twitter比较关注social graph的挖掘
use case & product
- homepage or related item recommendation
- user: follow
- item: text, image, video
- engagement: click, like, comment, share
objective
- increase the engagement
constraint
- scale of user and item
- latency
召回、精排、规则多样性重排、混排
- Fetch the best Tweets from different recommendation sources in a process called candidate sourcing.
- Rank each Tweet using a machine learning model.
- Apply heuristics and filters, such as filtering out Tweets from users you’ve blocked, NSFW content, and Tweets you’ve already seen.
- user
- demographics
- item
- text
- engagement
- impression, engagement
- context
- device
- time
- label
- dense
- sparse
- In-Network召回
- Out-of-Network 召回
- MaskNet
- 过滤已屏蔽用户的推文、NSFW内容和已看过的推文
offline
- recall@k, hit_rate
online
- ctr
- batch service or online service
- A/B testing