|
|
|
| e-Pub |
Previous | 
Home
Section: New Results
Natural Language Processing
Bridging Natural Language Processing and the Web of Data for Multimedia Question Answering
Participants : Elena Cabrio, Fabien Gandon, Yoann Moise.
Differently from search engines, the goal of Question Answering (QA) is to return precise answers to users' natural language questions, extracting information from both documentary text and advanced media content. Up to now, QA research has largely focused on text, mainly targeting factual and list questions. The goal of our work was instead to exploit structured data and metadata describing multimedia content on Linked Open Data to provide a richer and more complete answer to the user, combining textual information with other media content.
We implemented an extension of our QAKiS system (http://qakis.org ) to boost the answer visualization adding multimedia content. More specifically, once QAKiS outputs the textual answer(s) to the question asked by the user, the user can click on ”more details” to have further information on the retrieved entity. Three main types of additional (and multimedia) content are then displayed: i) additional textual information providing a description of the retrieved entity (extracted from DBpedia), and a structured Information Card containing a set of relevant properties of the entity; ii) images (extracted from Flickr) and relevant videos (extracted from YouTube); iii) entity geo-localization: a pointer on a map is shown for questions asking about a place, together with its points of interest. The results of this research have been published at ESWC 2014 - Demo/poster paper [84] .
SMILK - Social Media Intelligence and Linked Knowledge
Participants : Elena Cabrio, Fabien Gandon, Fabrice Jauvat.
Automated Natural Language Processing (NLP), Web Open Data (Linked Open Data) and social networks are the three topics of the SMILK ANR LabCom including their coupling studied in three ways: texts and Linked Data, Linked Data and social resources, texts and social resources. The purpose of this LabCom is indeed to develop research and technologies on the one hand, retrieve, analyze, and reason about linking data from textual Web resources and other to use open Web data taking into account the social structures and interactions in order to improve the analysis and understanding of textual resources.
As a first step in this direction, during the internship of Fabrice Jauvat we have developed a prototype of a system that - given free text (in particular in the cosmetics domain, extracted from a forum, a magazine, or a Web site) - can first recognize the named entities launching in parallel the RENCO system (developed by our partner in the LabCom), and NERD (http://nerd.eurecom.fr/ ), and then connect them to DBpedia, so that additional information on the entity can be extracted. For instance, if the word ”J'adore” is detected in the text, it is recognized as a named entity and connected to its DBpedia page, so that information about the fact that it is a perfume, that its brand is Dior, and so on, can be automatically extracted and shown to the user.
Ontology-Based and Natural Language Chatbot System in the Commercial Domain
Participants : Amine Hallili, Catherine Faron Zucker, Elena Cabrio, Fabien Gandon.
This work is done within a Cifre PhD Thesis colocated in the Wimmics team and with SynchroNext Company located in Nice. The work consists in modelling and implementing ontology-based natural language Chatbot in commercial domain which consists of
-
The design of a commercial knowledge base using the websites' APIs and web services (e.g. Amazon1, eBay2, BestBuy3),
-
Interpreting and handling links between users' natural language questions by constructing relational graph,
A. Hallili attended the ESSLLI summer school where a poster was accepted [93] .
Editor of Formal Lexicographic Definitions
Participants : Maxime Lefrançois, Romain Gugert, Alain Giboin, Fabien Gandon.
Last year a prototype of a GUI of an editor of formal dictionary definitions aimed at lexicographers was developed based on the formalism of Units Graphs and on Meaning-Text Theory. This year, the prototype was demonstrated during the IC 2014 conference [60] . The prototype was also described in a paper reporting the knowledge engineering methodology for representing lexicographic definitions it supports [39] .