Data for "Towards an Anatomy of Search Engine Component Performances" - TREC 07-08-09-10
We selected a set of alternative implementations of each component and by using the Terrier open source system we created a run for each system defined by combining the available components in all possible ways.
We considered three main components of an IR system: stop list, Lexical Unit Generator (LUG) and IR model. We selected a set of alternative implementations of each component and by using the Terrier open source system we created a run for each system defined by combining the available components in all possible ways. The components we selected are:
- stop list: nostop, indri, lucene, smart, terrier
- LUG: nolug, weak Porter, Porter, snowball Porter, Krovetz, Lovins, 4grams, 5grams, 6grams, 7grams, 8grams, 9grams, 10grams;
- model: the vector space model (i.e., TFIDF and LemurTFIDF), the probabilistic model – comprehending the BM25 models and the Divergence From Randomness (DFR) models (i.e., BB2, DFIZ, DFRee, DLH, DPH, IFB2, InB2, InL2, InexpB2 and PL2) – and the language models (i.e., DirichletLM, HiemstraLM, Js KLs and LGD).
Content of the directories:
- code: the Matlab code for running the analyses and reproducing the experiments from the Grid of Points contained in the
datadirectory. It requires the MATTERS library. - data: the Grid of Points for the following Adhoc collections: TREC 7 (
T07directory), TREC 8 (T08directory), TREC 9 (T09directory), TREC 10 (T10directory). Each directory contains a.matfile for each of the following evaluation measures: AP, P@10, RPREC, nDCG, nDCG@20, ERR, ERR@20, Twist and RBP. Files have to be opened with theserloadcommand of the MATTERS library.
Ferro, N. and Silvello, G. (2017). Towards an Anatomy of Search Engine Component Performances. Journal of the Association for Information Science and Technology. http://onlinelibrary.wiley.com/doi/10.1002/asi.23910/full.