Refinitiv Labs Preps for SentiMine Release

Refinitiv Labs—the innovation division of Refinitiv—is working to make SentiMine, a discoverability tool for unstructured content, available to users on the Eikon terminal and Refinitiv’s workflow solution, Workspace.

The tool, catered to investment managers and investment bankers, is currently available to about two dozen users under early-access. The aim is to eventually support hundreds, then thousands, of users, said Sanjna Parasrampuria, head of Refinitv Labs in Asia, who was speaking on the Waters Wavelength Podcast.

SentiMine helps users review and interpret unstructured financial content such as equity research reports, news, company filings, and earnings calls, using topic modeling, natural language processing, and deep learning. Parasrampuria said that in the interim, the lab is seeking feedback from its early-access users as it scales the product. 

“In the beginning, I had 50 customers who said, ‘Hey, we heard about this, can we look at it?’ [but at the time we had to say], ‘No, right now I can’t support 20 users at a time on this.’ This was from several months ago. Since then, we have already taken several steps and strides forward. So, we will be speaking with more customers to have access to this at an early access version soon enough. My team is working day and night to get that piece out so that more people on a request basis are in a position to get their hands on this,” she said. 

Refinitiv Labs will continue scaling the product until it’s official release, which will be in a few months, at which point the team should be positioned to support 500 SentiMine users concurrently on the app.

Refinitv Labs is aiming to make SentiMine first available on Eikon, where, according to Parasrampuria, most of its buy-side community is already active, and then later on to its Refinitiv Workspace workflow solution, which is seeing more sell-side play.

“Workspace is the evolution of Eikon, and it has a later version of UX experiences, interface, and workflow integration that will make it a lot more intuitive. We’re hoping that we’re in a position to provide SentiMine to our investment banking community in Workspace, which is already moving there,” she added.

The Nuts and Bolts

Refinitiv Labs spent nine months working on and developing SentiMine, starting with interactions between the data giant and a handful of existing clients. Parasrampuria said the goal was to put all the unstructured content for equities in one place, and surface key topics and sentiment. 

This helps users like research analysts and portfolio managers who scour these types of datasets to focus on acting on the information, rather than “just stay on top of it.”

“Most of the time, people read the headline and never have the time to get into the details. Here, you’re in a position to prioritize your time significantly,” she said. 

Using SentiMine, asset managers can view the key topics impacting their portfolios and sentiment surrounding those themes. Then, from a signal standpoint, they can compare any contrarian sentiment against the consensus view.

“It’s not just at face value; it shows the exact sentence that makes me believe this is why the person has this type of view,” she said.

The tool aims to help asset managers better determine whether they should continue holding their positions, invest more, or hedge against emerging risks. Across the aisle, investment bankers can get a quick, aggregated view of a topic or stock to help them prepare for pitching an idea of a company they want to cover or help with onboarding from an M&A scenario.

At the heart of SentiMine is a customized version of Google’s transformer-based model BERT—or Bidirectional Encoder Representations from Transformers—which allows the model to gain context when understanding text in documents. 

“Obviously, this is also a challenge for many other companies trying to do the same things. We can’t use things off the shelf because we sit in this nuanced world of finance, wherein our dictionary of speech is very unique to the world of finance. And many of these things that sit in the world of open source are built on not just financial context. So we had to custom-make a whole bunch of things. We created our own financial ontology that can be overlaid on top of this so that it’s sensible to a very advanced asset manager user who’s not going to be okay with us referencing operating cost as something else, for example,” Parasrampuria said. 

She added that SentiMine also uses entity recognition to identify what company is being talked about, as well as deep learning and neural networks to identify certain other pieces and layers of information that BERT isn’t able to solve alone.

She credited her team for the work done on SentiMine so far. 

“My team is a 100 times smarter than me and I’m working with the best scientists that understand tech and how they apply to finance. Most of them come from asset management and hedge fund backgrounds themselves, so they also appreciate it from the context of an end-user,” she said.