Trend Analysis of Belief-State History with Discrete Wavelet Transform for Improved Intention Discovery

Software Product Lines (SPL) have emerged as a new paradigm of software development. By means of mass production of customized software products, SPL has the potential to significantly reduce development time and cost while improving the quality of software systems. Currently, there is still a severe shortage of tools that support the decision-making process for software clients to interactively "order" software products due to the difficulty of software customization, especially via dialogue in natural language. While most of the existing approaches use POMDP-based dialogue management, this thesis research proposes to introduce historical information of belief states into the POMDP model and to analyze its trend with discrete wavelet transformation (DWT). Accordingly, a new algorithm is developed to improve the accuracy of intention discovery with trend analysis, and to reduce the dialog length by switching POMDP policies between contextual control modes according to the anticipated knowledge of different users. The efficiency and accuracy of the proposed method are examined by experiments with simulation.

Name thesis:degree_name

M.Sc.

Level thesis:degree_level

Masters

Discipline thesis:degree_discipline

Computer Science

Grantor

University of Windsor

Year dc:date.issued

2016

Author dc:creator

• Mulpuri, VIjaya Krishna

Advisor dc:contributor.advisor

• YUAN, XIAOBU

Licence

• http://creativecommons.org/licenses/by-nc-nd/4.0/

Language dc:language.iso

en_CA

Handle dc:identifier.uri

https://hdl.handle.net/20.500.14776/6312

OAI identifier oai:identifier

oai:uwindsor.scholaris.ca:20.500.14776/6312