Sentimentator is a tool for creating fine-grained sentence-level annotations for research in Sentiment Analysis. Annotations are needed for compiling data sets for machine learning tasks.
Getting sentimentator up and running should be straightforward with Docker. In your file system, navigate to the sentimentator directory. On the command line, build the Docker image:
docker build -f Dockerfile -t sentimentator .
After building the image, run that image and expose port 5000. This example assumes your
image is called sentimentator. Run:
docker run -d -p 5000:5000 sentimentator
to run a container, or:
docker run -it --rm -p 5000:5000 -v $(pwd):/app --name sentimentator sentimentator
to run a container and use a shell inside the container. To use the shell, open another command line window and run:
docker exec -it sentimentator <editor_name>
and navigate to localhost:5000 on a Web browser.
Initially the database is empty. You can populate it with demo data by executing the database initialization script in a running Docker container. In a shell run:
python3 data_import.py <lang_name> <lang_filename> <alignment_filename>
<lang_name>: two-letter language code, for example en, fi, or sv.
<lang_filename>: name of the .txt file from which to import each line as a sentence to the database.
<alignment_filename>: name of the .txt file from which to import the OPUS metadata for each sentence. The data_import.py script imports metadata for a pivot language by extracting the sentence id and the path to the OPUS document id for each sentence from the parallel line in the alignment file. Example alignment file:
# en/0/3509138/6031634.xml.gz fi/0/3509138/6003289.xml.gz 5 3
# en/0/3509138/6031634.xml.gz fi/0/3509138/6003289.xml.gz 6 4
# en/0/3509138/6031634.xml.gz fi/0/3509138/6003289.xml.gz 9 6
[...]
For example, to import English, in a shell inside a running container run:
python3 data_import.py en data/en/en.txt data/en/en.ids
OPUS parallel corpus files are available for download at http://opus.nlpl.eu/OpenSubtitles2018.php in the MOSES/GIZA++ format.
The demo dataset used in this project is taken from the OPUS corpus as presented in this paper: P. Lison and J. Tiedemann, 2016, OpenSubtitles2016: Extracting Large Parallel Corpora from Movie and TV Subtitles. In Proceedings of the 10th International Conference on Language Resources and Evaluation (LREC 2016).