how to slove this problem: the number of negative instance is far more than positive instance

[yuye2133]

In a rank task, one query maybe have many documents, if <query, document_(i)> label 0 or 1, the number of 0 will be far more than 1, and then the trainning model cannot fit test set very well.

[zhangzibin]

@yuye2133 Actually your problem is label imbalance. You should try some strategies to deal with that.

[yuye2133]

但是这类Rank问题不是都存在这种不平衡问题吗？请问你是怎么处理这种不平衡的啊？

[gallupliu]

同求，理解应该是模型对问题打分 然后根据打分排序

[zhangzibin]

@yuye2133 @gallupliu I use resampling usually. The best strategy depends on the problem.