Using AI tools for content alerts

A typical journal ETOC selector screen

Probably for as long as journals have existed, publishers have tried to keep potential readers informed of new issues. Although details of new journal issues are delivered digitally, as ETOC alerts (the acronym for Electronic Table of Contents Alerts), in truth there is not very much that is digital about the information. Is there a way they can be improved using AI?

Just to refresh your memory: with each new issue of a journal, an ETOC alert is mailed to all users who have registered an interest. In principle, this makes admirable sense. There is ample evidence that academic researchers use search and discovery tools to find out about new content. One of the most remarkable statistics from the most recent survey of How Readers Discover Content in Scholarly Publications is that 45% of articles read by academics were found via discovery – in other words, not by references from the papers they read, or from word of mouth. There is an element of serendipity, since the researcher doesn’t know what she is looking for, but it is within a domain of interest.

ETOC alerts attempt to provide this kind of serendipitous discovery, Unfortunately, an academic’s area of interest will typically be considerably narrower than the scope even of one journal. The chances are that of all the articles in one issue of a journal, the researcher is typically only interested in one or two. Unfortunately, using ETOC alerts, there is no way of refining the choice presented: either you sign up for all the articles in the journal, or you get no alerts at all.

But is there a better way of doing things? The UNSILO concept extraction system, which automatically identifies concepts from articles in a corpus, enables a very precise match of articles with other articles: not just “cancer”, nor even “lung cancer”, but, say, “squamous cell carcinoma”. Using this precise matching system, the tool can be used to find other articles or chapters that are about that same topic. If the system is “fed” with one document, it can find other related documents, and in addition, it can filter by recency to show only papers that have been published in the last few weeks or months.

Even this system has its drawbacks. When as a user you are confronted with a lengthy list of terms to choose from to select your specialty, you soon realise this process is both time-consuming and still not sufficiently detailed. Plus, there is the problem that researchers have to be able to state their interest in terms that the system understands (they might think of themselves as a specialist in “pulmonary disorder”, while the system files this topic under “lung diseases”).

Selecting your specialty via a dropdown menu: typically not detailed enough for precise alerts to be provide


Better still is when the researcher simply designates a recent article or articles they have read and asks the system to find “more like this”. No taxonomy is required; all the matching is done in the background by comparing concepts in the article.  

In this way, the traditional ETOC alert can be transformed into something far more precise and personalized. The result is more relevant for the researcher (instead of receiving, say, the titles of ten articles, of which only two are relevant, they receive just the alerts for the relevant articles), and also provides an advantage for the publisher. With an ETOC alert that lists, say, ten articles, it’s likely that 80% of the recommendations will not be relevant. Using the UNSILO tools, that figure can be reduced to a few percent at most.

In this way, alerts for readers can move from the spam folder to the most-accessed emails in a researcher’s inbox.


Receive an email every time we publish a new blog post