Adaptive predictive-questionnaire by approximate dynamic programming

Frédéric Logé; Erwan Le Pennec; Habiboulaye Amadou-Boubacar

doi:10.18420/muc2020-ws111-264

Adaptive predictive-questionnaire by approximate dynamic programming

Frédéric Logé
, Erwan Le Pennec
, Habiboulaye Amadou-Boubacar

Air Liquide
Ecole polytechnique
INRIA

Research output: Contribution to journal › Conference article › peer-review

Abstract

As too much interaction can be detrimental to user experience, we investigate the computation of a smart questionnaire for a prediction task. Given time and budget constraints (maximum q questions asked), this questionnaire will select adaptively the question sequence based on answers already given. Several use-cases with increased user and customer experience are given. The problem is framed as a Markov Decision Process and solved numerically with approximate dynamic programming, exploiting the hierarchical and episodic structure of the problem. The approach, evaluated on toy models and classic supervised learning datasets, outperforms two baselines: a decision tree with budget constraint and a model with q best features systematically asked.

Original language	English
Journal	Lecture Notes in Informatics (LNI), Proceedings - Series of the Gesellschaft fur Informatik (GI)
DOIs	https://doi.org/10.18420/muc2020-ws111-264
Publication status	Published - 1 Jan 2020
Event	Mensch und Computer 2020, MuC 2020 - Workshop on 7. Mensch-Maschine-Interaktion in sicherheitskritischen Systemen - Human and Computer 2020, MuC 2020 - Workshop on the 7th Human-Machine Interaction in Safety-Critical Systems - Magdeburg, Germany Duration: 6 Sept 2020 → 9 Sept 2020

Keywords

Approximate dynamic programming
Planning
Questionnaire design

Access to Document

10.18420/muc2020-ws111-264

Cite this

@article{132db9b3d5cd4cf18c14d8aca33d3bc5,

title = "Adaptive predictive-questionnaire by approximate dynamic programming",

abstract = "As too much interaction can be detrimental to user experience, we investigate the computation of a smart questionnaire for a prediction task. Given time and budget constraints (maximum q questions asked), this questionnaire will select adaptively the question sequence based on answers already given. Several use-cases with increased user and customer experience are given. The problem is framed as a Markov Decision Process and solved numerically with approximate dynamic programming, exploiting the hierarchical and episodic structure of the problem. The approach, evaluated on toy models and classic supervised learning datasets, outperforms two baselines: a decision tree with budget constraint and a model with q best features systematically asked.",

keywords = "Approximate dynamic programming, Planning, Questionnaire design",

author = "Fr{\'e}d{\'e}ric Log{\'e} and \{Le Pennec\}, Erwan and Habiboulaye Amadou-Boubacar",

note = "Publisher Copyright: {\textcopyright} Proceedings of the Mensch und Computer 2020 Workshop on «Workshop on User-Centered Artificial Intelligence (UCAI 2020)». Copyright held by the owner/author(s); Mensch und Computer 2020, MuC 2020 - Workshop on 7. Mensch-Maschine-Interaktion in sicherheitskritischen Systemen - Human and Computer 2020, MuC 2020 - Workshop on the 7th Human-Machine Interaction in Safety-Critical Systems ; Conference date: 06-09-2020 Through 09-09-2020",

year = "2020",

month = jan,

day = "1",

doi = "10.18420/muc2020-ws111-264",

language = "English",

journal = "Lecture Notes in Informatics (LNI), Proceedings - Series of the Gesellschaft fur Informatik (GI)",

issn = "1617-5468",

publisher = "Gesellschaft fur Informatik (GI)",

}

TY - JOUR

T1 - Adaptive predictive-questionnaire by approximate dynamic programming

AU - Logé, Frédéric

AU - Le Pennec, Erwan

AU - Amadou-Boubacar, Habiboulaye

N1 - Publisher Copyright: © Proceedings of the Mensch und Computer 2020 Workshop on «Workshop on User-Centered Artificial Intelligence (UCAI 2020)». Copyright held by the owner/author(s)

PY - 2020/1/1

Y1 - 2020/1/1

N2 - As too much interaction can be detrimental to user experience, we investigate the computation of a smart questionnaire for a prediction task. Given time and budget constraints (maximum q questions asked), this questionnaire will select adaptively the question sequence based on answers already given. Several use-cases with increased user and customer experience are given. The problem is framed as a Markov Decision Process and solved numerically with approximate dynamic programming, exploiting the hierarchical and episodic structure of the problem. The approach, evaluated on toy models and classic supervised learning datasets, outperforms two baselines: a decision tree with budget constraint and a model with q best features systematically asked.

AB - As too much interaction can be detrimental to user experience, we investigate the computation of a smart questionnaire for a prediction task. Given time and budget constraints (maximum q questions asked), this questionnaire will select adaptively the question sequence based on answers already given. Several use-cases with increased user and customer experience are given. The problem is framed as a Markov Decision Process and solved numerically with approximate dynamic programming, exploiting the hierarchical and episodic structure of the problem. The approach, evaluated on toy models and classic supervised learning datasets, outperforms two baselines: a decision tree with budget constraint and a model with q best features systematically asked.

KW - Approximate dynamic programming

KW - Planning

KW - Questionnaire design

U2 - 10.18420/muc2020-ws111-264

DO - 10.18420/muc2020-ws111-264

M3 - Conference article

AN - SCOPUS:85138799779

SN - 1617-5468

JO - Lecture Notes in Informatics (LNI), Proceedings - Series of the Gesellschaft fur Informatik (GI)

JF - Lecture Notes in Informatics (LNI), Proceedings - Series of the Gesellschaft fur Informatik (GI)

T2 - Mensch und Computer 2020, MuC 2020 - Workshop on 7. Mensch-Maschine-Interaktion in sicherheitskritischen Systemen - Human and Computer 2020, MuC 2020 - Workshop on the 7th Human-Machine Interaction in Safety-Critical Systems

Y2 - 6 September 2020 through 9 September 2020

ER -

Adaptive predictive-questionnaire by approximate dynamic programming

Abstract

Keywords

Access to Document

Fingerprint

Cite this