Annonce en lien avec l’Action/le Réseau : aucun
Laboratoire/Entreprise : Laboratoire d’Informatique Fondamentale et Appliquée de Tours
Durée : 5 à 6 mois
Contact : Patrick.Marcel@univ-tours.fr
Date limite de publication : 2019-04-01
Contexte :
Can data analysis be fully automated and eventually an Artificial Intelligence (AI) makes the decision? The debate around AI, especially Machine Learning (ML), and their supposed capacity at automating decision making, is very in- tense these days. In the database (DB) community, and more particularly in the data warehousing (DW) community, there is long tradition of having the decision maker at the center of the data analysis process. At the inverse of auto- mated application of algorithms [4], DW has been, since its inception, all about facilitating the task of interactive exploration of a dataspace, and not let e.g., an algorithm automatically mine this space for patterns. One could even say that DW is the ancestor of the Human-In-the-Loop Data Analysis phenomenon [2].
This internship topic follows up from the reinvention of OLAP described in [9, 10], and ambitions to automatize further interactive data analysis, while letting the end user in command. This reinvention of OLAP introduces an analytics model redefining what a query is, with respect to both what users ask the system, what the answer entails, and how this answer is computed. An implementation is currently being done.
The work introduced in [10] opens several major research questions. A first question is How to facilitate the understanding of data? This demands to precisely define what are the answers to complex sequences of high level intentions, and package them into coherent data stories accessible to even non expert users.
As an answer to this question, authors propose that answers to intentional operators are no longer traditional sets of tuples, but dashboards including data, charts, informative summaries of KPI performance, as well as concise representations of knowledge hidden in the data.
The long term ambition is to automatically generate such dashboards based on past and current user interactions, and using data mining techniques.
Sujet :
The main challenge of this internship is to define how to structure dashboards in a context where the interactive data analysis is a sequence of possibly complex queries, each being a composition of intentions, in a personalized way [8].
The detailed objectives are:
1. Study of the intentional operators proposed in [10].
2. Literature review about dashboard representation.
3. Propose a dashboard model adapted to complex intentional queries.
4. As a proof of concept, generate dashboards for a set of user explorations.
Profil du candidat :
Applicants are expected to be 2nd year master students in Computer Science.
Formation et compétences requises :
Applicants are expected to be skilled in databases, machine learning, programming and be fluent in English. A first experience in research is a plus.
Adresse d’emploi :
The recruited student will be supervised by Veronika Peralta and Patrick Marcel at the University of Tours, in the campus of Blois (3 place Jean-Jaurès, 41000 Blois).
Document attaché : Intership_data_stories.pdf