2018-JDDC大赛第4名的解决方案
MIT License
JDDCJD Dialog Challenge, 2018430201810184
feel free | |
---|---|
4 | |
0.744771 | |
17 | |
0.254222 |
./preliminary
./final
TFIDFTF-IDF (Term Frequency-Inverse Document Frequency) TF-IDFNLPTF-IDFtop 10
TF-IDFTF
IDF
1323Q1A1Q2A2Q3+A3
tfidf1tri-grams2top10A
1
1BM252word2vec3LSItfidf
BM25TFIDF0.3seq2seqTransformerseq2seq+attention+dropout+beam search0.56~0.6transformer+beam search0.7
contextcontextcontextQn-1Qn+AnQA10%Q2Q3+A3QAQAAA[3, 200]QA
transformer+beam searchTransformer state-of-the-art
2Transformer3
EncoderDecoderEncoder6Multi-Head AttentionFeed Forward Neural Networkresidual connectionlayer normalizationsub-layer512Decoder6Multi-Head AttentionFeed Forward Neural NetworkMasked Multi-Head Attentionresidual connectionlayer normalizationAttentionTransformerinput embeddingoutput embeddingpositional encoding
Transformer CNNsRNNsattentionTransformerTensor2Tensor (https://github.com/tensorflow/tensor2tensor) Tensor2TensorTransformer450QAbeam searchsize460.01
Q1A1Q2A2Q3A3QA11
QAQ1A1Q2A2Q3A3
TFIDFA3top10top10A3
10deltaBleuDQNbleuDQN10010top10deltaBleu2
3
tfidf_baselinebaseline
contextcontextcontext100
15015150contextA3
context
contextAcontextcontext
AcontextB12Bcontext43B
B****CcontextB
1context
A | context | |
B | context | |
C | context |
contextCQ2Q3+A3
1seq2seq2transformer3BM25TFIDF
0.30.6
seq2seqSequence to Sequence Learning with Neural Networks0.5seq2seq50.56seq2seq
seq2seqTransformer2017WMTseq2seq
seq2seqTransformerTransformerseq2seqQATransformerAttention Is All You Needbase modelQA1200.61QA450Attention Is All You Needbig modelTransformer0.69Transformer20checkpoints5checkpointsaverage0.73beam size0.74
1baseline
2context
3
4
5
6checkpointaverage
========================================
End-to-EndSeq2SeqfancyEnd-to-End Pipeline (NLU-DM-NLG)
NLPNLPNER
** **
http://www.wildml.com/2016/07/deep-learning-for-chatbots-2-retrieval-based-model-tensorflow/
GALLEY M, BROCKETT C, SORDONI A. deltaBLEU: A Discriminative Metric for Generation Tasks with Intrinsically Diverse Targets[J]. arXiv:1506.06863 [cs], 2015.
VASWANI A, SHAZEER N, PARMAR N. Attention Is All You Need[J]. arXiv:1706.03762 [cs], 2017.
VASWANI A, BENGIO S, BREVDO E. Tensor2Tensor for Neural Machine Translation[J]. arXiv:1803.07416 [cs, stat], 2018.
BORDES A, BOUREAU Y-L, WESTON J. Learning End-to-End Goal-Oriented Dialog[J]. arXiv:1605.07683 [cs], 2016.
SUTSKEVER I, VINYALS O, LE Q V. Sequence to Sequence Learning with Neural Networks[J]. arXiv:1409.3215 [cs], 2014.
QIU M, LI F-L, WANG S. AliMe Chat: A Sequence to Sequence and Rerank based Chatbot Engine[C]//Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers). Vancouver, Canada: Association for Computational Linguistics, 2017: 498503.