One part of being a content writer for future technology and science-related topics is reading through long research papers on various topics published by scientists and researchers. However, after the first two paragraphs, I often hear the Windows shut-down tone in my head (not kidding!). So, for people like me who like to read short summaries rather than long research papers, a team of AI-scientists has developed a model that can take a research paper and present it in the form of a short gist.
Researchers at the Allen Institute for Artificial Intelligence developed this new artificial intelligence (AI)-based model that can convert a long text document into a short summary and present it in a TL;DR (Too Long; Didn’t Read) format. It is a pretty useful tool as it provides users with only the relevant information from a long passage of text.
Now, to develop the model, initially the researchers “pre-trained” it on the English language. Following this pre-training session, they formulated a SciTLDR data set that contained around 5,400 summaries of research papers on computer science.
Moreover, to make the model less dependent on domain knowledge while summarizing, the researchers further trained it on more than 20,000 titles of research papers.
So, this new summarizing model uses its training and artificial intelligence (AI) to separate the significant parts of texts from the abstract, introduction, and conclusion sections of research papers. Then it uses these parts of the text to create a short summary of the paper.
In their initial tests, the researchers found that the model was able to present documents of over 5,000 words in 21-word short summaries.
The team rolled out the summarizing AI-model to the Institute’s own Semantic Scholar search engine. However, it only works on papers related to computer science topics, as of now.
You can try it out on the Semantic Scholar search engine and read more about the model in the official paper published by the team.