New paper out! Congrats to the lead author @CharlineBour, for this great work👏

This work represents the last part of my PhD in a collaboration of @Epiconcept, @Cesp_Inserm and @UnivParisSaclay👌 A huge thanks to my co-authors : @vkhetan_iit, @xtannier, Md Imbessat Hassan Rizvi, @tczerni, @stormlogo, @CharlineBour, Andrew Fano, @GFaghe 🙏🙏

Secondly, cause-effect pairs were identified in causal sentences with several models tested. Lastly, in a semi-supervised approach cause-effect pairs were aggregated to form a cause-effect network, which was visualised in D3. Check it out ->

A cause-effect dataset was manually labeled and augmented using active learning. First, sentences containing causal information (causal sentences) were detected by fine-tuning a BERTweet model.

Our paper on extracting both explicit and implicit cause-effect relationships in patient-reported diabetes tweets using deep learning has been published. Check it out 👇👇

Can we use social media data to emulate cohort studies for epidemiological research? This is one of the questions we are trying to address in the Deep Digital Phenotyping research lab at @LIH_Luxembourg But, why? [1/13] 🧶

30 millions de tweets américains passés au crible de l’#IA pour faire ressortir les grands thèmes associés à la détresse liée au #diabète selon les patients https://t.co/VWayI2svkH

Congrats to 👏 👏 👏 To Adrian Ahne @adrahne, having completed his PHD at Epiconcept, and who is now a Doctor in digital Epidemiology! 🙌 We are all very proud of his success! 👇

And a last thanks to all my colleagues at @EtudeE4N for their warm welcome and cheerfulness and the friends I met there which made those 3 years very joyful☺️☺️

An additional thank you to my thesis committee Dr. Tubert-Bitter, Prof. Bringay, Dr. Diallo (@gayodiallo ), Dr. Charles and Dr. Hulman (@adamhulman ) for having accepted to evaluate my work and their rich reflexions on my work and interesting discussions!

I would like to express my appreciation to Prof. Tannier @xtannier whom I met during my studies at @PolytechSorbonn and who continued to advice me on #NLP related matters in my thesis!🙏

A sincere thanks to my company supervisors @tczerni and @stormlogo for their rich advise during those years and I am particularly grateful to @Epiconcept for their financial support of this thesis but also the wonderful people I met there!🤗🤗🤗

Yesterday I successfully defended my thesis after three exciting and intense years🥳🥳 I would like to particularly thank my thesis director @GFaghe for his constant support, the freedom he gave me and outstanding supervision. What a journey!!

Tremendous thanks to my great co-authors: @GFaghe, @xtannier , @tczerni and @stormlogo ! And also a big thank you for the financial support for this project from @Epiconcept !!

Using active learning we reduced the manual annotation effort Due to the streaming character our methodology is memory efficient and capable of handling large datasets. Next step 👉 Test on a sample of end-users

In an iterative process a user acts on the tree, explores and discovers concepts until a user-defined clustering solution is obtained. Interpretability is increased through the human-system interaction and head words visualisation.

The underlying machine learning algorithm clusters documents in a hierarchical tree manner consisting of classification nodes (user-defined concepts) and clustering nodes (clusters documents based on head words that best describe the documents having passed the node).

A user interacts via an interactive user interface to increase user-friendliness.

New paper alert!! 🥳🥳 We proposed a methodology and developed a prototype for a clinical decision support systems to explore unstructured clinical text data and target topics of interest for health professionals without programming skills.

Always good to see the @ColiveVoice ad on screen ✊😎! Whether you have already participated or not yet, please 🙏 (re)contribute to develop vocal biomarkers for #digitalhealth monitoring. Everybody can participate, regardless of their health status. 👉 https://t.co/GHA4bK8T7x