Research team : LORIA - Knowledge Information and Web Intelligence

Name :

Acronym :

Additional address :

Address :

Postal code :

City :

Website :

Research team manager :

Number of members :

Description :

General objective The objective of the researches conducted in the KIWI team is the increase of the quality of online services (intranets, digital databases, information portals, etc.) brought to an identified or non-identified active user. With this aim, we study the interaction between users and e-services or e-resources for a large audience and risk-free services. Scientific objective We aim at designing decision support methodologies. The characteristics of the problem is the dynamicity and uncertainty of the environment, the sparsity of data that are heterogeneous, uncertain and noisy.
Scientific context More specifically, the goal is to improve the quality of the interactions between the general audience and systems of research and access to information. There are several possible approaches to assist the active user by either creating adaptive interfaces that facilitate the exploration and the searches on the Web (these systems rely on social navigation), or by developing sites providing personalized content, or by developing statistical tools that suggest keywords for improving searches,
etc. Our approach consists in providing each user with items likely to interest him/her. Contrary to the personalized content, this solution does not require to adapt resources to the potential readers.
One way to perform service personalization is to use recommender systems: systems that recommend to users the adequate resources that fit their objectives, attempts, profiles, etc. These systems can also be viewed as information retrieval (IR) systems as they search the adequate resources in the whole set of resources. In the Kiwi team, our research activity is dedicated to recommender systems and information retrieval systems.
Feasibility and hypothesis The data that reflects the interaction between users and services is usage data, such as clickstream, logs, gaze traces, etc. Some complementary data called observation data such as GPS or time information, for example, can also be available. We have to mention that we face a paradox that influences the way we solve the problem of e-personalization: the amount of information available about each user is sparse, whereas the total amount of information available is huge as the number of users or e-services or e-resources is enormous.
Our research relies on the following hypotheses: stability of preferences and behaviors within a small period of time. A user is never unique because of behavior either by himself or by another user. The interaction between a user and a e-service is influenced by the context of the interaction, eventually by the culture of the user. We assume that each factor influencing the interaction is directly or indirectly observable in the traces of usage or in the observation data.
Our approach As it is irrelevant to model users individually due to the lack of available individual data within a reasonable amount of time, we decided to collectively model users. These models not only reflect the shared features of a community of similar users while characterising their differences.
It is the reason why we focus on social learning, sociology and ergonomy of interaction and collective intelligence.The technics used The techniques used to perform this goal are mainly statistics, graph theory and natural language processing.

E-education positioning :

Name of the laboratory :

URL publication :

Scientific field :

Affiliation :

Teaching level(s) :

Projet(s) :