How Personalized is Your Summary? An Empirical Study on Student Lecture Note Summarization

Anonymous

Abstract: Personalized text summarizaiton is one of the research hotspots of text summarization. Existing work has focused on how to incorporate the conditions of personalization into the methodology, and very little work has been done to analyze the embodiment of personalization from user real data perspective. In this paper, we conduct a personalized summarization collection experiment, which recruited 200 students who wrote 2,000 lecture note summaries. An analysis of structure, modality, content length, reference usage, and abstractiveness about summaries is conducted to address the question, "How personalized can human-written summaries be?" Additionally,we introduce a personalized text summarization dataset, and benchmark the state of art summarization models, the result indicate that the abstractive models show better performance on our dataset.

Paper Type: short

Research Area: Summarization

Contribution Types: Data resources, Data analysis

Languages Studied: English