mponza / Wikipediarelatedness

Branch for an immediate application of the Two-Stage Framework over every kind of graph. Typing only two commands.

Setting Up

In your working directory, just type:

git clone https://github.com/mponza/WikipediaRelatedness.git
cd WikipediaRelatedness
wget https://piccolo.link/sbt-0.13.17.zip; unzip sbt-0.13.17.zip; rm sbt-0.13.17.zip

for downloading this repository and sbt.

Indexing

First, you need to index and pre-process several resources for running the Two-Stage Framework. This can be automatically done with:

src/main/bash/build.sh path/to/graph.tsv path/to/two-stage-data

where graph.tsv is the graph in tsv format (or tsv.gz) and two-stage-data is the directory that will host all resources that will be indexed for running the Two-Stage Framework.

Running

You can compute the Two-Stage Framework relatedness over a set of query nodes by simply typing:

src/main/bash/query.sh k path/to/two-stage-data path/to/queries.tsv path/to/queries2rel.tsv

where k is the size of the subgraph (the value used in the paper was fixed to 30), two-stage-data is the same directory provided in the Indexing step and queries.tsv are the list of query nodes in tsv format of which the relatedness needs to be computed and saved in queries2rel.tsv.

Examples

For a toy-example on a very small graph please check src/main/bash/example.sh.

For using the Two-Stage Framework in your own code just check Main class for examples.

Datasets of Entity Relatedness Pairs

You can find the datasets WikiSim and WiRe in src/main/resources/datasets/WikiSim.csv and src/main/resources/datasets/WiRe.csv files, respectively.

Citation and Further Reading

If you find any resource (code or data) of this repository useful, please cite our paper:

Marco Ponza, Paolo Ferragina, Soumen Chakrabarti
A Two-Stage Framework for Computing Entity Relatedness in Wikipedia
In Proceedings of 26th International Conference on Information & Knowledge Management (CIKM 2017)

License

The code in this repository has been released under Apache License 2.0.