en-de
en-es
en-fr
en-sl
en
en-zh
0.25
0.5
0.75
1.25
1.5
1.75
2
Online Learning with Implicit User Preferences
Published on 2012-01-244572 Views
Many systems, ranging from search engines to smart homes, aim to continually improve the utility they are providing to their users. While clearly a machine learning problem, it is less clear what the
Related categories
Presentation
Online Learning with Implicit User Preferences00:00
Today‟s Knowledge-Based Systems00:46
Learning for Search Engines01:51
Overview02:56
The Most Basic Learning Problem04:57
Approaches to Implicit Utility Elicitation05:41
How does User Behavior Reflect Retrieval Quality? -106:57
How does User Behavior Reflect Retrieval Quality? -208:14
Absolute Metrics: Metrics09:09
Absolute Metrics: Results - 110:13
Absolute Metrics: Results - 211:45
Approaches to Utility Elicitation12:05
Paired Comparisons: What to Measure?12:46
Paired Comparisons: Balanced Interleaving13:20
Balanced Interleaving: Results -115:32
Balanced Interleaving: Results -218:30
Problem: Learning on Operational System18:42
Regret for the Dueling Bandits Problem21:01
First Thought: Tournament22:44
Algorithm: Interleaved Filter 223:51
Assumptions -125:46
Assumptions -226:15
Lower Bound26:58
Dueling Bandits: Infinite K -127:02
Dueling Bandits: Infinite K -227:41
Experiment: Web Search28:05
Who does the exploring? -129:45
Who does the exploring? -230:34
Who does the exploring? -330:49
Strategy for Within-Ranking Feedback31:09
User Study31:45
Strategy for Between-Ranking Feedback -133:44
Strategy for Between-Ranking Feedback -233:53
Does Accuracy Depend on Quality?34:48
Preference Online Learning Model34:51
Preference Perceptron Algorithm37:04
Preference Perceptron Regret Bound38:55
Conclusions42:20