Exploiting future information for next point-of-interest recommendation
The next Point of Interest (POI) recommendation has recently received increased attention from recommender system researchers and the general public. The goal is to predict user preferences based on historical check-in records to understand human behaviour better and suggest the next potential locat...
Saved in:
Main Author: | |
---|---|
Other Authors: | |
Format: | Final Year Project |
Language: | English |
Published: |
Nanyang Technological University
2022
|
Subjects: | |
Online Access: | https://hdl.handle.net/10356/157235 |
Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
Institution: | Nanyang Technological University |
Language: | English |
Summary: | The next Point of Interest (POI) recommendation has recently received increased attention from recommender system researchers and the general public. The goal is to predict user preferences based on historical check-in records to understand human behaviour better and suggest the next potential locations to users. Over the years, various approaches have been developed for the next POI recommendation, such as Matrix Factorization (MF), Recurrent Neural Networks (RNNs), and Graph Neural Networks (GNNs). In the next POI recommendation scenario, many complex models are built based on Long-Short Term Memory (LSTM) like ST-RNN and LSTPM, which introduce contextual information such as spatial or temporal factors to improve model performance. This report proposed an improved model PFST-LSTM based on the published model ATST-LSTM and LSTPM. It combines the attention mechanism of the ATST-LSTM model with potential future preferences, which are inspired by the time slot concept of the LSTPM model. Therefore, it not only considers the relationship between sequences but also adds more information about potential future preferences apart from temporal and spatial information. Experimental result shows that the potential future preference significantly impacts the proposed model and helps it achieve better prediction performance. |
---|