TLDR or TL; DR – this AI sums up research in one sentence
New software provides one-sentence summaries of research papers. This should help scientists to sift through studies more quickly and find relevant results.
TLDR or TL; DR is a common Internet acronym for “too long; didn’t read «. And now also the name of a software supported by artificial intelligence that generates one-sentence summaries of research work. The result is fed by information in the abstract, the introduction and the conclusion of a paper.Different questions are come in our minds like what is TLDR or TL; DR.
The free tool has been active since this week on the Semantic Scholar search engine , which specializes in academic publications. Both projects were developed by employees of the non-profit Allen Institute for Artificial Intelligence (AI2). So far, the software has only generated sentences for ten million papers in the field of computer science, which Semantic Scholar covers. For publications in other disciplines, there should be summaries from next month, says Dan Weld, who heads the Semantic Scholar group at AI2.
Preliminary tests suggest the tool will help readers sort search results faster, he says. “People really seem to like it.”
The team of authors already described the tool in April 2020 in a preprint on the preprint server arXiv . The study was then assess at a conference that took place in November and has now been accepted for publication. The researchers have made their code freely available; and there is also a demo website where anyone interested can try out the tool.
Not perfect, but a good start
“This type of tool will become a standard feature of scientific search in the near future. Given the need, I’m actually amazed that it took so long to put it into practice, ”says Jevin West, an information scientist at the University of Washington in Seattle, who tested the tool at the request of Nature. “It’s not perfect, but definitely a step in the right direction.”
Snappy sentences on Twitter about various publications had inspired Weld to develop the TLDR or TL; DR program. Like other speech generation software, the tool uses neural networks that have been trained on large amounts of text. The team has worked through tens of thousands of research papers so the network has learned to generate succinct sentences. The researchers then refined the software to summarize the content. To do this, they used a new data set of a few thousand computer science papers with appropriate summaries. The team also collected non-technical examples to improve the software’s performance in 16 other areas. Particular attention is pay to bio-medicine.
In addition to TLDR or TL; DR software, there are other tools: since 2018, for example, the “Paper Digest” website has been offering summaries of work. However, it seems to extract key phrases from the text rather than generating new ones like TLDR or TL; DR, says Weld. So far, his program has primarily been aim at experts who already understand the technical jargon of a paper. However, according to Weld, the team is working on producing summaries for interested laypeople.
The researchers also plan to license the technology to publishers and expand their service to offer personalized research briefings. “We are now getting to the point where AI methods can create summaries that are acceptable to humans,” says Weld.