Candidate Summary Elimination Approach for Turkish Hybrid Text Summarization

The importance of text summarization in the field of natural language processing has increased significantly in recent years. With increasing amounts of text data, it is crucial to develop effective methods for automatic text summarization to extract information quickly and accurately. Text summarization methods are generally divided into extractive and abstractive. In this study, both methods were used together in a hybrid manner. Luhn, LSA, TextRank, LexRank, SumBasic, PageRank and Hidden Markov Model were used as extractive methods, and the transformer-based BERT2BERT model was used as the abstractor method. The summaries were first summarized separately from the raw text using subtractive summarization methods. The obtained extractive summaries were summarized again by the abstractive model according to the two highest-scoring extractive method summaries, based on the score consisting of the combination of the ROUGE score, BERTScore and METEOR scores at certain weights, which we collectively call "BIG3". As a result of the study, it was observed that the summaries created with our hybrid model achieved higher scores in the evaluation criteria.