LSTM Sequence to Sequence Model for Dynamic Title Generation

To make research communication more efficient, this research work introduces a smart way to create titles for research papers automatically from their corresponding titles. Leveraging a diverse dataset of research papers, the study employs a comprehensive pipeline involving data preprocessing, neural network modeling, and evaluation. Our methodology employs advanced natural language processing techniques, including LSTM-based sequence-to-sequence models and word embeddings, to capture the relationship between abstracts and titles. The trained model is rigorously optimized, and its performance is evaluated using established metrics. An accuracy of 78.82% and perplexity of 1.005 is achieved. This means that the model is able to generate titles that were very similar to the human-generated titles. The results demonstrate the feasibility of automated title generation, providing titles that closely resemble human-generated counterparts. This research offers valuable insights into enhancing scholarly publishing and knowledge dissemination across academic domains.