Stephan Brunessaux

A Platform for Semantic Annotations and Ontology Population Using Conditional Random Fields

Ontologies are widely used for organising and sharing knowledge. But elaborating these resources is a heavy and time-consuming task. This paper is two-fold: it describes EADS DCS text-mining platform, in particular, its service to annotate documents with semantic tags and it presents its extension for incremental learning of ontologies. Domain experts are assisted in the ontology population task by recent machine learning techniques (i.e. conditional random fields). Comparisons are made between annotations from the ontology and from a trained CRF model, so as to detect candidate instances. An iterative process controlled by the experts results in knowledge discovery and constitution of an accurate ontology.

Exploring multiple feature combination strategies with a recurrent neural network architecture for off-line handwriting recognition

The BLSTM-CTC is a novel recurrent neural network architecture that has outperformed previous state of the art algorithms in tasks such as speech recognition or handwriting recognition. It has the ability to process long term dependencies in temporal signals in order to label unsegmented data. This paper describes different ways of combining features using a BLSTM-CTC architecture. Not only do we explore the low level combination (feature space combination) but we also explore high level combination (decoding combination) and mid-level (internal system representation combination). The results are compared on the RIMES word database. Our results show that the low level combination works best, thanks to the powerful data modeling of the LSTM neurons.

Evaluation with the VIRTUOSO platform: an open source platform for information extraction and retrieval evaluation

This paper describes a software architecture designed to enable the evaluation of information processing and retrieval systems. The overall objective of our project is to provide an open technical framework for the integration of tools for collection, processing, analysis and communication of open source information1. However, enabling the integration of heterogeneous components does not make sense without a proper way to compare the capabilities of multiple tools. Thus, as part of the architecture the VIRTUOSO platform offers an evaluation framework which allows one to deploy and run evaluation kits for different use-cases.

A step toward an adaptive composition of query suggestion approaches

The problems of comparing search support tool in interactive information retrieval (IIR) and of selecting the right one have always been difficult due to the inherent dependency to users. Using an adapted evaluation protocol, we study in this paper different suggestion approaches. The results show that the performance are changing for different users and also during the search sessions. As a consequence, they also show that the selection of a particular support tool has to use new grounding. In this way, we propose a system that allows to combine independent suggestion mechanisms based on an analysis of user behavior and considering the search session time as a key factor instead of using only static rules.

Events Extraction and Aggregation for Open Source Intelligence: From Text to Knowledge

Due to the considerable increase of freely available data, the discovery of relevant information from textual content is a critical challenge. The work presented here takes part in ongoing researches to develop a global knowledge gathering system. It aims at building knowledge sheets summarizing all the pieces of information we know about events extracted from text. For this sake, we define a global process bringing together different methods and components from multiple domains of research.

A Semantic Extension for Event Modelisation

Social networks and fast blogging services are interesting information sources. Anyone can contribute and submit information that matters to him/her. Among these many digital information sources, some appeared on the Internet as public means to be aware and to follow disasters and their evolutions. However, automatic raw information processing is more or less complex, error producer, relevant and reliable. This paper introduces a formal event modelisation extended with semantic properties to represent these information. We explain how to combine this extended model with reasoning methods on structured knowledge in order to discover unobvious relations hidden in information. Finally, we present a prototype based on this model and we discuss results returned by our experiments.

Measures for topical cohesion of user communities on Twitter

Nowadays, Online Social Networks (OSN) are commonly used by groups of users to communicate. Members of a family, colleagues, fans of a brand, political groups: the demand for a precise identification of these groups is increasing from brand monitoring, business intelligence and e-reputation management. However, a gap can be observed between the communities detected by many data analytics algorithms on OSN, and effective groups existing in real life: the detected communities often lack of meaning and internal semantic cohesion. Most of existing literature on OSN either focuses on the community detection problem in graphs without considering the topic of the messages exchanged, or concentrates exclusively on the messages without taking into account the social links. In this article, we support the hypothesis that communities extracted on OSN should be topically coherent. We therefore propose a model to represent the interaction between users on Twitter, the reference on micro-blogging OSN, and metrics to evaluate the topical cohesion of the detected communities. As an evaluation, we measure the topical cohesion of the groups of users detected by a baseline community detection algorithm, using two measures inspired from the classification domain, and one measure inspired from the NLP domain. A detailed analysis is performed on a big tweet dataset, from which a user graph is built. Introduced measures are compared with statistics to better picture the experiment, and yield interesting insights on a social and textual corpus.

Topical cohesion of communities on Twitter

Abstract Nowadays, Online Social Networks (OSN) are commonly used by groups of users to communicate. Members of a family, colleagues, fans of a brand, political groups... There is an increasing demand for a precise identification of these groups, coming from brand monitoring, business intelligence and e-reputation management. However, a gap can be observed between the communities detected by many data analytics algorithms on OSN, and effective groups existing in real life: the detected communities often lack of meaning and internal semantic cohesion. Most of existing literature on OSN either focuses on the community detection problem in graphs without considering the topic of the messages exchanged, or concentrates exclusively on the messages without taking into account the social links. In this article, we support the hypothesis that communities extracted on OSN should be topically coherent. We therefore propose a model to represent the groups of interaction on Twitter, the reference on micro-blogging OSN, and two metrics to evaluate the topical cohesion of the detected communities. As an evaluation, we measure the topical cohesion of the groups of users detected by a baseline community detection algorithm.

Relevant Sources of Information Are Not Necessarily Popular Ones

The constant growth of the Web in recent years has made more difficult the discovery of new sources of information on a given topic. This is a prominent problem for Experts in Intelligence Analysis (EIA) who are faced to the search of pages on specific and sensitive topics. Because of their lack of popularity or because they are poorly indexed due to their sensitive content, these pages are hard-to-find with traditional search engines. In this article, we describe a new Web source discovery system called DOWSER (Discovery Of Web Sources Evaluating Relevance). The goal of this system is to provide users with new sources of information related to their needs without considering the popularity of a page unlike classic Information Retrieval tools. The expected result is a balance between relevance and originality, in the sense that the wanted pages are not necessary popular. DOWSER is based on a user profile to focus its exploration of the Web in order to collect and index only related Web documents. As requests can be insufficient to express sensitive and specific needs, the users information needs are specified using users interests represented by DBPedia resources [1] and keywords, both extracted from Web pages provided by the user. A series of experiments provides an empirical evaluation of DOWSER.

Extracting Contextonyms from Twitter for Stance Detection.

Opinion mining on tweets is a challenge: short texts, implicit topics, inventive spellings and new vocabulary are the rule. We aim at efficiently determining the stance of tweets towards a given target. We propose a method using the concept of contextonyms and contextosets in order to disambiguate implicit content and improve a given stance classifier. Contextonymy is extracted from a word co-occurrence graph, and allows to grasp the sense of a word according to its surrounding words. We evaluate our method on a freely available annotated tweet corpus, used to benchmark stance detection on tweets during SemEval2016.