There are two types of related feedback:
"Real" related feedback:
1. System return Results
2. Users provide some feedback
3. Based on these feedback, the system returns a number of different, better results
Related feedback on "assumptions"
1. The system obtains the result but does not return the result
2. The system improves query based on these results
3. Get results based on improved query and return
Rocchio ' s Modified Query
Modified query vector = Original query vector + Mean of relevant documents found by Original Query-mean of Non-relevant Documents found by original query
Q0 represents the original query
R represents the associated Document set
s represents an unrelated document set
Local Context Analysis (LCA)
Algorithm process:
the user enters a query and uses this query to Search article: Find the article most relevant to this query, using a 300-word sliding window to get the article.
Then find the candidate term: first of all, the article in part-of-speech tagging, select all the nouns as candidates.
Calculate the weight of a term:
The meaning of En (c,w): When C,w is independent of each other, there are simultaneous expectations of N (nw/n) (nc/n)
Why would co_degree lose one? Prevents a very small NC condition.
Select a new term based on the weights to add a query and refine the query to get new results.
[IR Course notes] Query Refinement and relevance Feedback