User intention prediction for trigger-action programming rule using multi-view representation learning

Trigger-action programming (TAP) allows users to create rules for automating smart devices and Internet services, such as “If a person is leaving, then turn off the air conditioner”. User intention, as the internal driving force behind rule creation, plays a crucial role in understanding user demand...

Celý popis

Uložené v:

Podrobná bibliografia
Vydané v:	Expert systems with applications Ročník 267; s. 126198
Hlavní autori:	Wu, Gang, Hu, Liang, Hu, Yuxiao, Xing, Yongheng, Wang, Feng
Médium:	Journal Article
Jazyk:	English
Vydavateľské údaje:	Elsevier Ltd 01.04.2025
Predmet:	Internet of Things Multi-view representation learning Trigger-action programming User intention prediction Multi-view representation learning Trigger-action programming User intention prediction Internet of Things
ISSN:	0957-4174
On-line prístup:	Získať plný text
Tagy:	Pridať tag Žiadne tagy, Buďte prvý, kto otaguje tento záznam!

Popis
Shrnutí:	Trigger-action programming (TAP) allows users to create rules for automating smart devices and Internet services, such as “If a person is leaving, then turn off the air conditioner”. User intention, as the internal driving force behind rule creation, plays a crucial role in understanding user demands. While existing research can easily identify users’ explicit demands, such as turning off the air conditioner, it often fails to recognize implicit intentions, such as energy saving. Predicting user intention presents several challenges, including (1) contextual dependencies between triggers and actions that lead to different intentions when users combine various functions, (2) identifying user intention in short function description texts, and (3) entities with similar functions that do not reflect user intentions due to a lack of domain-specific knowledge. To address these challenges, we propose a multi-view representation learning-based framework, MvTAP, jointly in (1) user view, (2) developer view, and (3) knowledge view. In MvTAP, we first design different methods to learn the representation of the three views, respectively. We then propose a transformer-based method to fuse the representations of these views. User intention prediction is formulated as a multi-label classification task. The effectiveness of MvTAP is validated using a real-world IFTTT dataset. Experimental results show that the proposed framework can effectively predict user intentions in TAP rules. [Display omitted] •Multi-view representation learning framework for user intention prediction in TAP.•Design user, developer, and knowledge views for enriched intention understanding.•Propose a transformer-based method for fusing representations from different views.•Extensive experiments verify the effectiveness of the proposed framework.
ISSN:	0957-4174
DOI:	10.1016/j.eswa.2024.126198