Experience and numerical traces of interactions
- knowledge acquisition, modelisation and engineering,
- trace engineering,
- interactive knowledge discovery from traces,
- interactive knowledge acquisition from traces,
- Case-based reasoning.
My research works aim at studying user's experience by observing its interactions
with a numerical environment and the exploration of its traces.
Traces testify a past activity and constitute a container that may be analysed with different objectives :
studying complex phenomena such as, in particular, human learning or to assist user's task.
Exploration of numerical traces of activity
The aim is to set up a trace analysis laboratory in order to answer to various questions related to human learning (but not limited to) : users modeling, characterization of failures, etc.
One way to explore traces is knowledge discovery and more particularly the extraction of serial episodes.
The work rely on a system based on modeled traces (see the web page of kTBS - Kernel for trace based system) implemented in the Tweak team. A trace based system is a knowledge base resulting from the interpretation of traces.
We propose to assist the collection and exploitation of digital traces, as well as their transformation in order to memorize their interpretation and to acquire new knowledge in interaction with the user.
DisKit implements a knowledge discovery process that extracts
a set of frequent serial episodes from one or more traces.
DisKit encapsulates the DMT4SP algorithm that extracts sequential episodes and sequential rules from a sequence of events, and supports pre- and post-processing steps with additional constraints.
Transmute is a web based application designed to provide a user with tools in order to assist the interpretation of the traces.
Initially designed to visualize and transform traces interactively, it has been enriched with knowledge discovery features: parameterization and launch of a the miner and display results, assistance to interpretation, visual and interactive manipulation of the results and storing the resulting model.
It relies on DisKit that supports the knowledge discovery process.
Knowledge Acquisition from Traces with Interactive Exploration
Katie aims at generating a trace model from raw data interactively.
The model is built inductively from the traces using formal concepts analysis (FCA).
However, There may be disagreements between a data schema produced by FCA and what the analyst knows about the domain, due to noisy data or anomalies within data.
The disagreement analysis confronts the implications between concepts with those of the domain model in an iterative and interactive way.
Differential adaptation in case-based reasoning
In collaboration with Alain Mille (LIRIS/Tweak), Jean Lieber and Amedeo Napoli (LORIA/Orpailleur).
Differential adaptation is a generic approach to adaptation in case-based reasoning (CBR), where the solution of a target case is build by taking into account differences between source and target problems and dependencies between the source problem and its solution. Adaptation operators are inspired by differential calculus and manipulate variations between problems to generate variations of the solution, and thus calculate the solution of the target problem. KBS reference
Interactive knowledge discovery from traces
Damien Cram's thesis, in collaboration with Alain Mille (LIRIS/Silex). The thesis focuses on the interactive discovery of knowledge in interaction traces in the field of collaborative knowledge management.
Interactive and Opportunistic Knowledge Acquisition in Case-Based Reasoning.
Amélie Cordier's thesis, in collaboration with Alain Mille (LIRIS/Silex) and Jean Lieber (LORIA/Orpailleur)
The thesis addresses the problem of opportunistic knowledge acquisition in case-based reasoning in the revise and learn steps.
(Human Observatory Based on analysis of e-learning traces, ANR).
Open Food System, programme investissements d’avenir.
Coprat (2009, Région Rhône Alpes)
COmmunautés de PRatique Assistées par les Traces.
Procogec (2007-2010, ANR)
PROgiciel COllaboratif de GEstion des Connaissances.
Dimocode (2007-2009, CNRS)
Diffusion par Internet des MOdèles pour la Conception Optimale des Dispositifs Energétiques.