KGEvetPred

Introduction

The evolution and development of breaking news events usually present regular patterns, leading to the happening of sequential events. We propose a framework with a pipeline procedure from event extraction to event prediction. Considering the different event domains, we offer a domain-aware event prediction method which has been shown superiority over existing approaches.

This project mainly includes two major tasks:

• Event Evolution Knowledge Graph Generation
• Event prediction

For a detailed description and experimental results, please refer to our paper: Event prediction based on evolutionary event ontology knowledge.

For the dataset, please refer to: Evolutionary Event Ontology Knowledge(EEOK).

Event Evolution Knowledge Graph Generation

Description

This task first analyzes the news data, extracts the event chain and event elements from it, and finally generates an event graph.

Files

• /corpus：stores raw data
• /front：stores front-end related code
  • more details in front-end readme.md
• /utils：stores generation event tools. There are detailed comments in this part of the file, you can directly open the file to view the function description.
  • ltp analyzer
  • ltp formatter
  • graph manager
  • tools

Precautions

There are still some imperfections in the event analysis part, and TODO is used to mark the part that can be improved.

Event prediction

Description

1. Generate experimental data:
   • In /corpus is the generated experimental data
   • /scripts/make_dataset.ipynb is the script to generate experimental data
2. Model
   • /models contains model documents
   • There are 7 models in total
   • Model training can use /train.py
   • /start.sh is to use the server to run the model training script
3. Visualization of results
   • /scripts/draw.py is a script for visualizing model training results, /scripts/getscore.py is a script for obtaining model score data, which is used to assist drawing

Precautions

1. Create a log folder in the running directory
2. When testing the model (in the model code file) in a single file, you need to comment out the register line

Others

1. Many files are not uploaded on github, you can find me to copy them directly, please refer to /.gitignore
2. The required packages are in /requirements.txt
3. /my_logger.py is the logger script, and the log is stored in /log

Citation

If this repo helps you, please cite our paper.

@article{MaoLPLHGHW21,
  author    = {Qianren Mao and
               Xi Li and
               Hao Peng and
               Jianxin Li and
               Dongxiao He and
               Shu Guo and
               Min He and
               Lihong Wang},
  title     = {Event prediction based on evolutionary event ontology knowledge},
  journal   = {Future Gener. Comput. Syst.},
  volume    = {115},
  pages     = {76--89},
  year      = {2021},
  url       = {https://doi.org/10.1016/j.future.2020.07.041},
  doi       = {10.1016/j.future.2020.07.041},
  timestamp = {Fri, 18 Dec 2020 10:25:23 +0100},
  biburl    = {https://dblp.org/rec/journals/fgcs/MaoLPLHGHW21.bib},
  bibsource = {dblp computer science bibliography, https://dblp.org}