My thesis

My thesis aims to study how representation learning methods can be applied to user- generated data in order to extract relevant user representations. The application we are interested in is the generation of user profiles using documents a user may have in his personal document repository.

Firstly, I intent to formalize this task as the generation of a synthetic review made of relevant units selected among documents written by the user on different occasions. This bears similarity with automatic text summarization. The goal the extractive summarization systems, in the multi-document scenario, is to extract sentences out of a set of documents that are relevant with respect to a given information. The main difference with the latter is, however, that the user profile should be modeled in the extraction module and that the extraction unit might not be a sentence.

My collaborations

Throughout my research activities I have developed valuable collaborations with people I appreciate and who have influenced me a lot.

My projects

I have spent two years working as a software engineer in the BioASQ project, in LIP6 in Paris. During those years I developed the BioASQ Participants Area which is an online platform that serves the data for the challenge, performs the evaluations of the participants' submissions and provides APIs for easier exchange of data.

I am still following the challenge by supporting it in my free time. Amongs others, a main reason is that I am deeply interested in the topics of text classification. Also, in this framework, we have collected high quality data that serve my research purposes.