Machine Assisted Reading

Jan22 '17

with Zhe Yu, NC State


Machine assisted reading (MAR) is a project designated to reduce effort of systematic literature reviews (SLRs) with machine learning algorithms.


SLRs are extremely helpful in summarizing and understanding research works in target field. It is suggested that SLRs should be conducted frequently by every software engineering researcher. However, it usually takes a huge amount of time and effort (months’) to conduct an SLR. The objective of MAR is to reduce this effort and thereby facilitate the conduction of SLRs.


Our current results show that MAR can reduce the effort of primary study selection to about 1/10 by sacrifising 0.1 recall. The tool is available online here. Now we are exploring more possibilities:

  • How to reuse previous knowledge to further boost primary study selection?
  • Can we estimate the effort before the review process is done?
  • Is unsupervised learning helpful in MAR?
  • How to deal with concept drift in the review process?
  • How to allocate review tasks to and synthesize results from multiple reviewers?