|
User Centric Data Exploration| old_uid | 17895 |
|---|
| title | User Centric Data Exploration |
|---|
| start_date | 2019/10/17 |
|---|
| schedule | 11h |
|---|
| online | no |
|---|
| summary | Traditional data exploration is an iterative process that generally starts with fuzzy initial user needs, and a prescriptive querying language such as SQL to extract meaningful information. Since their inception, most DBMSs have notoriously been optimized to ensure that massive queries could be answered very fast (see TPC benchmarks for example), ignoring new analytical needs, like for instance data enthusiasts' interactions to produce dashboards or insights from the data.
Our research group investigates user-centric approaches for data exploration, from the automatic elicitation of user intents, the evaluation of the exploration results based on what was interesting for the analyst and what she learned from interactions with the data, the proposal of new interactive clustering approaches, to the definition of a first user centered benchmark for exploration quality assessment.
This talk describes some of the approaches investigated, and is organized in three main parts. First, we present two approaches for modeling user profiles: (i) user intent elicitation and its application to bundling exploration queries for recommendation purpose, and (ii) subjective interestingness modeling and its application to estimate to which extent an interaction could be valuable for a user.
Next, we present a method to automatically assess the quality of explorations of multidimensional data.
Finally, our talk concludes with a description of an on-going work on the implementation of intentional exploration primitives. |
|---|
| responsibles | Piwowarski |
|---|
| |
|