jinsung-jeon / lightmove

LightMove: A Lightweight Next-POI Recommendation for Taxicab Rooftop Advertising(CIKM21)

Mobile digital billboards are an effective way to augment brand awareness. Among various such mobile billboards, taxicab rooftop devices are emerging in the market as a brand new media. Motov is a leading company in South Korea in the taxicab rooftop advertising market. In this work, we present a lightweight yet accurate deep learning-based method to predict taxicabs’ next locations to better prepare for targeted advertising based on demographic information of locations. Considering the fact that next POI recommendation datasets are frequently sparse, we design our presented model based on neural ordinary differential equations (NODEs), which are known to be robust to sparse/incorrect input, with several enhancements. Our model, which we call LightMove, has a larger prediction accuracy, a smaller number of parameters, and/or a smaller training/inference time, when evaluating with various datasets, in comparison with state-of-the-art models.

Usage

Install the environment using yaml file

conda env create --file environment.yaml

Model parameter

• pretrain : 0(Train), 1(Train with pretrained model), 2(Test)
• model_method : 0(G0E) 1(L2E) 2(G2E) 3(G5E)

Train model

• Change localpath to your path

python main.py --data_name  'foursquare2' --data_path '/localpath/LightMove/data/' --save_path /localpath/LightMove/train/ --pretrain 0 --loc_emb_size 100 --uid_emb_size 60 --tim_emb_size 10 --hidden_size 100 --dropout_p 0.3 --model_method 1 --epoch_max 50 --learning_rate 0.005

Train with pretrained model

python main.py --data_name  'foursquare2' --data_path '/localpath/LightMove/data/' --save_path /localpath/LightMove/test/ --pretrain 2 --loc_emb_size 100 --uid_emb_size 60 --tim_emb_size 10 --hidden_size 100 --dropout_p 0.3 --model_method 1 --epoch_max 50 --learning_rate 0.005

Test model

python main.py --data_name  'foursquare2' --data_path '/localpath/LightMove/data/' --save_path /localpath/LightMove/test/ --pretrain 2 --loc_emb_size 100 --uid_emb_size 60 --tim_emb_size 10 --hidden_size 100 --dropout_p 0.3 --model_method 2 --epoch_max 50 --learning_rate 0.005

Authors

Jinsung Jeon, Minju Jo, Seunghyeon Cho, Noseong Park(Yonsei University)
Soyoung Kang(NAVER Clova)
Seonghoon Kim, Chiyoung Song(Motov Inc., Ltd.)

lightmove's Issues

Some questions about results

Hi, I'm interested in your paper, and I use the DeepMove source code and gowalla dataset to generate "gowalla.pk" file. But I got terrible results, ==>Test Top1 Acc:0.0047 Top5 Acc:0.0123 Top10 Acc:0.0159 MRR:0.0086 Loss:11.3840. I use the G0E method and default learning rate 0.005, other parameters follow the script file you provided, could you explain why that happens?