有两个版本，一个是www18的：https://www.comp.nus.edu.sg/~xiangnan/papers/www18-tutorial-deep-matching.pdf
一个是sigir18的：https://www.comp.nus.edu.sg/~xiangnan/sigir18-deep.pdf
好像都403了。。可以看这个http://www.hangli-hl.com/uploads/3/4/4/6/34465961/wsdm_2019_tutorial.pdf
sigir的这个比较新。。看之
这里讲深度学习部分，传统部分见：https://daiwk.github.io/posts/dl-match-for-search-recommendation-traditional.html
AAAI’16 Convolutional Neural Network Architectures for Matching Natural Language Sentences
AAAI’16 Text Matching as Image Recognition
IJCAI’16 Match-SRNN: Modeling the Recursive Matching Structure with Spatial RNN
SIGIR’17 End-to-End Neural Ad-hoc Ranking with Kernel Pooling
WSDM’18 Convolutional Neural Networks for So-Matching N-Grams in Ad-hoc Search
EMNLP 2016 A Decomposable Attention Model for Natural Language Inference
Deep Matrix Factorization
Autoencoders Meeting CF
Collaborative Denoising Autoencoder
Deep Collaborative Filtering via Marginalized DAE
Deep User-Image Feature
Attentive Collaborative Filtering
Collaborative Knowledge Base Embeddings
要求\(Head + Relation \approx Tail\)，也就是说，想让这两个向量尽量是同一个向量，那cos相似度就没啥用了，因为cos只能表示夹角尽量小，可能向量的长度会差很远，所以呢，可以用L1(曼哈顿距离)或者L2距离(欧几里得距离)！！！
直接用欧氏距离，relation向量是通过attention学到的