SummarFlex: Exploring Personalized News Filtering and Reading with Query-Focused Hierarchical Summarization

Abstract

News readers often need to filter articles of their interests and seek needed information in the selected article. However, they may face clickbait titles and redundant content, which could reduce their news filtering efficiency and reading experience. In this paper, we develop a technical prototype SummarFlex to explore news filtering and reading with query-focused hierarchical summarization. In SummarFlex, users can filter articles with titles generated by GPT that are relevant to the queried keywords. When reading a selected article, users can specify keywords to generate a highlevel summary of the full content and click a sentence to get an expanded summary hierarchically. A within-subject study (N = 20) shows that compared to a baseline tool without the summarization features, SummarFlex significantly improves users' efficiency in finding interested articles and comprehension of needed information in the article. We conclude with design insights for supporting news filtering and reading with machine-generated content.