Awesome

SERM

This is the code of CIKM 17 short paper: "SERM: A Recurrent Model for Next Location Prediction in Semantic Trajectories".

About

In this repository, we propose a RNN-based model, namely SERM, to predict next location of LBSN users. SERM considers both squencial relations and semantic influence to generate the prediction result.

Files

There are four python files in the root path.

Filename	Description
config.py	All configurations in SERM model.
eval_tool.py	Evaluation functions and tools.
geo_data_decoder.py	Preprocessing for both New York Foursquare and Los Angelos Geo-Tweets.
train.py	Training procedure of SERM. This is the entrace of SERM model.

Package	Description
model	Source code of SERM model.

Dataset and external data

There are four floders to store the dataset and external data. Some large files are not available in this repository. Please download on link: https://pan.baidu.com/s/1NKZ4Tq86VIP0Ae5gSGGVcw password: 59z1

Folder	Description
data	Path of New York Foursquare and Los Angelos Geo-Tweets datasets.
features	Features generated by preprocessing and decoder procedure.
pretrained	Pretrained model of SERM which is able to reproduce the experimental result.
word_vec	Pretrained Glove word embeddings.

Reference

@inproceedings{YaoZHB17,
  author    = {Di Yao and
               Chao Zhang and
               Jian{-}Hui Huang and
               Jingping Bi},
  title     = {{SERM:} {A} Recurrent Model for Next Location Prediction in Semantic
               Trajectories},
  booktitle = {Proceedings of the 2017 {ACM} on Conference on Information and Knowledge
               Management, {CIKM} 2017, Singapore, November 06 - 10, 2017},
  pages     = {2411--2414},
  year      = {2017},
  url       = {https://doi.org/10.1145/3132847.3133056},
  doi       = {10.1145/3132847.3133056},
}