Using K-Means and Variable Neighborhood Search for Automatic Summarization of Scientific Articles.

This work presents a method for summarizing scientific articles from the arXive dataset using Variable Neighborhood Search (VNS) heuristics to automatically find the best summaries in terms of ROUGE-1 score we could assemble from scientific article text sentences. Then vectorizing the sentences using BERT pre-trained language model and augmenting the vectors with topic embeddings obtained by applying the K-means algorithm. Finally, training the Random Forest classification model to find sentences suitable for the summary and compile a summary from the selected sentences. The described algorithm produced summaries with high ROUGE-1 scores (0.45 on average), so we are heading for further developments on a larger dataset.