This benchmark continues our previous work P. Huynh, P. Papotti, 2017 on fact checking by providing the first comprehensive and publicly available infrastructure for evaluating fact checking methods across a wide range of assumption about the facts and the reference information.
It is an additional material of the publication "A Benchmark for Fact Checking Algorithms Built on Knowledge Bases", from P. Huynh, P. Papotti, accepted to International Conference on Information and Knowledge Management (CIKM), 2019
-
The benchmark server is written by Nodejs. Follow
package.jsonto install dependencies. -
For the Knowlegde graph, we employ the DBPedia repository from
KGMiner [2], including enties, predicates and facts in the form of triples.- Download: https://www.dropbox.com/s/yqlom6t5ehze0uq/data.zip?dl=1
- Extract the zip file and place the files
node_dict.tsv,edge_dict.tsvin folder../data/dbpedia/graphs/and fileedges.tsvin folder../data/dbpedia/graphs/graph_chi/- node_dict.tsv: contains entities and corresponding indices. Ex: Paris 12345, France 25672
- edge_dict.tsv: contains predicates. Ex: capital 100
- edges.tsv: contains fact triples in the form of (s, o, p). Ex: 12345 25672 100 represents for (Paris, France, capital)
- To load the graph server: please navigate to KG-Miner folder and run ./run_server.sh
-
Follow base folders for downloading and installing benchmarking systems:
Knowledge Linker (KL) [1],Discriminative Predicate Path (KGMiner) [2],Subgraph Feature Extraction (SFE) [3],Parallel Graph Embedding (Para_GraphE) [4],Rule Discover [5]Each system has some modifications from the original version, according to the experiments considered in this work.