QAmp

Svitlana Vakulenko, Javier D. Fernandez, Axel Polleres, Maarten de Rijke and Michael Cochez. Message Passing for Complex Question Answering over Knowledge Graphs. CIKM. 2019

Requirements

• Python 3.6

• tensorflow==1.11.0

• keras==2.2.4

• pyHDT (for accesssing the DBpedia Knowledge Graph)

• elasticsearch==5.5.3 (for indexing entities and predicate labels of the Knowledge Graph)

• pymongo (for storing the LC-QuAD dataset)

• flask (for the API)

Datasets

• LCQUAD 5,000 pairs of questions and SPARQL queries

Setup

It is not trivial to set up the environment. You need to:

1. Create virtual environment and install all dependencies (to install CUDA, TF, Keras and friends follow https://medium.com/@naomi.fridman/install-conda-tensorflow-gpu-and-keras-on-ubuntu-18-04-1b403e740e25)

conda create -n kbqa python=3.6 pip
conda activate kbqa
pip install -r requirements.txt

2. Install HDT API:

git clone https://github.com/webdata/pyHDT.git
cd pyHDT/
./install.sh

3. Download DBPedia 2016-04 English HDT file and its index from http://www.rdfhdt.org/datasets/
4. Follow instructions in https://github.com/svakulenk0/hdt_tutorial to extract the list of entities (dbpedia201604_terms.txt) and predicates
5. Index entities and predicates into ElasticSearch
6. Download LC-QuAD dataset from http://lc-quad.sda.tech
7. Import LC-QuAD dataset into MongoDB

sudo service mongod start

Run

see notebooks

Benchmark

python final_benchmark_results.py

Citation

@inproceedings{DBLP:conf/cikm/VakulenkoGPRC19,
  author    = {Svitlana Vakulenko and
               Javier David Fernandez Garcia and
               Axel Polleres and
               Maarten de Rijke and
               Michael Cochez},
  title     = {Message Passing for Complex Question Answering over Knowledge Graphs},
  booktitle = {Proceedings of the 28th {ACM} International Conference on Information
               and Knowledge Management, {CIKM} 2019, Beijing, China, November 3-7,
               2019},
  pages     = {1431--1440},
  year      = {2019},
  url       = {https://doi.org/10.1145/3357384.3358026},
  doi       = {10.1145/3357384.3358026},
  timestamp = {Mon, 04 Nov 2019 11:09:32 +0100}
}