shuuchen / keras_mn

A Keras implementation of end-2-end memory networks https://arxiv.org/pdf/1503.08895.pdf

keras_mn

Memory networks are used for automatic question answering.

The model is trained using story, question, answer triplets and tested using story, question to predict the answer.

Requirements

• nltk==3.2.3
• matplotlib==2.0.2
• Keras==2.0.6
• numpy==1.12.1

Train/Test data

bAbI Tasks Data 1-20 (v1.2) https://research.fb.com/downloads/babi/

Test result

The model is tested against all 10,000 samples in test data. The first 10 results show that only one mis-prediction happens.

Story	Question	Answer	Prediction
john travelled to the hallway . mary journeyed to the bathroom .	where is john ?	hallway	hallway
daniel went back to the bathroom . john moved to the bedroom .	where is mary ?	bathroom	bedroom
john went to the hallway . sandra journeyed to the kitchen .	where is sandra ?	kitchen	kitchen
sandra travelled to the hallway . john went to the garden .	where is sandra ?	hallway	hallway
sandra went back to the bathroom . sandra moved to the kitchen .	where is sandra ?	kitchen	kitchen
sandra travelled to the kitchen . sandra travelled to the hallway .	where is sandra ?	hallway	hallway
mary went to the bathroom . sandra moved to the garden .	where is sandra ?	garden	garden
sandra travelled to the office . daniel journeyed to the hallway .	where is daniel ?	hallway	hallway
daniel journeyed to the office . john moved to the hallway .	where is sandra ?	office	office
john travelled to the bathroom . john journeyed to the office .	where is daniel ?	office	kitchen

Model plot

License

Released under the MIT license. See LICENSE for details.