|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Linking, navigation and analytics
Providing real-time insight during political debates in a second screen application
Participants : Vincent Claveau, Guillaume Gravier, Gabriel Sargent.
Joint work with Institut Eurecom, Wildmoka and AVISTO Telecom in the framework of the FUI project NexGenTV.
Second screen applications are becoming key for broadcasters exploiting the convergence of TV and Internet. Authoring such applications however remains costly. Within the NexGenTV project, we developed a second screen authoring application that leverages multimedia content analytics and social media monitoring. A back-office is dedicated to easy and fast content ingestion, segmentation, description and enrichment with links to entities and related content. From the back-end, broadcasters can push enriched content to front-end applications providing customers with highlights, entity and content links, overviews of social network, etc. The demonstration operates on political debates ingested during the 2017 French presidential election, enabling insights on the debates [12].
http://www.nexgentv.fr/communication/events/discover-our-new-politics-debates-live-video-edition
Information extraction in clinical documents
Participants : Clément Dalloux, Vincent Claveau.
Joint work with Claudia Moro (Pontifícia Universidade Católica do Paraná, Brazil) and Natalia Grabar (Univ. Lille)
Extracting fine-grained information from clinical texts is a keystone for numerous medical applications. For instance, in clinical trial protocols eligibility criteria are expressed through texts in an unstructured way. This year, we have developed an annotated corpus of clinical trials and made it available to the community. Based on this corpus, we proposed automatic methods to extract numerical information [20] and to handle the variation of the units used [43]. In such medical applications, detecting negation, uncertainty, and the scope on which they apply is important. Thus, we have also developed an annotated corpus, made it available to the community, and we have proposed an automatic tool based on recurrent neural networks [37], [41] and made it available as a web service.
Semi-supervision for information extraction
Participants : Vincent Claveau, Ewa Kijak.
Many NLP problems are tackled as supervised machine learning tasks. Consequently, the cost of the expertise needed to annotate the examples is a widespread issue. Active learning offers a framework to that issue, allowing to control the annotation cost while maximizing the classifier performance, but it relies on the key step of choosing which example will be proposed to the expert. This year, we examined and proposed such selection strategies in the specific case of conditional random fields (CRF) which are largely used in NLP. On the one hand, we proposed a simple method to correct a bias of some state-of-the-art selection techniques. On the other hand, we built an original approach to select the examples, based on the respect of proportions in the datasets. These contributions were validated over a large range of experiments implying several datasets and tasks, including named entity recognition, chunking, phonetization, word sense disambiguation [19].
Linking multimedia content for efficient news browsing via explorable news graphs
Participants : Rémi Bois, Guillaume Gravier, Pascale Sébillot.
Joint work with Maxime Robert, Éric Jamet (Univ. Rennes 2) and Emmanuel Morin (Univ. Nantes) in the framework of the CominLabs project Linking Media in Acceptable Hypergraphs.
As the amount of news information available online grows, media professionals are in need of advanced tools to explore the information surrounding specific events before writing their own piece of news, e.g., adding context and insight. While many tools exist to extract information from large datasets, they do not offer an easy way to gain insight from a news collection by browsing, going from article to article and viewing unaltered original content. Such browsing tools require the creation of rich underlying structures such as graph representations. These representations can be further enhanced by typing links that connect nodes, in order to inform the user on the nature of their relation. We propose an efficient way to generate links between news items in order to obtain an easily navigable graph, and enrich this graph by automatically typing created links. User evaluations are conducted on real-world data in order to assess for the interest of both the graph representation and link typing in a press reviewing task, showing a significant improvement compared to classical search engines [15], [16].
Multimodal detection of fake news
Participants : Vincent Claveau, Cédric Maigrot, Ewa Kijak.
Social networks make it possible to share rapidly and massively information, including fake news, hoaxes or rumors. Following our previous work in the frame of the Verification Multimedia Use task of Mediaeval 2016, we have explored the use of multimodal clues to detect fake news in social networks [38]. This year, we have studied the interest of combining and merging many approaches developed by the MediaEval participants in order to evaluate the predictive power of each modality. We have proposed several fusion strategies making the most of their potential complementarity [39].