Jim Longo, VP Product & Development, HighWire
AI. and machine-learning are some of the biggest buzzwords in publishing today, and we expect to see these technologies disrupting the industry massively over the next few years. If you ran a search on Google with these terms, you would be presented with almost 400m results. So what is a publisher to do?
According to a report released earlier this month by UNSILO on AI in Academic Publishing, 70% of respondents are using or trialing some form of AI tooling, and 45% of those who are not plan to introduce it over the next 12 months. Despite this, almost half of the industry has no dedicated machine-learning staff, and 10% of respondents whose organizations are using AI were unable to say what AI was being used for.
The challenges are multiple, including a lack of expertise in this area in the publishing community. Not entirely surprising, given the newness of the technology, the speed of the industry’s reaction to change, and the demand for talent.
In fact, I would argue that publishers should not focus on building up in-house specialists, but rather look to their technology partners and vendors to identify and incorporate best-of-breed third-party tools and services. Publishers have real business challenges that they’re trying to solve, and their focus is and should be on these; it’s of little importance to editorial staff on a day-to-day basis how those problems are solved, as long as they are solved.
Our role at HighWire is to seek out the best minds in AI who have practical solutions to these problems, trial efficacy of their AI offerings, and then built partnerships that enable us to plug these services and tools into our open platform. By maintaining an approach that favors openness and interoperability, both we and our customers can benefit from best-of-breed specialist technology without needing to build up dedicated resources in-house from scratch.
Here’s a brief overview of some of those best-of-breed A.I. partners we’re currently working with, and the use-cases we’re tackling together:
Meta is a program funded by The Chan Zuckerberg Initiative that we’ve been working with for several years.
- Challenge: One project that we’re working on is around horizon scanning: the identification of upcoming topics and trends. There’s so much material being published that it’s impossible for any reader or editor to track it all manually, let alone identify trends or hot-button topics.
- Solution: We’ve created a dashboard for publishers in conjunction with Meta that will bring forward and pull out trending content. In our experience so far, A.I. is good at picking out winning or losing trends, but those topics and papers in the middle of the bell-curve need more refining. Advances in machine-learning and algorithms will greatly increase that predictability over time.
Our newest partner, Semantic Scholar, funded by The Paul Allen Institute for Artificial Intelligence and is an AI-driven discovery tool.
- Challenge: With so much content constantly being published, researchers often fall prey to information overload and miss important new developments. Publishers can also struggle to keep on top of large content sets, for example, disambiguation around authors and institutions.
- Solution: Semantic Scholar ingests content from academic papers, runs a knowledge graph and machine-learning against the scientific corpus of information and creates a user interface that allows researchers to discover and expand their perception of topics. Unlike other discovery tools, they have APIs available to our publishers, so our publishers will be able to bring the content that Semantic Scholar has indexed and enhanced and apply it back to their corpus in order to improve experiences within their own website.
YewNo is an A.I. and analytics based company.
- Challenge: As revenues are squeezed and retaining reader attention becomes harder, publishers need to drive discoverability of useful and relevant content in order to show their value, and cross-sell monetizeable content to new audiences.
- Solution: YewNo works with publishers to analyse their content and extract concepts from it, which can then be appended to articles as expanded metadata to improve search and discovery. In an experiment with a book publisher where an average of 75 key terms were extracted from each book, the publisher saw improved discovery of those publications and an increase in revenue generated on those titles. We are currently running a trial with one of our academic publishing customers utilizing this technology.
UNSILO uses A.I. to automate and improve the manuscript review and submissions process.
- Challenge: At present, there is a lot of manual work needed to ensure a submission contains all requirements and components, which is done by editorial staff. This can take 5-10 minutes per paper per submission: not much on its own, perhaps, but a huge time-suck for a journal at-scale.
- Solution: We’re working with the folks from UNSILO on a number of projects, utilizing their Manuscript Evaluation tool to automate manuscript screening upon submission. Freeing up that time for editorial specialists means they can spend more time on value-add activities.
- Challenge: Expanding the pool of potential reviewers is a real problem for publishers. The demand for reviewers is skyrocketing and editors tend to fall back on their favourites and pick up the same people again and again, so this becomes a more limited knowledge-set over time. This not only drives bias, but also constrains the potential of scientific endeavor.
- Solution: We’re excited about the early results in our publisher trial using UNSILO’s Reviewer Finder, which looks at papers that have been published, looks at contributing authors, looks at the contributing authors’ network within a knowledge graph, and then finds and identifies people that are non-obvious reviewers and suggests them to the editors.
Access Innovations are specialists in taxonomy management.
- Challenge: The sheer size and scale of data and information held by some publishers, particularly in STM, can make helping researchers find, discover and navigate content effectively difficult – not to mention potentially adding huge manual overhead for editorial staff, who have to maintain diverse tag sets and taxonomies.
- Solution: Following a successful project working together to relaunch McGraw-Hill Education’s flagship site, AccessEngineering, we are currently exploring how we can leverage Access Innovations’ deep experience in taxonomies and AI to identify MeSH terms within articles and help map the articles identified onto a medical publisher’s internal taxonomy.
HighWire will continue to trial and report on different use-cases for AI, text and data mining and machine-learning, to investigate what best meets the current and future challenges of the sector. Improvements and integrations will be incorporated in future deployments to customers as part of our ongoing future-facing R&D roadmap. If you’re interested in any of the above partnerships or capabilities, please reach out to me or to your Account Director to discuss this further.