Paper Dissected and Recap #4 : which BERT for long text ?

Nowadays, BERT appears everywhere. It could vastly outperform LSTM thanks to avoiding the sequential dependency modeling by allowing each token in the input sequence to attend independently to every other token in the sequence (self-attention mechanism). However, this advantage, which brings its success, is also its bottleneck. The full self-attention operation limits the capability of…More

Paper Dissected and Recap #1 : “Joint Aspect and Polarity Classification for Aspect-based Sentiment Analysis with End-to-End Neural Networks”

As anyone working on Sentiment Analysis, one day they will want to dig deeper into a more fine-grained sentiments analysis of each aspect* (for example in restaurant, they are: food, service, ambiance,…) instead of solely predicting positive, negative, neutral sentiment of a piece of text. The problem thus becomes a little bit more complex, it…More

About me

Currently Applied Scientist / Research Engineer at, I obtained the Ph.D. in Computer Science from the University of Lorraine, fortunately worked under the supervision of Christophe Cerisara and Claire Gardent (CNRS). My research interests during the Ph.D are neural models with focus on summarization, text generation, information extraction and sentiment analysis tasks. During summer and winter…More