UUUSA

A Java implementation of the formalism described in the article "Universal, Unsupervised (rule-based), Uncovered Sentiment Analysis"

Prerequisites

Java 8

Maven (if you are compiling from source code)

You also will need MaltParser and the Stanford tagger if you plan to train a parser or a tagger to plug it into the system.

Build the JAR

Building from source coude

unzip uuusa-master.zip
cd uuusa-master
mvn assembly:assembly

If everything goes fine, you should see something like at the end of the log:

[INFO] ------------------------------------------------------------------------
[INFO] BUILD SUCCESS
[INFO] ------------------------------------------------------------------------
[INFO] Total time: 7.779 s
[INFO] Finished at: 2017-08-30T13:28:24+02:00
[INFO] Final Memory: 58M/1028M
[INFO] ------------------------------------------------------------------------

This will create a folder /target inside uuusa-master. Inside you should see two .jar files:

samulan-0.1.1.jar
samulan-0.1.1-jar-with-dependencies.jar

Take the file samulan-0.1.1-jar-with-dependencies.jar to execute uuusa as a standalone application.

Pre-built versions

We provide some versions of UUUSA, already as JAR's, so you do not need to build them:

samulan 0.1.0.jar This version is the system as used in Vilares et al. (2017a). Related data and resources used can be also found in our local repository: http://grupolys.org/software/UUUSA/samulan-0.1.0.jar

Data

Data/Resources used for our UUUSA model can be found here

Data/Resources used for our SISA (Syntactic Iberian Polarity classification) model can be found here

Obtaining a trained tagger or parser

We only provided a small set of pretrained taggers and parsers. You might want to use your own. To do so, you must consider some things:

To train a tagger:

Samulan supports models trained using the Stanford-tagger

Locate the trained model (.tagger) inside your PATH_SENTIDATA/

To train a parser:

Samulan supports parsers trained using Maltparser-1.7.1.

Locate the trained model (.mco) and the features xml inside the PATH_SENTIDATA/maltparser/

Execute the JAR file

Over a raw file

java -Dfile.encoding=UTF-8 -jar -Xmx2g USA_JAR -s EN PATH_SENTIDATA -r PATH_OPERATIONS_XML -i PATH_RAW_TEXT -p PATH_UUUSA_PROPERTIES_FILE -v [true|false]

The input must be formatted as a tsv file, where the last column contains the text to be analized.

Over a CoNLL file

java -Dfile.encoding=UTF-8 -jar -Xmx2g USA_JAR -s EN PATH_SENTIDATA -r PATH_OPERATIONS_XML -c PATH_PARSED_CONLL -p PATH_UUUSA_PROPERTIES_FILE -v [true|false]

Execution options

-i	Path to the raw file. Cannot be used together with -c and viceversa.
-c	Path to a CoNLL file containing the parsed files. You must specify an identifier above the first conll 	 graph of each text (### IDENTIFIER\n"). Check http://grupolys.org/software/UUUSA/en_parsed.conll for an example
														 
-s	Path to the Sentidata directory. Check http://grupolys.org/software/UUUSA/EN-SentiData/ for an example.
-e	Encoding. Default utf-8
-r	Path to the .xml file containing the rules
-o	Path to the output file with the predictions
-v	VERBOSE. true|false
-sc	Selects the type of classification. trinary|binary|so
-p	Path to the properties file
-spf	Path to the file where the parsed sentences in CoNLL format will be saved. Useful if you plan to run many experiments. Added in version 0.1.1.

References

Vilares, D., Gómez-Rodríguez, C., & Alonso, M. A. (2017a). Universal, unsupervised (rule-based), uncovered sentiment analysis. Knowledge-Based Systems, 118, 45-55.

@article{vilares2017universal,
  title={Universal, unsupervised (rule-based), uncovered sentiment analysis},
  author={Vilares, David and G{\'o}mez-Rodr{\'\i}guez, Carlos and Alonso, Miguel A},
  journal={Knowledge-Based Systems},
  volume={118},
  pages={45--55},
  year={2017},
  publisher={Elsevier}
}

If you use SISA, please also cite:

Vilares, D., Garcia, M., Alonso, M. A., & Gómez-Rodríguez, C. (2017b). Towards Syntactic Iberian Polarity Classification. 8th Workshop on Computational Approaches to Subjectivity, Sentiment & Social Media Analysis (WASSA 2017), Copenhagen, Denmark, 2017 (to appear)