The project contains a combined query set and extended assessment files from the Yahoo! SemSearch Challenge 2010/11 evaluation campaigns. The relevance labels have been acquired after an evaluation campaign on Amazon Mechanical Turk followed by a thorough quality control analysis (see details in the coming paper).
There have been made a few changes comparing to the initial YSC files:
- the both query sets from YSC 2010 and YSC 2011 campaigns are combined into a single query set
- the query file is formatted as follows: query_num\tquery, e.g. 1\t44 magnum hunting (\t means the tab symbol)
- misspellings in 9 queries were fixed using a feature "Search instead for" of Google Search
- the relevance labels are standardized according to the three-grade scale: 0 means irrelevance, 1 - fair relevance, 2 - excellent relevance
- 11% initial labels are fixed, the total number of labels has increased by 17%.
We recommend using the following trec_eval commands to compute some standard IR evaluation measures:
- NDCG: trec_eval -q -c -m ndcg.0=0,1=1,2=3 assess.txt your_results.txt
- MAP: trec_eval -q -c -l1 -m map assess.txt your_results.txt
- P@10: trec_eval -q -c -l1 -m P.10 assess.txt your_results.txt
If you have found these data helpful and used them in your research work, please cite the following paper:
Zhiltsov, N., Agichtein, E. Improving Entity Search over Linked Data by Modeling Latent Semantics. Proceedings of the International Conference on Information and Knowledge Management (CIKM 2013). ACM, 2013.