Automatic text summarization

Abstract

Automatic text summarization has been a rapidly developing research area in natural language processing for the last 70 years. The development has progressed from simple heuristics to neural networks and deep learning. Both extractive and abstractive methods have maintained their interest to this day. In this thesis, we will research different methods on automatic text summarization and evaluate their capability to summarize text written in Finnish. We will build an extractive summarizer and evaluate how well it performs on Finnish news data. We also evaluate the goodness of the news data to see can it be used in the future to develop a deep learning based summarizer. The obtained ROUGE scores tell that the performance is not what is expected today from a generic summarizer. On the other hand, the qualitative evaluation reveals that the generated summaries often are more factual than the gold standard summaries in the data set.