Xavier Amatriain,Quora Engineering VP. Take a look at his shared Quora recommendation system (REcommender Systems, which will also be shortened to Recsys) Build experience.

The mission of the site


Review images

Quora data


Review images

Lots of high quality text

Review images

A lot of data correlation

Review images

Quora’s recommendation system


Review images

On Quora, recommendations are used in many places.

Review images

model

Review images

Lessons learned from the recommendation system construction process


  1. implicit signals beat explicit ones (almost always)

Review images

Review images

(Note: Explicit signals refer to feedback collected directly, such as asking users to rate, or clicking on a thumbs down/thumbs down button. Implicit signals refer to information analyzed by user behavior, such as user logs.)

Review images

Review images

2.be thoughtful about your training data

Review images

Review images

3. your model will learn what you teach it to learn

Review images

Review images

Review images

4.explanations might matter more than the prediction

Review images

Review images

5. if you have to pick one single approach, matrix factorization is your best bet

Review images

Review images

6. everything is an ensemble

Review images

Review images

Review images

Review images

7. building recommender systems is also about feature engineering

Review images

Review images

Review images

Review images

Review images

8. why you should care about answering questions (about your recsys)

Review images

Review images

Review images

9. Data and models are great. You know what’s even better? The right evaluation approach!

Review images

Review images

Review images

Review images

10. You don’t need to distribute your recsys

Review images

(Does it have to be distributed? Not really)

Review images

Review images

conclusion


Review images

Slide link: http://www.slideshare.net/xamat/recsys-2016-tutorial-lessons-learned-from-building-reallife-recommender-systems

(after)

The recent hot,


Eliminate small Huang Tu big sword | X yuan pocket

Yahoo open source Pulsar large-scale Pub-sub messaging system

YouTube recommendation system based on deep neural network

Full explanation of the mark-sweep garbage collection algorithm

Recommended 10 architecture, big data aspects of dry goods