A new sequential classification to assist Ad auction agent in making decisions

One of the critical abilities of intelligent agents is making quick, accurate and wise decision within dynamic environments in a reasonable period of time. This paper introduces a new sequence classification method based on positive and negative sequential patterns. The historical log data from previous performance “TAC/Ad auction” has been used for the classification. We have applied Nconf: an “interestingness measure” to prune meaningless negative and positive candidates. Our experiments showed that agents equipped with this classifier can earn higher profit compared to non equipped ones. Our findings lead us to believe that with some additional tune during the run we may achieve more qualified agents.