AttGGSNN

AttGGSNN is an attention-based gated graph sequence neural network, which is built in an end-to-end model to formulate a VQA-like system.

The Example of Input data, question and target:

V means the multiple time-series sequences for all during a period of time before the predicted timestamp.
Q means the question of "Pairwise" companies.
A means the answer of the question we set.
Noted that this is a QA-like system, where the questions are embedded according to the keywords instead of the whole sentence, and the purposed model is used to forecast the future market information such as trend, volatility.

The System Overview:

Readme.docx is a document written in Chinese, and it shows how to run the codes.

This is the programs of my thesis for Master of Science in Information Management in National Chiao Tung University. The format of citation is : Wei-Chia Huang (2019). "A Question Answering System for Financial Time-Series Correlation Based on Improved Gated Graph Sequence Neural Network with Attention Mechanism". Master's Thesis of Institute of Information Management, Hsinchu: National Chiao Tung University. Retrieved from https://hdl.handle.net/11296/hu4b8r.

AttGGSNN_1DConv.ipynb is the main version of my work, which implement my proposed method.

RelationNetwork_Benchmark.ipynb is the benchmark model for my proposed model - Attention based GGSNN.

Rule-based_PairsTrading.ipynb shows you how I operate the experiment used for simulating the real world case. And it's a rule-based pairs-trading, a market neutral strategy aims to do statistical arbitrage.