trappmartin / PWMD_ICTIR2017

Position-Sensitive Word Mover's Distance

Position-Sensitive Word Mover’s Distance

Code accompanying the paper Retrieving Compositional Documents Using Position-Sensitive Word Mover’s Distance.

Twin Films Dataset

The twin dataset described in the publication is found under:

dataset

where the extracted IMDB mapping are found in

dataset/imdb-mappings.csv

the ground truth mappings of twin films described on wikipedia are listed in

dataset/twin-films.csv

and the extracted plot keywords are found in binary format (python pickle) under

dataset/plotKeywords.pkl

.

Example Code

An example implementation of the Position-Sensitive Word Mover’s Distance can be found under:

code/pwmd.py

To run the code, please ensure to install the following python packages in advance.

python 3
spacy
pyemd

Citation

Please use the following citation:

@inproceedings{Trapp2017b,
    title={Retrieving Compositional Documents Using Position-Sensitive Word Mover’s Distance},
    author={Trapp, Martin and Skowron, Marcin and Schabus, Dietmar},
    booktitle={Proceedings of the ACM International Conference on the Theory of Information Retrieval},
    year={2017},
    doi={10.1145/3121050.3121084}
   }

License

The code is licensed under the MIT License.