3 M F3 ^. S8 \7 _1 _" j8 f10. 小结 w6 _+ v4 W: _( A# W# ]
交互型语言匹配模型由于引入各种花式attention,其模型的精细度和复杂度普遍强于表示型语言模型。交互型语言匹配模型通过尽早让文本进行交互(可以发生在Embedding和/或Encoding之后)实现了词法、句法层面信息的匹配,因此其效果也普遍较表示型语言模型更好。8 M. b% @ c! s: v
: J# ]4 L ^/ d B
【Reference】# D4 h- Y7 M7 q. q
3 E6 F% ?. N7 H" O/ tARC-II: Convolutional Neural Network Architectures for Matching Natural Language Sentences 7 p6 S' ^: m, @8 [1 t; c6 O, R2 I 7 y, Z A7 Q7 k2 EPairCNN: Learning to Rank Short Text Pairs with Convolutional Deep Neural Networks- l6 ?3 u9 G" B* v4 N; U \! t8 c
7 f! x. N, J' K% j, v# z! s7 DMatchPyramid: Text Matching as Image Recognition2 s7 @5 Z3 w9 h0 D ~/ ^7 k9 {4 C; y1 [
& A: ^3 l7 `' M/ m& f! A2 C& E& t) z
DecAtt: A Decomposable Attention Model for Natural Language Inference 4 ]$ s# Z4 [ w: w ' j. w8 M7 ~* OCompAgg: A Compare-Aggregate Model for Matching Text Sequences6 }% A0 \0 l j) h0 [9 G3 E0 q% D4 t" D
0 x: G% z% @3 V: s- T( v7 _, ?5 R
ABCNN: ABCNN: Attention-Based Convolutional Neural Network / s7 _5 h/ d* e9 R- Zfor Modeling Sentence Pairs% Y# O" W z1 C4 B3 D+ B: O
7 {# h! u* J) | L: v) |/ ]/ |% Q$ v! EESIM: Enhanced LSTM for Natural Language Inference- `3 E1 I/ S3 d6 `
; z7 h/ C- A; s1 |; ^3 R2 X* g
Bimpm: Bilateral Multi-Perspective Matching for Natural Language Sentences7 B" k. m+ X5 @& l# Y; l' L. e% `" v
, t. _( q$ \; U5 ]1 G
HCAN: Bridging the Gap Between Relevance Matching and Semantic Matching ) B7 J" s$ L( X3 I- Y% q$ ]8 ^for Short Text Similarity Modeling 4 _( M' P1 U0 }$ u 5 b4 i1 d* V, n% x5 V6 ^$ I文本匹配相关方向打卡点总结(数据,场景,论文,开源工具) 7 |/ U0 r/ \0 I7 V. B; Q4 p; l$ u* \: d9 d+ v
谈谈文本匹配和多轮检索& R7 j% y. G8 l* i* A, B8 [ M* `