Rank Manipulation for Relational DBs

1
Rank Manipulation for Relational DBs

• Learning unavailable ranking functions for the
  purpose of manipulation
• Much effort already in information retrieval
• People want their Web documents to appear first
  in search engine results
• Apply existing techniques from information
  retrieval or develop new techniques for the top-k
  over relational database case
• Internal ranking functions sometimes use weighted
  attributes where the weights arent specified by
  users of the system
• Business owners goal and strategy
• Have their products appear higher in the rankings
  by focusing on weights
• Change product line attributes to achieve better
  rankings
• An approximation might suffice
• Combating rank manipulation
• Want to be fair to all businesses that have
  products listed

William Conner, CS 497 KCC
2
Rank Manipulation for Relational DBs

• Example of learning an unavailable ranking
  function
• What can be learned from the ranked output of a
  top-k query?
• Example query inexpensive car that is fuel
  efficient
• Ranked output
• Car is 8000, gets 25 MPG, all other attributes
  are X (ranked 3)
• Car is 8025, gets 40 MPG, all other attributes
  are X (ranked 47)
• Maybe something can be learned, maybe not
• How can this information be exploited?
• Used car dealer might purchase more inexpensive
  cars for their dealership rather than investing
  in cars that are more fuel efficient
• Only works if the improved rankings offset any
  potential financial losses that result from
  changing the product line
• On the other hand, find complex ranking functions
  that keep potential manipulators guessing without
  taking a big hit in performance

William Conner, CS 497 KCC