September 27, 2019
gk was proud to take part in the AI & Data Science in Trading Conference in London, held September 16-17. Anthony Tassone, gk’s CEO, chaired the Natural Language Processing stream on day one of the conference which included presentations about:
Teaching machines to understand cyberslang – Mike Chen, Director of Dynamic Equity at PanAgora Asset Management.
Unlocking internal data using NLP – Peter Hafez, Chief Data Scientist at RavenPack.
Following these presentations, Anthony moderated a panel discussion about challenges and opportunities in widespread NLP adoption with Mike Chen, Peter Hafez and Bennett Saltzman, Senior NLP Data Scientist and Lead Model Developer at Amenity Analytics.
The panel discussion uncovered many interesting insights and opinions about natural language processing including:
- The adoption of NLP and its growth over the last few years
- How it is applied from one organization to another
- What problems AI experts are helping to solve through NLP
- What outputs clients are looking for
- The challenge of accessing real data and obtaining approval from legal and compliance teams
Here are some of the highlights captured from the discussion:
What outputs clients are looking for
“The sell-side is leveraging NLP to automate tedious tasks and to generate insights on their customer data. This is one of the biggest priorities of firms. They’re being bombarded by the buy-side and they have a ton of unstructured data hitting these desks and they can’t cope with it for a number of reasons. There’s a lot of data just lying on the ground from audio files to Bloomberg chat data and we’re seeing a ton of effort toward operational alpha to extract quotes and trades from these conversations so that we can make meaningful decisions later. It could be things like just understanding what percentage of my conversations are social or why doesn’t a quote turn into a trade? That’s probably the number one question we see from the sell-side.” – Anthony Tassone
Gaining a competitive edge
“There’s no doubt that a lot of firms have to go through the process of saying let’s try and build it because that’s how we create our competitive advantage, we don’t want the same NLP engine as everybody else. And then they realize, the business side is now asking us to deliver but we’re still several years away from it, we should probably try to get in a vendor and deliver something and then we can start sophisticating the overall internal offering by layering our own stuff on top of it.” – Peter Hafez
Converting data into alpha
“There’s a lot of steps in converting data via NLP into alpha. First of all, you’ve got to clean the data, you have to detect the entity, you have to detect the event, you need to link all the events together and all the entities together. And you don’t need just one data set, you need multiple data sets to extract alpha and so linking all of these together is a huge task.” – Mike Chen
Challenges of hiring data scientists
“It’s a challenge internally and externally for some of our clients who want to onboard someone who can be a champion in-house to manage that relationship with us. Internally I think what we really look for is a smart, curious person, I don’t think that’s too complicated of a personality to fit. We actually have a unique position in NLP, we take people who are domain experts in given fields so whether that’s finance, insurance, media; train them up on our NLP platform which takes six months, and so it’s a big investment for us; and then once we have been that educator, turn them loose and let them start creating models for different tasks. So, our approach doesn’t rely on people coming in with a significant background of AI expertise.” – Bennett Saltzman
It was a fantastic conference and gk was glad to take part and lead the NLP stream.
In case you missed it, here’s Anthony’s pre-conference interview with the team behind the AI & Data Science in Trading London event.