NIPS ˙08 Workshop: Beyond Search - Computational Intelligence for the Web

Online Search and Advertising, Future and Present

author: Chris Burges, Microsoft Research
published: Dec. 20, 2008, recorded: December 2008, views: 449

Slides

Slides
0:00	Online Search and Advertising, Future and Present
0:25	Contents
1:17	~ Search and Advertising ~
1:24	Why Search Works…
2:13	Key Points
2:35	What‟s wrong with what we do now?
3:40	How might ads be targeted better?
4:42	User Models
7:06	Key Points
7:47	What About Search?
7:57	Search: Somewhere in the Near Future
10:40	Example : MS LiveSearch
10:59	Search: Somewhere in the Near Future
11:20	How to get the information we need, to build good models for users?
11:44	Key Points
12:28	Search Applications: And,Data Changes Everything (1)
14:28	Search Applications: And,Data Changes Everything (2)
16:27	Data Changes Everything (1)
17:13	Data Changes Everything (2)
17:54	Key Points
18:51	Key Points
19:02	How To Proceed?
20:55	The Eliza Effect
21:53	Our Prime Directive in Building Sam:
22:35	Let the Data do the Work
23:27	Using Category Graphs to Drive Dialog
24:08	Use Category Graphs to Build Models
25:15	Other Useful Sources of Data
27:24	Temporal Querying Behavior
28:28	We Are Not Alone
29:50	One Possible Sentence Generator
31:18	New Challenges for Machine Learning
33:02	Demo
36:51	~ Some New Results on Ranking ~
38:14	Empirical Optimality of lambda -rank
38:14	Some IR Measures
39:11	IR Measures, cont.
39:38	LambdaRank: Background
39:42	The RankNet Cost (1)
40:32	The RankNet Cost (2)
40:59	RankNet Cost ~ Pairwise Cost
41:14	Pairwise Cost Revisited
41:57	LambdaRank
42:10	The Lambda Function
42:54	Lambda Functions for MAP, MRR
43:18	Local Optimality
45:10	Data Sets
45:15	Which lambda-function to choose? (1)
45:27	Which lambda-function to choose? (2)
46:05	Sampe size matters
47:11	IR Measure Optimality -Conclusions
48:07	~ RSA, Factoring, and Optimization ~
48:26	Factoring biprimes as optimization
49:23	Circumstantial Evidence That Factoring is Not NP-hard
50:27	Is This The Best We Can Do?
50:39	RSA Challenge
51:10	Represent the Problem in Binary
51:50	First Trick: Linearization
52:02	Linearization, cont.
52:22	A Geometrical Problem (1)
53:26	Second Trick: Quantization (2)
53:44	More Simple Tricks
54:10	The Geometric View
54:37	Projections Lose Information
54:57	Conclusions
55:49	- questions

Report a problem or upload files

If you have found a problem with this lecture or would like to send us extra material, articles, exercises, etc., please use our ticket system to describe your request and upload the data.
Enter your e-mail into the 'Cc' field, and we will keep you updated with your request's status.

Lecture popularity: You need to login to cast your vote.

Description

Search engine companies are gathering treasure troves of user-generated data. It has already been shown that such data can be used to directly improve the user's online experience. I will discuss some ideas as to what online search and advertising might look like a few years hence, in light of the algorithms and data we have now. Moving from future to present, I will outline some recent work done by researchers in the Text Mining, Search and Navigation team at Microsoft Research; the work in TMSN touches many aspects of online search and advertising.