learning2rank
- pointwise: 一个文档一个point,计算得分。可以排序; 回归问题,分类问题都可。
- pairwise: 任何两篇的排序;
- listwise:所有文档的排序损失之和。
lightgbm
LightGBM is a gradient boosting framework that uses tree based learning algorithms.
https://lightgbm.readthedocs.io/en/latest/
基于lightgbm的learning2rank
语料
https://webscope.sandbox.yahoo.com/catalog.php?datatype=c
格式:
0 qid:1 10:0.89028 11:0.75088 12:0.01343 17:0.4484 18:0.90834 21:0.77818 27:0.71505 29:0.77307 30:0.75925 39:0.65244 43
:0.79394 44:0.88406 45:0.87946 66:0.63696 69:0.24961 70:0.41479 71:0.82664 74:0.78587 77:0.70475 83:0.55661 85:0.7212 8
6:0.93081 91:0.34635 98:0.70029 101:0.96287 108:0.19608 122:0.85512 123:0.18873 124:0.46599 127:0.43434 129:0.048097 13
3:0.44747 139:0.92407 145:0.46578 146:0.57682 147:0.63382 149:0.84404 154:0.73041 155:0.50316 159:0.30671 170: