搜索引擎indri系列：评价 (Evaluation)

本文使用trec_eval。

注意编译生成可执行文件trec_eval才可以!!! linux系统下，进入trec_eval的目录，控制台输入 make即可编译。

命令行格式如下：

$ ./trec_eval -h
trec_eval [-h] [-q] [-m measure[.params] [-c] [-n] [-l ]
   [-D debug_level] [-N ] [-M ] [-R rel_format] [-T results_format]
   rel_info_file  results_file 

其中-m 可以有四种选择：

'official': the main measures often used by TREC
'all_trec': all measures calculated with the standard TREC results and rel_info format files.
'set': subset of all_trec that calculates unranked values.
'prefs': Measures not in all_trec that calculate preference measures.

我们选择 all_trec，它会对计算所有measure，结果如下：

user$ ./trec_eval -m all_trec path-to-qrels/qrels.251-300   path-to-retrieval-result/input.anu5aut1

runid                   all anu5aut1
num_q                   all 50
num_ret                 all 50000
num_rel                 all 5524
num_rel_ret             all 1879
map                     all 0.1537
gm_map                  all 0.0237
Rprec                   all 0.1948
bpref                   all 0.1691
recip_rank              all 0.4216
iprec_at_recall_0.00    all 0.4679
iprec_at_recall_0.10    all 0.2869
iprec_at_recall_0.20    all 0.2399
iprec_at_recall_0.30    all 0.2027
iprec_at_recall_0.40    all 0.1760
iprec_at_recall_0.50    all 0.1557
iprec_at_recall_0.60    all 0.1191
iprec_at_recall_0.70    all 0.0954
iprec_at_recall_0.80    all 0.0703
iprec_at_recall_0.90    all 0.0390
iprec_at_recall_1.00    all 0.0316
P_5                     all 0.2520
P_10                    all 0.2540
P_15                    all 0.2240
P_20                    all 0.2160
P_30                    all 0.1893
P_100                   all 0.1384
P_200                   all 0.0929
P_500                   all 0.0593
P_1000                  all 0.0376
recall_5                all 0.0695
recall_10               all 0.1029
recall_15               all 0.1160
recall_20               all 0.1483
recall_30               all 0.1653
recall_100              all 0.2590
recall_200              all 0.3060
recall_500              all 0.3944
recall_1000             all 0.4560
infAP                   all 0.1537
gm_bpref                all 0.0144
Rprec_mult_0.20         all 0.2690
Rprec_mult_0.40         all 0.2369
Rprec_mult_0.60         all 0.2343
Rprec_mult_0.80         all 0.2154
Rprec_mult_1.00         all 0.1948
Rprec_mult_1.20         all 0.1720
Rprec_mult_1.40         all 0.1537
Rprec_mult_1.60         all 0.1413
Rprec_mult_1.80         all 0.1305
Rprec_mult_2.00         all 0.1239
utility                 all -924.8400
11pt_avg                all 0.1713
binG                    all 0.1399
G                       all 0.1399
ndcg                    all 0.3419
ndcg_rel                all 0.3171
Rndcg                   all 0.2778
ndcg_cut_5              all 0.2783
ndcg_cut_10             all 0.2860
ndcg_cut_15             all 0.2736
ndcg_cut_20             all 0.2763
ndcg_cut_30             all 0.2631
ndcg_cut_100            all 0.2650
ndcg_cut_200            all 0.2730
ndcg_cut_500            all 0.3125
ndcg_cut_1000           all 0.3419
map_cut_5               all 0.0619
map_cut_10              all 0.0793
map_cut_15              all 0.0862
map_cut_20              all 0.0940
map_cut_30              all 0.1005
map_cut_100             all 0.1291
map_cut_200             all 0.1389
map_cut_500             all 0.1494
map_cut_1000            all 0.1537
relative_P_5            all 0.2620
relative_P_10           all 0.2863
relative_P_15           all 0.2705
relative_P_20           all 0.2895
relative_P_30           all 0.2738
relative_P_100          all 0.2961
relative_P_200          all 0.3168
relative_P_500          all 0.3951
relative_P_1000         all 0.4560
success_1               all 0.3200
success_5               all 0.5400
success_10              all 0.6000
set_P                   all 0.0376
set_relative_P          all 0.4560
set_recall              all 0.4560
set_map                 all 0.0199
set_F                   all 0.0627
num_nonrel_judged_ret   all 16729