<html> <head> <meta http-equiv="content-type" content="text/html; charset=utf-8"> </head> <body text="#000000" bgcolor="#FFFFFF"> The App (<a class="moz-txt-link-freetext" href="http://www.litislab.fr/equipe/app/">http://www.litislab.fr/equipe/app/</a>) and MIND (<a class="moz-txt-link-freetext" href="http://www.litislab.fr/equipe/mind/">http://www.litislab.fr/equipe/mind/</a>) teams of the LITIS lab at INSA Rouen Normandy offer a postdoctoral position for 12 months as part of the SAPhIRS project. Keywords: machine learning, deep learning, recurrent neural networks Description of the project and postdoctoral missions: Social networks are regularly used to express opinions on public and political events or to disseminate opinions on sensitive topics (hate speech, hooliganism, racism and nationalism, etc.). The objective of the SAPhIRS project is to study opinion propagation within social networks: to identify the key mechanisms for disseminating information and opinion and to identify leaders of influence. Particularly in Security field, we will focus on Twitter detection and analysis of hamessages calling for hatred or violence, monitoring their spread and detection of influential actors. As part of this project, we propose a 12-month postdoctoral position in machine learning for opinion mining, sentiment analysis and the detection of changes of opinion in Tweets. For this purpose, we plan to use state-of-the-art methods in NLP based on deep-learning neural networks, and especially recurrent neural networks with internal memory such as LSTMs or GRUs. In other words, the main tasks would be: <ul> <li>To annotate tweets automatically according to an opinion: supervised classification problem;</li> <li>To automatically identify messages containing the expression of radical ideas, in English, in French and in Arabic chat alphabet (transliteration of Arabic in Latin alphabet, also called arabizi or arabish): problem of supervised learning on unbalanced classes and possibly weakly supervised learning;</li> <li>To detect changes of opinion in user Tweets sequences: detection of anomalies and breaks in a time series. </li> </ul> The main difficulties come from the encoding of input data (short texts from Twitter, in French and Arabizi) for which language models remain to be defined, and in the design and learning of adapted recurrent models dedicated to these three tasks. Profiles: Candidates must have a doctorate in Machine Learning with, if possible, experience in NLP and/or Deep Learning. Knowledge of recurrent networks and Arabizi would also be a plus. Contractual conditions: The contract will be 12 months starting as soon as possible, with a gross salary of approximately 3500 €. The recruited person will work in the LITIS lab at INSA Rouen Normandy in Saint-Etienne-du-Rouvray. Application: CV, motivation letter, recommendation letters. Contact: <a class="moz-txt-link-abbreviated" href="mailto:alexandre.pauchet@insa-rouen.fr">alexandre.pauchet@insa-rouen.fr</a> <pre class="moz-signature" cols="72">-- Alexandre Pauchet Associate Professor HDR INSA Rouen Normandie - ASI Department - LITIS Lab Phone: +33(0)2 32 95 98 58 Web: <a class="moz-txt-link-freetext" href="http://asi.insa-rouen.fr/enseignants/~apauchet/">http://asi.insa-rouen.fr/enseignants/~apauchet/</a></pre> </body> </html>