A not-for-profit organisation inspired by research funders and led by scientists.
Part of the organisation’s mission is to bring open access and fast peer reviews to the science community, as well as to actively support researchers early in their careers. For most researchers the process of publishing their research will be frustratingly familiar: repeat submissions to publishers, followed by months of waiting with no guarantee that their paper will be accepted.
Those hoping for peer review through an organisation such as our client face better odds, but the process is still less than optimal. An editor is faced with finding ideal reviewers for the paper from a database containing 40,000 different individuals that, after filtering by subject area, reduces to about 2,000. Adding keywords to the search refines the population further, but still with no way of knowing how busy a potential reviewer is. The organisation asked if we could develop a more efficient way of modelling the data to identify potential reviewers more effectively.
We created a model that combined existing keywords with analysis of the natural language of the actual manuscript. Using a generative statistical model (latent Dirichlet allocation) and the doc2vec algorithm, the project created a suitability score that could be fed into the recommendations.
In the user interface Faculty created, an editor can easily see potential reviewers. Numbers illustrate suitability, a ‘clock’ shows the average speed of review, while reviewers who are early career researchers are highlighted in a different colour.
The model allowed editors to upload a manuscript and receive an overview of potential reviewers. Each name returned had additional data, including a suitability score, an indication of review time and whether they could be identified as an early career researcher.
The project gives editors a much quicker, more direct route to finding better reviewers. It helps established reviewers manage their time better, allows early-career researchers to gain recognition and, lastly, reduces the time a postdoc researcher spends waiting for their work to be reviewed.