Artificial Intelligence has now waded into the scientific literature domain. Iris.ai, a literature-exploration tool is the latest AI-based search tools in the town that offers targeted navigation of the knowledge landscape. This AI-based speed-reader came to life to keep a check on abundant scientific literature which is published on a daily basis world, with a statistical count of 2 papers every minute. The core problem that Iris AI is trying to solve is to reduce the time frame to parse and contextualise the sheer volume of scientific literature that’s being published.
While there are existing digital services making scientific literature searchable via keywords, such as Google Scholar, there are fewer options for navigating digital content in a way that automatically relates relevant research to reduce time. This is where Iris’s role comes into play.
How Iris Works
While conventional tools can only act largely as citation indices, AI-based ones have the capability to deliver a deeper view of the literature.
- Iris.ai has the capability to return a map of thousands of matching documents which are visually segregated by topic just by using a 300 to 500-word description of a researcher’s problem or the URL of an existing research paper.
- The results that it provides is much quicker and gives a precise overview of what should be relevant to a specific research question.
- Iris.ai groups multiple documents into varied topics defined by only the words they use.
- Iris.ai drills the connecting repositories collection or in short CORE collection, a searchable database of more than 134 million open-access papers, as well as journals to which the user’s library provides the overall access.
- Iris.ai then blends about three algorithms to create ‘document fingerprints’ that reflects word-usage frequencies, which are then used to rank papers according to its significances.
- The algorithms powering such tools typically perform only two functions that are it extracts scientific content and provides advanced services, such as filtering, ranking or grouping search results.
- In short, it uses a non-semantic neural topic modelling approach with a heuristic function for hierarchy building. It fetches the abstract from the scientific papers and starts with frequency analysis and pulls out the most essential words and expressions from that paper and then each of those words turned into a vector, in a multi-dimensional vector space.
- What Iris does is that it goes and finds related words from this whole other body of content, so it works with both synonyms and hypernyms and clusters these words and these bag of clusters of words are related to keywords found in text and finally the calculation of what is the most applicable label for this bag of words from clusters is done as a result.
According to Giovanni Colavizza, a research data scientist at the Alan Turing Institute in London, algorithms extracting scientific content leverages natural language processing (NLP) techniques, which seeks to interpret languages as humans use it. For example, developers can use supervised machine learning which involves tagging entities, such as a paper’s authors and references in training sets to teach algorithms to identify and extract them.
Colavizza further mentions that to provide more-advanced services, algorithms often construct ‘knowledge graphs’ that detail relationships between the extracted entities and show them to users. For example, the AI could easily suggest that a drug and a protein are related if they’re mentioned in the same sentence. The knowledge graph encodes this as an explicit relationship in a database, and not just in a sentence on a document, essentially making it machine readable.
- Even though Iris provides a result which is a map of related papers, the company is also planning to supplement those results by identifying hypotheses being explored in each paper as well
- It is also developing a parallel, blockchain-based effort called Project Aiur, which seeks to use AI to check every aspect of a research paper against other scientific documents, thus validating hypotheses
- Although Iris.ai is free for basic queries but costing upwards of $ 23,000 a year for premium access, which allows more-nuanced searches which thereby accelerates researchers’ entry into newer fields
Till now experts have been using free AI-powered tools such as Semantic Scholar which functions like Google Scholar.