
Supplementary Material README
=============================
This bz2 archive contains the supplementary for:

Ebert, Sebastian; Schütze, Hinrich (2014):
Fine-Grained Contextual Predictions for Hard Sentiment Words
In: Proceedings of the 2014 Conference on Empirical Methods in Natural Language
    Processing. EMNLP 2014

This archive is available at: http://www.cis.uni-muenchen.de/ebert/

Copyright (c) 2014 Sebastian Ebert
All rights reserved.



Data Description
----------------
contexts.* contain all contexts split into training (trn), development 
(dev1), and test (tst) set. All these files contain one context per line 
(column 1) together with that context's sense (column 2) from the paper and 
(training set only) the number of the cluster the context has been assigned to
(column 3). The cluster id corresponds to the cluster in centroid_sense.csv.

NEGATIVE-P.txt and NEUTRAL-P.txt contain the negative and neutral phrases found 
in the data.

centroid_sense.csv contains all 100 cluster centroids (column 2) and the sense
assigned to the clusters (column 1). This file version contains only the first
5 cluster centroids due to file size restrictions. A full version of the file
is available in the supplementary material archive of the above mentioned
website.

Columns are separated by <TAB>, whereas values within one column are separated 
by <SPACE>.
