Date of Graduation
Summer 2024
Degree
Master of Science in Computer Science
Department
Computer Science
Committee Chair
Razib Iqbal
Abstract
Voice-enabled virtual assistants have gained widespread popularity and are increasingly common in smart homes. To enhance customization and personalization in user experiences with these assistants, implementing a context detection feature is beneficial. This feature enables the virtual assistant to gather more information from the audio data of short voice conversations with users, helping it maintain awareness of the conversation and respond more aptly. In this thesis, I propose a novel context detection approach for virtual assistants in smart homes, named WERKS, which leverages user emotions. WERKS stands for word embedding with emotionally relevant keyword search. The WERKS approach incorporates emotion detection, keyword search, and word embedding from voice commands and short conversations to achieve effective context detection. This method comprises emotion detection, basic context detection, word embedding with emotionally relevant keyword search, and ensemble defined context classification layers. Evaluation of the WERKS approach on datasets has demonstrated that it can significantly improve context detection accuracy.
Keywords
audio, bigram, emotion detection, neural network, noun phrase, synset
Subject Categories
Computer Sciences
Copyright
© Brent Anderson
Recommended Citation
Anderson, Brent, "Context Detection With Word Embedding and Emotionally Relevant Keyword Search for Smart Home Environment" (2024). MSU Graduate Theses/Dissertations. 3985.
https://bearworks.missouristate.edu/theses/3985
