Currently, we provide a basic image to experiment with Army ANT 0.3. This image comes prepackaged with two indexes for a small subset of the INEX 2009 Wikipedia Collection. This subset, INEX 2009 3T-NL covers 3 topics (3T) and all the documents mentioned in the relevance judgments for these topics, excluding linked files (NL; i.e., no links). The instance includes the following engines, ranking functions and configurable parameters:
- Lucene
- TF-IDF
- BM25
- k1 = [1, 1.2, 1.8]
- b = [0.5, 0.75, 1]
- Divergence From Randomness (DFR)
- BasicModel: BM = [BE, G, P, D, In, Ine, IF]
- AfterEffect: AE = [L, B, Disabled]
- Normalization: N = [H1, H2, H3, Z, Disabled]
- Hypergraph-of-Entity
- Random Walk Score
- Length: l = [1, 2, 3, 4, 5, 6]
- Repeats: r = [10, 100, 250, 500, 750, 1000]
- Jaccard Score
- Random Walk Score
In order to experiment with the evaluation module, you will first need to download the INEX 2009 3T-NL subset, attached to release 0.3, thus obtaining the topics/2010-topics.xml
, with the test queries, and the assessments/2010/inex2010.qrels
with the relevance judgments.
In order to install Army ANT, please follow the instructions in army-ant-install. Running Army ANT will require a MongoDB docker instance in the same network on the default port.