EN FR
EN FR


Section: New Results

Linking, navigation and analytics

Opinion similarity and target extraction

Participants : Vincent Claveau, Grégoire Jadi.

Work in collaboration with Laura Monceaux and Béatrice Daille, LINA, Nantes.

In [19], we propose to evaluate the lexical similarity information provided by word representations against several opinion resources using traditional information retrieval tools. Word representation have been used to build and to extend opinion resources, such as lexicon and ontology, and their performance have been evaluated on sentiment analysis tasks. We question this method by measuring the correlation between the sentiment proximity provided by opinion resources and the semantic similarity provided by word representations using different correlation coefficients. We also compare the neighbors found in word representations and list of similar opinion words. Our results show that the proximity of words in state-of-the-art word representations is not very effective to build sentiment similarity.

In [20], we present the development of an opinion target extraction system in English and transpose it to the French language. In addition, we realize an analysis of the features and their effectiveness in English and French which suggest that it is possible to build an opinion target extraction system independant of the domain. Finally, we propose a comparative study of the errors of our systems in both English and French and propose several solutions to these problems.

Reinformation and fake detection in social networks

Participants : Vincent Claveau, Ewa Kijak, Cédric Maigrot.

Traditional media are increasingly present on social networks, but these usual sources of information are confronted with other sources called reinformation sources. These last ones sometimes tend to distort the information relayed to match their ideologies, rendering it partially or totally false. In [25], we conduct a study pursuing two goals: first, we present a corpus containing Facebook messages issued from both types of media sources; secondly, we propose some experiments in order to automatically detect reinformation messages. In particular, we investigate the influence of shallow features versus features more specifically describing the message content. We also developed a multi-modal hoax detection system composed of text, source, and image analysis [24]. As hoax can be very diverse, we want to analyze several modalities to better detect them. This system is applied in the context of the Verifying Multimedia Use task of MediaEval 2016. Experiments show the performance of each separated modality as well as their combination.

Multimodal video hyperlinking

Participants : Rémi Bois, Guillaume Gravier, Christian Raymond, Pascale Sébillot, Ronan Sicre, Vedran Vukotić.

Pursuing previous work on video hyperlinking and recent advances in multimodal content matching [32], we benchmarked a full video hyperlinking system in the framework of the TRECVid international benchmark [12]. The video hyperlinking task aims at proposing a set of video segments, called targets, to complement a query video segment defined as anchor. The 2016 edition of the task encouraged participants to use multiple modalities. In this context, we chose to submit four runs in order to assess the pros and cons of using two modalities instead of a single one and how crossmodality differs from multimodality in terms of relevance. The crossmodal run performed best and obtained the best precision at rank 5 among participants. In parallel, we also demonstrated that, in this framework, multimodal and crossmodal approaches offer significantly more diversity in the set of target proposed than classical information retrieval based approaches where all modalities are combined. We compared bidirectionnal multimodal embeddings [31] with multimodal LDA approaches as experimented last year in TRECVid  [49]. The former offers more accurate matching, the latter exhibiting slighlty more diveristy.

User-centric evaluation of hyperlinked news content

Participants : Rémi Bois, Guillaume Gravier, Pascale Sébillot, Arnaud Touboulic.

Work in collaboration with Éric Jamet, Martin Ragot and Maxime Robert, CRPCC, Rennes.

Following our study of professional user needs in multimedia news analytics [15], we developed a prototype news analytics interface that facilitates the exploration of collections of multimedia documents by journalists. The application, based on standard web technology, enriches classical functionalities for this type of applications (e.g., keyword highlights, named entity detection, keyword search, etc.) with navigation-based functionalities. The latter exploit a graph-based organization of the collection, established from content-based similarity graphs on which community detection is performed along with basic link characterization. We performed usage tests on students in journalism and on journalists where each user was asked to write a synthesis article on a given topic. Preliminary results indicate that the graph-based navigation improves the completeness of the synthesis by exposing users to more content than with a standard search engine.