Human-Like Entity Linking using Contextual knowledge
flasking.py starts a server on localhost:5000. Our current instance is running on http://flask.fii800.lod.labs.vu.nl. This accepts only POST requests with textual input as data. All additional parameters (weights, turning knowledge types ON/OFF) can be supplied as parameters in the query string. See flasking.py for supported parameters.
Our code is written in python 3. Apart from python dependencies, we use several more things needed to run this system properly:
- Pickle JSON with temporal views for december 2007. The keys of this JSON are wikipedia/dbpedia entities, while the values are integers representing their views in this month.
- A neo4j instance that contains all wikilinks, extracted using the code and following the instructions here: https://github.com/erabug/wikigraph
- A running Redis instance where we cache: the wikilinks seen once, the LOTUS calls, and additional computationally heavy information.
- (optional) Stanford CoreNLP instance - we decided to turn off this for our experiments due to efficiency and accuracy considerations. If you need it, you can use our instance at corenlp.fii800.lod.labs.vu.nl
For any considerations, suggestions, questions, or setup troubles, contact Filip Ilievski ([email protected]).