A Bichannel Transformer with Context Encoding for Document-Driven Conversation Generation in Social Media

Table 2

Quantitative evaluation results of document relevance and response quality.

Models	Doc_BLEU	NW	avg_len	PPL
SEQ [24]^§	24.88	0.23	7.31	15.62
SEQS [24]^§	27.96	0.34	7.21	19.53
D3G [26]^§	26.76	0.39	6.83	18.40
Transformer [25]^¶	27.55	0.33	7.91	13.70
Incremental transformer [44]^†	—	—	—	15.11
Incremental transformer (our impl)‡	26.96	0.42	8.52	11.01
BCTCE	28.23	0.36	9.16	17.80

The marks in this table are the same as those in Table 1.