Compositional Controlled Summarization

Creating condensed yet informative summaries of scientific documents while allowing users to tailor these summaries to their preferences is crucial for personalized information retrieval and better understanding of research papers. While past studies have tackled controlling summary attributes individually, the challenge of simultaneously managing multiple attributes has received less attention. In response, we introduce an innovative toolkit called CocoScisum. This toolkit is dedicated to controlled summarization of scientific documents and is tailored to cater to the unique requirements of the scientific community, featuring the ability to adjust key attributes specifically of length and keyword inclusion concurrently. Our toolkit is available on GitHub ( with an introduction video (


Yixi Ding (Reserach Engineers)
Dr. Yanxia Qin (Reserach Fellow)
Dr. Qian Liu 
Min-yen Kan