NERD for NexGenTV

My graduation thesis “NERD for NexGenTV” documents the contribution I gave to the NexGenTV project during my internship at Graduate school and research centre EURECOM. The project aims to enrich the experience of the television viewer on a support device, enriching the TV transmissions with complementary information. My thesis focuses on two NexGenTV subtasks: Named Entity Recognition (NER) and Named Entity Disambiguation (NED): the final goal was to automatically detect, recognize and link all spottable named entity presented in a corpus of subtitle transcripts of French political debates. My work presents two multilingual ensemble methods that combine the responses of web services NER and NED in order to improve the quality of the predicted entities. Both methods represent the information got by the extractor responses as real-valued vector (features engineering) and use Deep Neural Networks to produce the final output.