Abstract: The paper describes a method of extracting and exploring web aggregates based on an experimental user centered tool called TARENTe. The method consists of three phases. The first one is the extraction of the basic structure of aggregates made of a core and a border using a topical content and a link topology analysis. The general structure is next ripen and stabilized with visual and statistical tools to describe more precisely the aggregate. The last phase is the aggregate exploration and it is designed to detect local patterns using specific contextualized visualizations. Finally, end-user navigation interfaces reveal the discovered organization of aggregates by visually manifest it .
Abstract: In this paper, we propose to use tools of exploratory spatial analysis to study the mobility of foreign visitors in a city like Paris. Using digital traces from mobile phones, we explore several aspects of mobility: how people move from one place to another, but also how these places are linked together. The exploration depends on a methodology which uses visualiza- tions to reveal patterns in large heterogeneous data-sets.. This allowed us to gain new insights on how to analyze noisy and imprecise spatio-temporal data, and to formulate a reliable hypothetical model of tourist urban mobility.
Abstract: Geographical information systems (GIS) associate innovative statistical and visual tools and play a large role in the exploration of geographical data spaces. In the numeric world, web is a new space that remains to be explored. Heterogeneous and highly dynamic with very large dataset, its analysis requires adapted tool as much as a specific methodology. There- fore, we have developed a web information system (WIS) based on GIS in terms of knowledge extraction and visual exploration of datasets. We conduct both an experimental study and a theoretical research to draw perceptive and cognitive principles that take place during the ex- ploration based on visual tools. Thus we were able to improve and better choose these tools. We propose in this paper a study of the french picard web territory that uses methods to joint web data and geographical data in the experimental WIS and theoretical principles we discovered.
Abstract: This paper presents two different perspectives of the web: a global one that corresponds to the classical approach of search engines and a the local one that we propose as an alternative approach. The search engines perform their indexation operation on the whole web in an automatic way and display their results according to it by proposing a perfectible visualization. We will review the usability of these visualizations while examining the way search engines build their hierarchies. That leads us to reconsider the notion of context and the way models of the web influence our vision of it to finally propose a new model strongly related to its perception through alternative visualizations.
Abstract: How to extract and visually explore the topology of an open, large scale, hypertext system such as the web ? We address this issue by developing an experimental tool for extracting, exploring and analyzing Aggregates of web documents. This tool, called TARENTe, includes a crawling technology, and algorithms for both content analysis and authority graphs calculations (as Kleinbergâs HITS), linked with visualization solutions. We provide series of experimental results on different topics that allow us to describe the webâs structure in terms of topic Aggregates.
Abstract: In reply to our crucial need for making sense of the world outside, one can now rely on a huge quantity of numerical data, ready to be stressed in every possible way by a new kind of explorer only limited by imagination. Tools and skills are no more jungle-ready but made of clever and innovative design and use of infographics and visualizations that are able to reveal the hidden [1]. New fields like EDA and visual analytics encourage the use of an expert and highly visual process to formulate and validate hypothesis on a dataset [2][3].
Once hypothesis are discovered and validated throught visualizations, they become knowledges of the system and can be used for further investigation. The key role of diagrams as part of a process of knowledge discovery increases the need for diagrams to be well conceived and well used so as not to engage on a discovery path leading at best to a dead-end, at worst to a false conclusion. To design relevant and efficient diagrams, one can count on several semiologic works like Bertin[4] or Tufte[5] but these semiologies are still widely based upon the designers experience and are difficult to apply to the extraordinary variety of recent visualizations. Furthermore, they focus on a presenting knows use of diagrams where a process of exploration requires a specific revealing unknows use.
To address these concerns, we propose an experimental enactive semiology that encompasses both perception and cognition to explain graphics reading. It provides a way of analyzing every kind of graphics from a perceptual point of view knowing how it might affect the user. This semiology provides a set of primary graphic structures (list, table, chart, graph and map) and properties of how human perceive and understand them. Then we propose a direct application of these properties and structures to enhance EDA or visual analytics by presenting a new process of data exploration that encompass current visual mining processes by promoting a systematic use of diagrams and visualizations given their perceptual properties. The process is based upon a loop of hypothesis creation and validation that gradually become more and more accurates to form a model of the dataset. The process itself can be managed through a visualization software following the semiology guidelines. However not automatic, It gives a human explorator an insightfull way of mining large dynamic and heterogeneous datasets without getting lost, building a good-fitting model of it, while maximizing his efficiency and minimizing his errors. It also allows him to communicate the discovered knowledges given the specificity and expectations of the target audience. Examples are from various web exploration and experimental work.