divyag9 / acl2017 Goto Github PK

We propose a simple yet effective text- based user geolocation model based on a neural network with one hidden layer, which achieves state of the art performance over three Twitter benchmark geolocation datasets, in addition to producing word and phrase embeddings in the hidden layer that we show to be useful for detecting dialectal terms. As part of our analysis of dialectal terms, we release DAREDS, a dataset for evaluating dialect term detection methods.

DAREDS is a dataset consisting of dialect words and their dialect region extracted from Dictionary of American Regional English (DARE) available online at http://www.daredictionary.com. It is available at the data directory.

We experiment with three Twitter geolocation datasets available at https://github.com/utcompling/textgrounder or alternatively https://www.amazon.com/clouddrive/share/kfl0TTPDkXuFqTZ17WJSnhXT0q6fGkTlOTOLZ9VVPNu or in preprocessed pickle from from https://www.dropbox.com/sh/vofn0awjcjxhwbc/AABHekl2pmFk2Q3qdVO60JTKa?dl=0.

usage: geodare.py [-h] [-dataset str] [-model str] [-datadir DATADIR] [-tune]

optional arguments:
  -h, --help            show this help message and exit
  -dataset str, --dataset str
                        dataset name (cmu, na, world)
  -model str, --model str
                        dialectology model (mlp, lr, word2vec)
  -datadir DATADIR      directory where input datasets (cmu.pkl, na.pkl,
                        world.pkl) are located.
  -tune                 if true, tune the hyper-parameters.

The preprocessed pickle files (e.g. na.pkl) are the vectorized version of the geolocation datasets and are available upon request.

@InProceedings{rahimi2017a,
  author    = {Rahimi, Afshin  and  Cohn, Trevor  and  Baldwin, Timothy},
  title     = {A Neural Model for User Geolocation and Lexical Dialectology},
  booktitle = {Proceedings of ACL-2017 (short papers) preprint},
  year      = {2017},
  address   = {Vancouver, Canada},
  publisher = {Association for Computational Linguistics}
}

Afshin Rahimi [email protected]