Combining language models and BiLSTM for next-device prediction

Sheng Zhang; Fan Tang; Tianqi Zhang; Sen Fan

doi:10.1117/12.2636585

8 May 2022 Combining language models and BiLSTM for next-device prediction

Sheng Zhang, Fan Tang, Tianqi Zhang, Sen Fan

Proceedings Volume 12249, 2nd International Conference on Internet of Things and Smart City (IoTSC 2022）; 122492C (2022) https://doi.org/10.1117/12.2636585
Event: 2022 2nd International Conference on Internet of Things and Smart City (IoTSC 2022）, 2022, Xiamen, China

Abstract

In the smart home scenario, the heterogeneity of IoT devices makes it difficult to discover the potential relationship between devices, and the existing device search or recommendation methods cannot provide the next device used by the user according to the user's device usage order. We propose a next-device prediction method that combines a Natural Language Processing (NLP) language model and BiLSTM. This method extracts the device activity sequence from the IoT device activity log, and selects two embedding methods, Word2Vec for static word embedding and BERT for dynamic word embedding, to represent the correlation of devices. The BiLSTM network is trained to predict the next device a user will use. The experimental results show that, compared with deep learning models such as SPEED-LSTM, CNN-BiLSTM and BiLSTM-ATT, Word2Vec-BiLSTM and BERT-BiLSTM models have better prediction accuracy, precision, recall and F1 evaluation indicators. Effect. The correct rate of using Word2Vec-BiLSTM on the HH101 dataset is 74.86%, and the correct rate of the BERT-BiLSTM model is 77.43%.

Citation Download Citation

Sheng Zhang, Fan Tang, Tianqi Zhang, and Sen Fan "Combining language models and BiLSTM for next-device prediction", Proc. SPIE 12249, 2nd International Conference on Internet of Things and Smart City (IoTSC 2022）, 122492C (8 May 2022); https://doi.org/10.1117/12.2636585

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available