Test_NOTE

CIKM competition

TASK:  Design a language-independent algorithm to automatically detect the intent of unlabeled queries in the dataset.

Target categories : “VIDEO”, “NOVEL”, “GAME”, “TRAVEL”, “LOTTERY”, “ZIPCODE”, and “OTHER”.

Dataset ：

     dataset D1:   970,000 queries that are of length more than 2 bytes and occur at least 10 times a day.

     dataset D2:   D1 can label to severl queries, yet D2 limit each cell can be labeled only up to 2 categories, and filter out those bigger than 2. Reduced down to about 7W queries.

     dataset D3: Collect “click” and “session” data from user logs.

       dataset D4:  code the chinese characters into digit code.

     dataset D5:  50% data to train, and with some miss values, write the results to D5.