An application for sentiment analysis.
The application runs on Linux and Windows with Python 3 and TensorFlow installed. You can use make init to install TensorFlow with pip.
To run the application use runner.py. It requires a file with word embeddings, one word embedding per line. It can be generated with an external tool (e.g. using fastText) or downloaded separately (e.g. see GloVe).
Please run runner.py --help for full list of options.
Training set should contain sentences, one per line, with either 1 or 0 in the end, e.g.:
This is an example of a positive sentence. 1
This is an example of a negative sentence. 0
Example:
$ ./runner.py -r embeddings.vec -t training_set.txt
The model can be run to evaluate sentences in a file or interactively:
$ ./runner.py -r embeddings.vec -s sentences.txt
$ ./runner.py -r embeddings.vec -i
The application can also be run as a simple HTTP server. flask is required for the server to work. Clients are able to manage models, train, and run them. The server is processing one request at a time, so that long-running operations like training should not be run frequently.
It's also possible to load a few models at the same time, so running a model doesn't block for too long. The cache size is configurable, please refer to server.py --help for details.
Example:
$ ./server.py -r embeddings.vec -c 5