Contextual text prediction for Microsoft Teams and Outlook

We are very excited to announce our 17th Data Science Meetup. On Tuesday, 22.06.2021, from 18:00 to 19:00, our colleague Stojan Trajanovski will give an interesting talk about "Contextual text prediction for Microsoft Teams and Outlook"

Bio:
Stojan Trajanovski is an applied & data scientist at Microsoft, working on large-scale machine learning models for different AI scenarios such as AI features for Microsoft Office (e.g., Outlook, Teams). His research focuses are applied machine learning (at scale), optimization algorithms, network science & complex networks, game theory and applied graph theory. He was a research scientist at Philips Research in Eindhoven, The Netherlands working on product transfers, patents, or publications in the medical AI domain from 2016 to 2019. Previously, he also spent some time as a postdoctoral researcher at the University of Amsterdam (UvA) and at Delft University of Technology in The Netherlands. He obtained his PhD (with cum laude honors / given to 5% graduates) from Delft University of Technology (TU Delft) in 2014. He graduated with distinction from the MSc. program in Advanced Computer Science at the University of Cambridge, United Kingdom in July 2011. He holds an engineering degree from Ss. Cyril and Methodius University in Skopje. He has published more than 40 papers at renowned journals and conferences such as: IEEE/ACM Transactions on Networking, Physical Review, IEEE Transactions on – {Mobile Computing, Parallel and Distributed Systems, Control of Networked Systems, Biomedical Engineering}, NAACL, IEEE Infocom, IEEE CDC, AAMAS, ICNP, NeurIPS workshops and others. Even before, he successfully participated at the International Mathematical Olympiad (IMO), winning a bronze medal.
See https://tstojan.github.io for more details.

Abstract:
Email and chat communication tools are increasingly important for completing daily tasks. Accurate real-time phrase completion can save time and bolster productivity. Modern text prediction algorithms are based on large language models which typically rely on the prior words in a message to predict a completion. We examine how additional contextual signals (from previous messages, time, and subject) affect the performance of a commercial text prediction model. We compare contextual text prediction in chat and email messages from two of the largest commercial platforms Microsoft Teams and Outlook, finding that contextual signals contribute to performance differently between these scenarios. On emails, time context is most beneficial with small relative gains of 2% over baseline. Whereas, in chat scenarios, using a tailored set of previous messages as context yields relative improvements over the baseline between 9.3% and 18.6% across various critical service-oriented text prediction metrics.