency parsing. Additionally, our model outperformed the benchmark techniques in terms of precision, recall, and F-score.