Skip to Main content Skip to Navigation
Theses

Collaboration humain-machine à l’aide de motifs dialogiques pour la réalisation d’une tâche complexe : application à la recherche d’information

Abstract : This thesis offers a conventional model of the structure of a human-machine collaborative task to design an interactive system helping a human performing complex tasks. More specifically, we focus on tasks whose resolution is highly opportunistic and can not be planned. We introduce a task representation model based on dialogue games, dialogue patterns making it possible to describe the interaction structure with sequences of conventionally acceptable dialogue acts. We organize these dialogue patterns in states, structures describing the expected behaviors from each interlocutor during the different sub-tasks of the collaborative task. These states group dialogue patterns together in a coherent set in regard to the sub-task they are associated to and enrich them with locally relevant rules. They make it possible to describe the effects of the execution of a dialogue pattern by the interlocutors in a given context of the task. The states are used by the system to link an utterance of the human to the current state of the dialogue and of the task and to take the initiative of behaviors relevant with the interaction and helpful for the task. The decisions taken by the system are based on the concept of maturity, a value associated to each state representing the system’s ability to take initiatives in this state. The decision model of the system is designed to be resilient and give as much freedom as possible to the user. This model is implemented in CoCoA, a system that collaborates with a human to assist him performing a complex task. From the study of a human-human dialogue corpus on a medical collaborative information retrieval task, we use our model to describe the task. This use case is implemented with CoCoA and an evaluation is performed by simulating a human user with an information need and varying its actions according to behaviors identified in the corpus.
Complete list of metadatas

Cited literature [304 references]  Display  Hide  Download

https://tel.archives-ouvertes.fr/tel-02923983
Contributor : Abes Star :  Contact
Submitted on : Thursday, August 27, 2020 - 4:00:09 PM
Last modification on : Friday, October 23, 2020 - 5:02:39 PM

File

Louvet.pdf
Version validated by the jury (STAR)

Identifiers

  • HAL Id : tel-02923983, version 1

Citation

Jean-Baptiste Louvet. Collaboration humain-machine à l’aide de motifs dialogiques pour la réalisation d’une tâche complexe : application à la recherche d’information. Intelligence artificielle [cs.AI]. Normandie Université, 2019. Français. ⟨NNT : 2019NORMIR13⟩. ⟨tel-02923983⟩

Share

Metrics

Record views

60

Files downloads

21