Interview with Joan Plepi, a PhD candidate in deep learning for NLP

Joan Plepi Interview PhD candidate

Today we interview Joan Plepi, a PhD candidate researcher in our lab, focusing on user personalization techniques and applying them to Natural Language Processing problems on social media, or conversational systems. Read on to learn more about his exciting research path, his interests, and life in Marburg.

Could you explain your research to someone outside your field?

I am currently working on user personalization for different NLP tasks mainly in social media, like sarcasm detection, hate speech detection, misinformation spreading, conversations etc. The main goal of my research is to understand how users’ personal background, writing style, opinions, perception of different situations, and interaction with other peers, affects the performance of the NLP models in the tasks mentioned above. I believe that these users’ characteristics have different influences based on the different attributes of each NLP task. Hence, I am working on different ways to model the users’ characteristics on different levels, and their interactions with the audience by using graph neural networks, and analyzing these effects on downstream NLP tasks.

What excites you about it?

First, it is a relatively new area of NLP, and there is a lot to explore. I believe that working on personalized models for different tasks helps in improving the explainability of models output. Another thing that makes it interesting is that you are working on real-life scenarios and data, you get to understand how people’s social interactions with one another, impact their opinions or perception of different situations or topics. And yeah, there is a lot of problem-solving and analyzing, programming, and nice discussions with my colleagues.

What’s your research path, how did you get into the field?

I finished my bachelor’s in Computer Engineering in Tirana, and continued my master’s in Bonn for Computer Science. During my master’s, I focus on machine learning, and deep learning mainly on vision systems and robotics, as well as knowledge graphs. During my thesis, I decided to focus on natural language processing, because in my opinion language is complex to analyze and is quite dependent on many factors. In addition, I was also working on several projects at Fraunhofer Institute involving named entity recognition, question answering, and developing a framework with several pipelines in Java to process data in different formats, and use those to train and evaluate NLP models. While my work during the thesis developed my research skills, my job at Fraunhofer helped develop my technical skills and gain knowledge of different NLP tools and libraries.

What do you like at the lab and Marburg?

What I like the most at our lab, is the friendly environment that we have created. We are not only colleagues that discuss our work and our research topic, but also friends that talk about many things outside our work during our free time. On the other hand, Marburg is a pleasant small town, with various attractive places and its lovely river where you can always go for a walk. In addition, it has the Oberstadt with the pleasant traditional German houses and its wonderful castle from where you can take some nice pictures of the city from above.

What keeps you busy outside your university job?

During the week, outside the university job, I like to take a run along the river, go to the gym, or play basketball with my friends. During the weekend, I enjoy going on trips across Germany or hiking.

What would you like to achieve? What would you wish for your research field?

I would like to advance my research in the area that I am working. My wish is to assist in creating and making AI models, that help and support us in our everyday life in various aspects.