From: ted.dunning@gmail.com   
      
   You say that purchase is not necessarily what you are trying to   
   predict, but that is really just a bit of a red herring.  Purchase and   
   revenues happened to be what I worked on most, but any other form of   
   measurable outcome is reasonable.   
      
   Much more importantly, however, purchase was only what I was   
   predicting, not what I was using as input.  As input, I used direct   
   monitoring of use which correlates much better to preference than does   
   purchase and vastly more than any outcome of an artificial task such as   
   rating or assigning a value.   
      
   There is a long tradition in marketing that people don't tell you what   
   they are thinking and that they are really bad reporters of their own   
   motivations.  I believe that there was an instructive study in which   
   posters were presented to experimental subjects who were asked which   
   poster they liked.  Some of the subjects were asked to explain why they   
   liked the poster they selected, others were not asked.  Subsequent   
   followup showed that those who had to explain their preference did not   
   select the same posters some months later as much as those who did not   
   have to explain their preference.  I don't have the reference,   
   unfortunately.   
      
   The inescapable conclusion is that explained preference and unexplained   
   preference are different things cognitively.   
      
   This study conforms very directly to my own experience which showed   
   that the only really good data for determining likelihood that somebody   
   would like to play a bit of music was a history of what they had   
   already played.  Ratings were like explanations and, worse, people who   
   want to play music don't want to make ratings.  Browsing is also a poor   
   indicator of preference.  Even purchase is a highly compromised   
   indicator.   
      
   *That* is my first point ... that you have to use direct consumption to   
   get really, really good results.   
      
   My second point is that you have to have a really solid and objective   
   goal and practical goals are often composite and complex.  Revenue is   
   simply and example of a target variable that lots of people seem to   
   care about.   
      
   I guess it comes down to a question of tense.  When you say "I solve   
   this by ...", do you mean that you have already built a system that   
   succeeds for this task (which would justify the present tense)  or do   
   you mean that you PLAN to build a system using this method (which would   
   only justify the future tense).  In my case, I use the past tense.  I   
   did build such a system and know from experience that there are many   
   slips between paper design and accurate predictions.   
      
   [ comp.ai is moderated.  To submit, just post and be patient, or if ]   
   [ that fails mail your article to , and ]   
   [ ask your news administrator to fix the problems with your system. ]   
      
   --- SoupGate-Win32 v1.05   
    * Origin: you cannot sedate... all the things you hate (1:229/2)