Machine-generated visual summaries in a crowded research space

Machine-generated visual summaries in a crowded research space

One of the most immediate and effective ways to showcase new research is to display it as a poster at conferences. Presenting your research in a striking, visual format can be a powerful way to get audience attention. The challenge is, even with a captive audience, the time an author has to pique the interest of peers and publishers is limited. And now that authors are having to compete for attention at virtual conferences, the problem is exacerbated. The need for clear, concise packaging of research to communicate the context, key findings and results to your peers is even more important when the space to display this information is reduced to the size of a laptop screen.

And even when in-person conferences are possible again, another challenge for authors is getting mileage out of the poster after the event is over. The investment of time and energy needed to crystalise an extensive piece of research, months or even years in the making, into an attention-grabbing headline and a couple of hundred words is significant. Yet there’s huge potential to re-use poster contents well beyond conferences.

Condensing complex research into accessible snapshots accompanied by images isn’t only beneficial to specialists searching for emerging research in their field, it can also play a really important role in making scientific developments more available and accessible to lay readers.

Realising the full potential of a poster

When it comes to the contents of the poster itself, authors face a number of hurdles, one of the most common being selecting the material that will have the most impact and best convey the essence of the research. We asked authors: What do you find most difficult about producing a poster to showcase your research? Here’s how they responded1:

Survey conducted by Scholarcy on LinkedIn & Twitter, January 2021


Harder still, is getting the material to fit the space available. Because they don’t want to leave anything significant out, authors often make the mistake of cramming the space full of too much text or too many images which, rather than encouraging the audience to linger, gives them that well-known ‘tl;dr’ feeling and they quickly move on to the next one.

The general rule of thumb is that audiences “will not approach your poster if its subject is not clear from 3 metres away” and “all elements should be visible from at least 1.5m”.2  Most readers will only give a poster a few seconds of their attention before deciding whether or not it’s of interest. Another challenge is getting the headline right. If the wrong handful of words are selected here, this can be the first thing to turn audiences away.

The risk of turning people away with a bad poster layout, a weak headline or too much text is bigger than a missed opportunity from a conference. There’s a huge amount of high-quality research that has the potential to have a significant and positive impact on health, sustainability, climate change and much more that can quite easily be lost behind a badly presented poster. If novel insights and findings aren’t conveyed in an immediate way or get over complicated with too many words, the consequence is that important work is lost or overlooked. The risk is just as great with badly produced virtual posters as it is with physical ones.


The risk of getting it wrong

Researchers are competing in such a crowded space today, that without some visibility and traction for their work, the chances are that they will give up and put their efforts into something else. Success or failure can hinge on how well they communicate the importance of their findings and how much this resonates with the wider community. One aspect that might hamper this communication is determining what to include in a poster without being too subjective. What if the results that are being highlighted are not the most significant? Or an important aspect of the methodology, that could be the difference between someone paying attention or not, is missed. The impact of this quickly snowballs – overlooked studies; lost, potentially significant findings; and demoralised academics – often because the optimal information wasn’t selected to promote the work. Worse is that the poster could give a misleading picture about the research which can result in it being misinterpreted by the scientific community, media and wider public.


Shining a light on quality research: how templates and tech can help

Authors of research, for very good reasons, are often too close to their work to accurately and dispassionately condense it. Not only that, but it could be argued that agonising about headline summaries and the ‘right’ content to include under section headings is not the best use of their time. Campaigns such as #betterposter founded by Mike Morrison, Ph.D. candidate in organizational psychology at Michigan State University, are helping many researchers address this challenge. A clear and simple poster template can do a lot to focus the mind and shine a light of the quality of a piece of research.

Example donated by @drElsje (Better Scientific Poster)

You only have to search for #betterposter on Twitter to see how this campaign has helped the research community. The power of a clean, simple formula to draw the most impact from complex work cannot be underestimated. But it still doesn’t address the challenge to authors of condensing their hard-won results into a few succinct bullet points.

This is where natural language processing can come into its own. Applying machine intelligence to parse complex research literature and distil it into succinct summaries and key findings accompanied by an appropriate image and then present this in the poster layout shown above, which the author can then easily edit, is now possible.


Automated poster generation technology has arrived

Automated poster generation technology can not only remove the time and frustration of presenting work in this format for an author, it can also objectively extract significant information and present it in the most concise way. This has the potential to unlock overlooked or disregarded work that has been hidden behind a bad poster.

Speaking about his motivation to develop this kind of technology, Phil Gooch, founder of Scholarcy said:

“The key innovation of ‘better poster’ was to have the main headline or takeaway from the paper as a prominent part of the poster, with the rest of the poster summarising the key information on background, methods, results.

And I wondered whether this process could be semi-automated to help authors create these ‘better posters’ directly from their manuscripts. The idea was to allow authors to ‘write once, publish many’.”

So an author writes up their research once, and the same information can be used to:

  • create the conference poster;
  • create the paper for submission to the conference proceedings;
  • and then create other promotional material such as Twitter threads and plain language summaries

To achieve this, we leverage Scholarcy’s existing ‘flashcard’ technology and our new ‘smart synopsis’ tool engine. Our NLP engine already identifies the key takeaway in the paper, and we present this as the main headline to the poster. Our new infographics engine then identifies the figure in the paper that best matches this headline. If the paper contains no figures, then we locate an image on Wikipedia that best illustrates the headline.

This is just the start though. In future, we’ll be hooking up with OpenAI’s DALL-E engine to synthesise new images from the key information in the paper, allowing us to generate infographics for authors to use as a starting point for visual abstracts and more.

And we’re working with some key providers of scholarly communication technology to integrate this service into new products. Stay tuned!

Share this post