Flora Sakketou

My research activity is primarily focused on developing optimization algorithms and deep learning architectures and applying them to large scale Natural Language Processing (NLP) problems.

During my PhD, I explored the application of constrained optimization frameworks to deep learning models. By utilizing such frameworks, I was able to incorporate external knowledge into the learning process and enhance the performance of unconstrained methods. I developed algorithms that leverage various external sources of information and can be applied for natural language processing tasks, specifically for producing word embeddings and machine translation. Some of my work in NLP is inspired by my previous experience with recommender systems where I exploited the information that comes from the users’ social connections in addition to their rating behavior in order to provide better recommendations.

My latest research direction is towards user modelling and more specifically, identifying the nuances in a user’s written expression over time. Such subtle changes in language usage can indicate the change of their opinion over a certain issue or the change of their psychological state. Therefore my main aim is to design less biased, more socially balanced models by modelling the users’ psychological and sociological characteristics.