Identifying problematic dialogs in a human-computer dialog system

Abstract

In this thesis, we present the development of an automatic system that identifies problematic dialogues in the context of a Human-Computer Dialog System (HCDS). The system we developed is a type of application in Pattern Classification domain. In this work, we propose a probabilistic approach that predicts user satisfaction for each turn of dialogue. To do so, all the features used in our system are automatically extracted from the utterance. A robust and fast machine learning scheme, Hidden Markov Model (HMM) is used to build our desired system. In order to evaluate the system performance, we experimented on two publicly distributed corpora: DARPA Communicator 2000 and 2001. We evaluated the system using a 10-fold stratified cross-validation. Our results show that the system could be used in real life applications.

Date	10 Dec 2010
Original language	American English
Awarding Institution	École de technologie supérieure
Supervisor	Pierre Dumouchel (Supervisor) & Narjès Boufaden (Co-supervisor)