15
Intelligently matching users to questions for reading and writing Nikhil Dandekar @nikhilbd 6/8/2016

Intelligently matching users to questions for reading and writing

Embed Size (px)

Citation preview

Page 1: Intelligently matching users to questions for reading and writing

Intelligently matching users to questions for reading and writing

Nikhil Dandekar @nikhilbd

6/8/2016

Page 2: Intelligently matching users to questions for reading and writing

Quora’s Mission

“To share and grow the world’s knowledge”

● Millions of questions & answers

● Millions of users

● Over a million topics

● ...

Page 3: Intelligently matching users to questions for reading and writing

Matching users to questions for reading

Page 4: Intelligently matching users to questions for reading and writing

Feed ranking

Goal: Present most interesting stories for a

user at a given time

Page 5: Intelligently matching users to questions for reading and writing
Page 6: Intelligently matching users to questions for reading and writing

Feed ranking

Challenges:

● Millions of potential stories

● How do you measure and predict interestingness?

● Real-time ranking

Page 7: Intelligently matching users to questions for reading and writing

How do you measure interestingness?

click

upvote

downvote

expand

share

Page 8: Intelligently matching users to questions for reading and writing

Weighted sum of actions =

∑a va p(a | x) =

vclick p(click | x) + vupvotep(upvote | x) + vdownvote p(downvote | x) + ….

Predict this value for new stories. Rank stories by this

predicted value.

How do you measure interestingness?

Page 9: Intelligently matching users to questions for reading and writing

Machine Learning for Feed ranking

● Essential for getting good ranking

● Main sets of features:

○ User

○ Story

○ Interactions between the two

Page 10: Intelligently matching users to questions for reading and writing

Matching users to questions for writing

Page 11: Intelligently matching users to questions for reading and writing

Question Answering

Asker goal: Get a great answer to the

question in the shortest amount of time

Answer writer goal: Find the best questions

that you can write a great answer to

Page 12: Intelligently matching users to questions for reading and writing

Question Answering

Solution: Route the question to the best

person who can answer the question

Page 13: Intelligently matching users to questions for reading and writing
Page 14: Intelligently matching users to questions for reading and writing

Machine Learning for “Ask to Answer”

● Target

○ 1 if answer request sent and answer has goodness score of above

0.5

○ 0 otherwise

● Main sets of features:

○ Asker

○ Answer writer

○ Question

○ Various interactions between the three

Page 15: Intelligently matching users to questions for reading and writing

Questions?