Section: Overall Objectives
AxIS is carrying out research in the area of Information and Knowledge Systems (ISs) with a special interest in evolving large ISs such as Web-based Information Systems. Our core goal is to provide knowledge, methods and tools to support better design, evaluation and usage in the digital world, i.e. to improve the overall quality of ISs, to ensure ease of use to end users and also to contribute to user-driven open innovation as a way to foster innovation,
Our researches are organized to support the disruptive process of continuous innovation in which design is never ended and relies on very short test-adapt-test cycles. According to the constant evolution of actual and future ISs, to reach this goal, it is necessary to involve the users in the design process and to empower them, so that they can become codesigners as co-creators of value. This is a new way to anticipate the usage and its analysis and also to consider maintenance very early in the design process.
To achieve such a research, we have set up in July 2003 a multidisciplinary team that involves people from different computer sciences domains (Data Mining & Analysis, Software Engineering) and from cognitive sciences domains (Ergonomics, Artificial Intelligence), all of them focusing on information systems. Our goal is of course to improve efficiency of data mining methods but also to improve the quality of results for knowledge discovery in information systems. The originality of AxIS project-team is to adopt a cognitive and inter-disciplinary approach for the whole KDD (KDD: Knowledge Discovery From Databases) process and for each step (preprocessing, data mining, interpretation).
To address this challenge, relying on our scientific foundations (see our 2007 activity report, Section Scientific Foundations), we had a first 4 years step dedicated to the design of methodological and technical building blocks for IS mining (usage, content and structure) mainly in Web mining. The next four years were dedicated to provide original methods in mining data streams and evolutive data in the context of Web but also in sensor based applications. The last third period, we are putting our main efforts in the sub-objective 3 related to methods and tools for designing and evaluating user experience and supporting user-oriented innovation. In this period a great effort concerns the dissemination of our methods and tools inside the FocusLab experimental platform whose goal is to support the analysis of individual and collective user experience.
In this context, our team focused its effort on the technical and methodological environment needed to extract meaning from the huge amount of data issued from large and distributed information systems. Our ultimate goal is fed by research contributions from the three sub-objectives below:
Sub-objective 1 - Mining for Knowledge Discovery in Information Systems : Concerning Data Mining the specificity of our research is in two areas: methods and data. In traditional applications, a data mining process assumes that data to be mined is stored in a database with seldom (non frequent) updates. The extraction might take days, weeks, or even months, but due to the static nature of data, knowledge extraction can easily be deployed. When dealing with data streams, one only gets one look at data, which it changes over time. Due to the growing number of such emerging applications, the advanced analysis and mining of data streams is becoming more and more important, and it receives a great deal of attention. Mining data streams remains very challenging, because traditional data mining operations are impossible on data streams. Since data streams are continuous, high speed and unbounded, it is impossible to use traditional algorithms that require multiple scans.
In traditional Data Mining applications the representation of the data is a vector of Rp where p is the number of descriptors. In Web Mining the navigation must be represented by a ordered list of Rp vectors and it’s not easy to reduce this representation by one vector. At the start of AxIS the main challenge was to study different representations of the objects with the objective that the complex representation is closed to the initial representation. We proposed different non vectorial representations, called complex data. The main subject matters in sub-objective 1 are data stream mining, complex data clustering, semantic data mining.
Sub-objective 2 - Information and Social Networks Mining for supporting Information Retrieval : Related to information retrieval, we managed three main problems in the past: case-based recommender systems for supporting information search, expert finding whose goal is to identify persons with relevant experience from a given domain and entity extraction. Concerning social networks mining, our main subject matters are clustering methods for identifying communities inside social networks, expert finding and entity retrieval in Wikipedia. At the end of the nineties and in the early new millennium, many clustering methods have been adapted to the context of relational data sets (k-means approach and SOM by Hathaway, Davenport and Bezdek (1982, 2005), a divisive clustering by Girvan and Newman (2002), EM and Bayesian approaches by Handcock, M.S., Raftery, A.E. and Tantrum, J. (2007). The units are connected by a link structure representing specific relationships or statistical dependencies, the clustering task becomes a means for identifying communities within networks. Graphs are intuitive representations of networks.
With the last Web 2.0 technology developments of cloud computing, the improvement of web usability and web interactivity through rich interface, Ajax, RSS and semantic web, the concept of CAI (CAI: Computer Based Innovation) 2.0 is currently a major topic. In addition, HCI design and evaluation focus is no longer placed on usability but on the whole user experience. Experimentations take place out of labs with large number of heterogeneous people instead of carefully controlled panels of users. These deep changes require to adapt existing methodologies and to design new ones. So, to address these new requirements, we identified the following research :
Conceptual studies: state-of-the-art investigations covering the Living Lab landscape, the future internet domain landscape, the future user-open innovation for smart cities, user experience. These studies provide insight on methodological aspects for needs analyses, data gathering, evaluation, design, innovation methods.
FocusLab Experimental Platform (CPER Telius) (cf. section 6.6 dor the software part) is our delivery mechanism providing access to AxIS methodology and software for the scientific community.
For scientific foundations for some topics of sub-objectives 1 and 2, see our 2007 AxIS activity report.
All our research work (data and methods) is mainly applied in the context of Living Labs (cf. section 4.1 ).