ryujitamaki/smn

Sequential Matching Network

Keras implementation of the Sequential Matching Network model.

data

https://www.dropbox.com/s/2fdn26rj6h9bpvl/ubuntu_data.zip?dl=0

Experiments

ubuntu dialogue corpus

Models	R@1	R@2	R@10
SMN_last*	0.723	0.842	0.956
SMN_last	0.698	0.812	0.936

* means the results reported from the original papers

Reference

theano implementation(auther's code): https://github.com/MarkWuNLP/MultiTurnResponseSelection

paper: Wu, Yu, et al. 'Sequential Matching Network: A New Archtechture for Multi-turn Response Selection in Retrieval-based Chatbots.' ACL. 2017.

smn's Issues

dimension mismatch when calculate M1 matrix

Hi,
i try to run train.py but i got the following error message

File "train.py", line 28, in build_SMN
    word_match = dot([context_word_embedding, response_word_embedding], axes=(3, 2))
ValueError: Dimension 1 in both shapes must be equal, but are 10 and 50 for 'dot_1/MatMul' (op: 'BatchMatMul') with input shapes: [?,10,50,200], [?,50,200,1]

python 2.7
tensorflow 1.4.1
keras 2.1.3