Chatsum: An Intelligent Medical Chat Summarization Tool

During the recent pandemic, online meetings, voice, and chat-based services have become extremely popular and the only way to provide remote services. Medical advising is one such mandatory service where patients use text to chat with medical professionals over the internet or mobile media and optionally upload sensitive information such as images. With a view to improving the quality of service of the remote medical advising system of our industry partner, we propose Chatsum, a chat summarization system, which creates a summary of chats from the previous encounters for each patient. Chatsum can assist the doctors to serve better by significantly reducing the cognitive load and highlighting important information. We applied a variety of natural language processing techniques including a medical domain-specific annotation and information extraction tool cTAKES, to identify key information in conversations and trained a classification model to identify sentences to be included in the final summary. The usability and the quality of chat summaries generated by our tool have been validated by real doctors serving patients online using the current platform. We also investigated the importance of each feature type in our method and assigned an importance score to each sentence in a chat to create a visualization to reduce the cognitive load.