Machine learning and AI are ubiquitous these days, and people are finding all kinds of creative ways to use it in the realm of research. Here are a few ways Elsevier and our innovative collaborators are using it.
1. Machine learning can suggest new hypotheses for you to investigate.
Big data and machine learning platforms can also do more than fuel recommendation engines. They can become part of the scientific discourse by performing a task they’re particularly suited to – scanning vast amounts of information to unearth connections that lead to new hypotheses.
A pilot program between Elsevier and the Euretos AI platform aims to use these technologies to scan millions of journal articles and hundreds of databases to make connections and suggest new hypotheses for researchers to investigate. he project explores the power of combining big-data analytics with the human factor — aiming to find workflows in which humans and machines work together to accelerate the pace of research.
2. Our careers site uses machine learning to suggest relevant research positions
Mendeley Careers finds another use for machine learning: alerting researchers to the jobs that are most relevant to their interests and expertise. For most researchers, finding a new position means having to monitor everything from social media to jobs sites and the channels of prospective employers. However, Mendeley Careers can recommend open job postings based on your reading history. The site has more open job ads than any other academic publisher or research platform. I you choose to use it, it can draw on the abstracts and titles of articles you are reading in Mendeley to recommend highly relevant jobs without you having to do a thing.
3. Harvard scientists are using machine learning to spot duplicated or manipulated images in scientific articles.
Supported by Elsevier, a team of data analysts and research integrity experts at Harvard Medical School are working to develop a tool that can objectively indicate whether an image has been manipulated or is a duplication.
A manipulated image can be an indication of potential misconduct – it might be stretched, cropped, lightened, inverted or duplicated to change the meaning of the research. More commonly, it’s a symptom of bad data management. In either case it’s important to identify these instances early in the submissions process, as they can have significant consequences.
That’s no easy process, and the current solution is manual, with people combing the published literature looking for discrepancies in published images to see if they have been tampered with, or whether the same image was reused to represent the results of different experiments. Their assessment may not be as systematic and objective as it could be if a computer were undertaking the task. Plus, computers can sometimes detect anomalies the human eye cannot.
With input from Elsevier’s Research Integrity team, Harvard Medical School’s data analysts are developing a tool that uses machine learning to provide more objective information. Specifically, the tool is trained on a well-defined set of pairs of duplicate images (positive controls) and pairs of distinct samples (negative controls). This enables the algorithm to generate probability scores indicating whether two test images are too similar to be derived from different samples, without any further input needed from the user.
4. Personalized recommendations are helping scientists manage information overload.
Keeping up with scientific literature has been described as the single most important skill throughout a researcher’s career, with 28,100 active peer-reviewed journals and 2.5 million new articles published each year, according to The STM Report. However, just as auto-generated personalized recommendations have changed the way people find and interact with music and film, they now have the power to change the way researchers discover content and stay up to date and researchers electing to sign up for ScienceDirect Personalized recommendations have an alternative way to discover the content that is most relevant to them.
By applying machine learning to ScienceDirect’s usage logs, the recommender can model relationships between research articles. These capture not only the obvious relationships but also reveal more subtle ones, such as discovering that two articles from different domains that use different terminology are related to one another. The approach can help researchers make serendipitous discoveries that would have been difficult to retrieve using a traditional search engine.