twentiment

.. image:: https://secure.travis-ci.org/passy/twentiment.png?branch=master :target: https://secure.travis-ci.org/passy/twentiment

Research project on twitter sentiment analysis using the Naïve Bayes Classificator.

Installation

Install from PyPI (soon) or github with::

pip install -e git+https://github.com:passy/twentiment.git

Usage

First, start the twentiment server that loads the data from a JSON file. A sample is available in the repository <https://github.com/passy/twentiment/blob/623f4064469850b40b50db4707f12a07047f022b/samples/few_tweets.json>_.

::

twentiment_server samples/few_tweets.json

After that, you can use twentiment_client to query the server using the syntax GUESS my tweet to be scored.

There's a significantly larger samples database available with about two million tweets <http://ge.tt/1fThqCP/v/0>_.

Example

::

twentiment> GUESS hello world
OK 0.0
twentiment> GUESS This car is amazing.
OK 0.5
twentiment> GUESS My best friend is great.
OK 0.9285714285714286
twentiment> GUESS Whatever.
OK 0.0
twentiment> GUESS This car is horrible.
OK -0.5
twentiment> GUESS I am not looking forward to my appointment tomorrow.
OK -0.9852941176470597

Wishlist

(Ranked by importance)

* Give the server an option to fork the server process into the background
  and launch a shell like twentiment_client right away.
* Restructure the Classifier to allow adaptive retraining, i.e. provide a
  TRAIN command that adds new samples at runtime.
    * At the moment, most of the calculations are done at start-up time, so
      querying is rather cheap. Could be difficult to find a good balance.

* Persistence of the server state. Maybe through redis? Only important with
  TRAIN functionality.
* Add some sort of parallelism to the server, so querying doesn't block.
* Add a way of importing live training data from twitter (like from
  analysing emoticons)

Motivation

This is a project report for the Business Intelligence course. To increase the learning potential, I tried to reuse as little as possible from the excellent NLTK <http://nltk.org/>_ project and reimplemented the relevant parts myself.