ETHZ_IR_2015_Proj2_G9 Assignment Report
Creative Commons CC BY 4.0
We develop 3 term-based models(Naive tf, log tf, and BM25), unigram language model, and Pointwise Online AdaGrad approach to select top 100 documents of each query. We use MIN((TP+FN),100) as denominator when calculating AP. We also perform some methods in preprocessing and running stage to obtain better MAP, as well less running time of the whole program. 2 rounds of scanning are needed in our system.