Skip to content

A Spherical Hidden Markov Model for Semantics-Rich Human Mobility Modeling (AAAI 2018)

License

Notifications You must be signed in to change notification settings

WanzhengZhu/SHMM

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

10 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SHMM

This repo is the SHMM model implementation of A Spherical Hidden Markov Model for Semantics-Rich Human Mobility Modeling [AAAI2018].

Table of Contents

Introduction

SHMM, a multi-modal Spherical Hidden Markov Model is designed for semantics-rich human mobility modeling. Under the hidden Markov assumption, SHMM models the generation process of a given trace by jointly considering the observed location, time, and text at each step of the trace.

An example model output is shown below:

Code

All code are in the folder ./code.

Twitter Dataset

  • To run SHMM on Twitter Dataset, run ./code/run_Twitter.sh.

  • Hyperparameters for Twitter and Synthetic data are in ./code/run/*.yaml.

  • Preprocess:

    • Twitter data preprocessing code is in ./code/Twitter/preprocess. It is included in ./code/run_twitter.sh as well.

    • How to generate background data is in ./code/Twitter/Generate_background_tweet.txt.

  • Model Training:

    • The main model for SHMM is in ./code/gmove.
  • Post-process:

    • To plot figures, the code is in ./code/pre-post-process.

    • To analysis keywords for different states, the code is in ./code/Twitter/find_state_keywords.py.

Synthetic Dataset

  • To run SHMM on Synthetic Dataset, run ./code/run_Synthetic.sh.

  • Synthetic Data Generation:

    • The code is in ./code/Synthetic-Data. The main function is generate_VMF_data.m.
  • Synthetic Data Analysis:

    • The code is in ./code/gmove.

Data

Please download the dataset from this link, unzip it and rename the folder data and put it under ./.

Now, all raw data, processed data are in the folder ./data.

Twitter Dataset(LA)

  • Raw tweets is in ./data/tf-la/raw/raw_tweet.txt.

  • Background tweets (for better word embedding) is in ./data/tf-la/raw/background_tweet.txt.

  • Processed data for SHMM training is in ./data/tf-la/input/final.txt.

  • Processed data for Gmove training is in ./data/tf-la/input/sequences.txt, ./data/tf-la/input/words.txt.

  • Results will be saved in ./Results/Twitter-LA/Results_LA.txt

Twitter Data(NY)

  • It has the similar data arrangement as LA.

Synthetic Dataset

  • Synthetic dataset generation requires a template, which is in ./data/Synthetic/a_template.txt.

  • After synthetic data generation, synthetic data for SHMM training is in ./data/Synthetic/synthetic_data.txt.

  • Data parameters for generating the synthetic data is saved in ./data/Synthetic/data_para.txt for evaluation purpose.

Summarized Results

Summarized Results for Twitter dataset and Synthetic dataset are in the folder ./Results.

Citation

@inproceedings{zhu2018spherical,
  title={A Spherical Hidden Markov Model for Semantics-Rich Human Mobility Modeling.},
  author={Zhu, Wanzheng and Zhang, Chao and Yao, Shuochao and Gao, Xiaobin and Han, Jiawei},
  booktitle={AAAI},
  year={2018}
}

About

A Spherical Hidden Markov Model for Semantics-Rich Human Mobility Modeling (AAAI 2018)

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published