20 likes | 132 Views
Utility and Legal Discovery. (or: An Old View of IR & High Relevance) Fredric Gey UC Berkeley ICAIL DESI Workshop Meeting June 4, 2007 Stanford University. IR, Precision, Recall and Utility. Almost all IR research is “high precision”
E N D
Utility and Legal Discovery (or: An Old View of IR & High Relevance) Fredric Gey UC Berkeley ICAIL DESI Workshop Meeting June 4, 2007 Stanford University
IR, Precision, Recall and Utility • Almost all IR research is “high precision” • The typical web search user only wants to look at the top 10 documents (web urls) • Legal discovery is “high recall” (want to find “all” the relevant documents) • The usual IR model is cost == payback • $1 cost to examine a document -- $1 value of relevant document • If probability of relevance is 0.01, you pay $100 to find the next relvant document – “it’s time to give up” • However if utility value for the one relevant document (the ‘smoking gun’) at position 100,000 document is $100 million, the picture changes • Legal discovery search should use the utility model XMDR Mapping May 2007Sic-naics-mapping-xmdr.ppt