Relevance Feedback

1
Relevance Feedback

• Dr. Ying Xie

2
Web Search Process
Crawler
Information Need
Formulation
Indexing
Query Rep
Inverted index and web graph
Ranking
Ranked List
Learning
User Relevance Feedback
3
Outline

• Users Relevance feedback
• How VSM utilizes the relevance feedback
• How Language Model utilize relevance feedback

4
Users Relevance Feedback (1)

• User can explicitly provides relevance feedback

5
Users Relevance Feedback (2)

• Or, the system can guess the users relevance
  feedback by monitoring the users browsing
  behavior.
• - which URLs the user clicked
• - which docs the user download or printed
• - which URLs have the users eye focus for
  certain time.
• -

6
Users Relevance Feedback (3)

• Pseudo Relevance feedback
• The top n documents are relevant

7
How the system utilizes relevance feedback?

• Expand the set of query terms
• Adjust the weight of each query term

8

• How VSM utilizes the relevance feedback

9

• Document Representation
• Query Representation
• Ranking Function

10
What is the optimal query under VSM

• The optimal query should be the one that is as
  similar to the relevant documents as possible, as
  dissimilar to the non-relevant documents as
  possible.

11
Rocchio Method
12
Assignment 2

• Document collection
• Doc0 Modern information retrieval
• Doc1 Boolean retrieval model
• Doc2 Vector space model
• Doc3 Language Model for Information Retrieval
• Doc4 Vector Space model, Boolean model,
  Language model for information retrieval
• Users initial query
• Retrieval model
• Users relevance feedback
• Doc3, Doc4 are relevant. The rest are not
• Question 1. Please use Rocchios method to
  figure out the new query from users relevance
  feedback. Please set
• 2. Please compare the
  performance between the new query and the old
  query by drawing the Precision-recall graph for
  each of them