Reference article:
https://mp.weixin.qq.com/s?__biz=MzI3MTA0MTk1MA==&mid=2652000715&idx=4&sn=b7c68c03e7f507cdce3626c01b018259&chksm=f121253ac656ac2c8d82a852efc2527601adccfca917444b23d5114e9022079dc473a5bd6436&mpshare=1&scene=1&srcid=071314t1VnJ0jGFjEKYiMhFj&key=df6095916712388e205c3bb3c57943443c0ba39ac566db39833c1cbebdba4a9fc8d5d308b7261927e70c536d68e3a19caba940fd8016a156312edbcd75447d2e2508760cffd01cec189244826c54258f&ascene=0&uin=MTczOTkwNjE2MQ%3D%3D&devicetype=iMac+MacBookPro11%2C5+OSX+OSX+10.12.5+build(16F73)&version=12020810&nettype=WIFI&fontScale=100&pass_ticket=3V7zt%2BdPOCWOpz6om5o9Ce4Xw6OObLP2J%2Fuw5ZdePCUarF1Ov4Xv3uzzaTclcFRn
Briefly introduce the central idea :
The attention model is to set the weight according to the contribution of each word in the input sentence to each word in the output sentence. Input sentence , by f ( different per paper ) is calculated for each input word and the intermediate results hi possibility aligned , i.e., encode then Ci of each input word j of the output word i affect the size of the weight , and then decode sequentially output sentence to give For each word in , if it encounters a terminator , the end of the sentence will be output.