Taoufiq Dkaki

TetraFusion: information discovery on the Internet

The TetraFusion system described in this paper supports knowledge discovery from the World Wide Web by helping users perform data mining operations on sets of harvested URLs. Potential applications range from domain overviewing to science monitoring to competitive intelligence.

Combining mining and visualization tools to discover the geographic structure of a domain

Abstract Science monitoring is a core issue in the new world of business and research. Companies and institutes need to monitor the activities of their competitors, get information on the market, changing technologies or government policies. This paper presents the Tetralogie platform that is aimed at allowing a user to interactively discover trends in scientific research and communities from large textual collections that include information about geographical location. Tetralogie consists of several agents that communicate with each other on users’ demands in order to deliver results to them. Metadata and document content are extracted before being mined. Results are displayed in the form of histograms, networks and geographical maps; these complementary types of presentations increase the possibilities of analysis compared to the use of these tools separately. We illustrate the overall process through a case study of scientific literature analysis and show how the different agents can be combined to discover the structure of a domain. The system correctly predicts the country contribution to a field in future years and allows exploration of the relationships between countries.

Interactive multidimensional document visualization

Complementary approaches are used to improve information retrieval efficiency. One approach consists in focusing on the system inner processes used (e.g. document indexing, query-document similarity computing). Another approach can be to improve the user-system interaction [7]. Positive and negative relevance feedback type mechanism is among the most used and lots of studies have shown its efficiency. Classification is also used: grouping together documents according to their similarity and using those classes when consulting or querying the collection. When proposing an interactive process, powerful interfaces are needed. INQUERY [l] proposes a graphical view of the retrieved documents where similar documents are graphically close; the retrieved documents are listed to the user according to the class they belong to. VIBE [G] represents the information in a two dimensional space (document vs term) so that it is possible to graphically see which are the possible relevant documents according to some selected terms. Some complementary studies have shown that it is efficient to use the document structure or the factual information extracted from a document. ENVISION [5] represents the retrieved documents into two-dimensional tables. Rows and columns represent either factual information such as authors, date, or estimated relevance and cells are filled with document references, navigation through those tables is possible. In this poster we propose an interactive document visualization tool. It takes into account the fact that document relevance depends on the document features which are considered (information content, publication date, author affiliation, . ..). Those elements are crossed in order to discover the strong links that exist between them. Discovered relationships are then displayed on a 4-dimensional graphical view. This interface allows the user to graphically select the elements he is most interested in and to automatically construct new filters that will change interactively the set of the displayed documents.

Passage Retrieval Using Graph Vertices Comparison

In this paper, we describe an Information retrieval Model based on graph comparison. It is inspired from previous work such as KleinbergpsilaHits and Blondel et al.psilas model. Unlike previous methods, our model considers different types of nodes: text nodes (elements to retrieve and query) and term nodes, so that the resulting graph is a bipartite graph. The results on passage retrieval task show that high precision is improved using this model.

Information discovery from semi-structured sources - Application to astronomical literature

Textual information systems provide different kinds of information seeking that answer different user needs. Among them, knowledge discovery systems aim at providing global views and useful patterns from raw information. This paper presents a framework to discover knowledge from semi-structured documents and visualize it through graphical views. An application to astronomical literature is given.

Recherche et représentation de communautés dans un grand graphe: Une approche combinée

Ce travail concerne l’analyse, la comprehension et la representation de grands graphes. La progression des moyens de recueil et de stockage des donnees rend la taille de ces graphes croissante : le developpement de methodes permettant leur analyse et leur representation est donc un domaine de recherche dynamique et important. Dans cet article, nous developpons une methode de representation de graphes basee sur une classification prealable des sommets avant sa representation complete. La phase de classification consiste en l’optimisation d’une mesure de qualite specialement adaptee a la recherche de groupes denses dans les graphes. La representation finale est basee sur un algorithme de « forces » contraint. Deux exemples issus de l’analyse de reseaux sociaux sont presentes.

Enhancing High Precision by Combining Okapi BM25 with Structural Similarity in an Information Retrieval System.

In this paper, we present a new similarity measure in the context of Information Retrieval (IR). The main objective of IR systems is to select relevant documents, related to a user‟s information need, from a collection of documents. Traditional approaches for document/query comparison use surface similarity, i.e. the comparison engine uses surface attributes (indexing terms). We propose a new method which combines the use of both surface and structural similarities with the aim of enhancing precision of top retrieved documents. In a previous work, we showed that the use of structural similarity in combination with cosine improves bare cosine ranking. In this paper, we compare our method to Okapi based on BM25 on the Cranfield collection. We show that structural similarities improve average precision and precision at top 10 retrieved documents about 50%. Experiments also address the term weighting influences on system performances.

Knowledge discovery in bibliographic collections using concept hierarchies and visualization tools.Application to the astronomy domain

This paper presents new methods for knowledge extraction and visualization, applied to datasets selected from the astronomical literature. One of the objectives is to detect correlations between concepts extracted from the documents. Concepts are generally meta-information which may be defined a priori, or may be extracted from the document contents and are organised along domain ontologies or concept hierarchies. The study illustrated in the paper uses a data collection of about 10,000 articles extracted from the NASA ADS, corresponding to all publications for which at least one author is a French astronomer, for the years 1996 to 2000. The study presents new approaches for visualizing relationships between institutes, co-authorships, scientific domains, astronomical object types, etc.

Science watch by knowledge discovery in bibliographic databases

The purpose of Science Watch (SW) is to help searchers to remain aware of what is happening in their respective research-fields without drowning in the huge number of scientific publications. It mainly answers strategic questions such as: Who works with whom? (relationship between searchers) What journals may interest a research worker? What person, laboratory or university depar tment are more likely to collaborate with another person, laboratory or umversity depar tment? What is new? (new concepts, vanishing points of interest etc.) etc. To meet this goal SW includes raw-information selection, pre-selected-data analyses, and results interpretation. The results highlight relationships among the collected da ta leading to the responses to the questions above. Formal relationships, the only we can hope to discover without carrying out a poll, can be discovered from bibliographic da ta bases. In this case SW and Knowledge Discovery in Databases (KDD) are alike: they have the same information origin (databases) and the same goals (knowledge discovery). Technics and methods of KDD can be used to fulfil the goals of SW and vice versa. Actually SW becomes a par t of KDD when it only uses formal information coming from data