how to extend Click through model example to more than one rank list? RRS feed

  • Question

  • I want to extend the click through model example(http://research.microsoft.com/en-us/um/cambridge/projects/infernet/docs/Click%20through%20model%20sample.aspx) to support more than one rank list because advertising lists are more likely to change than search result list .

    suppose 2 rank lists for a query q:

    q            url1, url2, url3, ...

    clicks       50   40   30

    q            url2, url3, url4, ...

    clicks      45   42    28

    in the Click through model example, each rank list has 2 Variable arrays:

            private Variable<double>[] appeal;
            private Variable<double>[] relevance;

    but for these 2 rank lists, query is the same and if url is the same, the appeal and relevance should also be the same.

    one solution is creating only one Variable for each url, but it will make the graphical model change whenever the data changes.

    is there any better method?

    Tuesday, April 9, 2013 3:06 AM

All replies

  • For each url/document pair you will need a separate variable. There are efficient ways to implement this but it is difficult to advise on specifics without knowing more details. For example, in an offline scenario you could have a jagged array of appeals (and relevances) indexed by query (the outer index) and by all the docs that appear in any list for that query (the inner index). Then, for each impression, you could have an observed index array into the document array for that query.


    Tuesday, April 30, 2013 9:11 AM