News Mood

Monitor of mood of current headlines, the web frontend is up (and collecting data) on http://hatch01.cs.unc.edu:11235

Sentiment classification is done via 6 one-vs-rest naive Bayes classifiers (one for each emotion)

Data

Originally trained with data from Rada Mihalcea's (EECS @ UMich) Affective Text database.

Currently collecting more data on the live site.

Due to limited data for a multiclass classifier (6 outcomes with ~1250 lines of data), the accuracy is not very high right now.

With 3-fold cross validation the accuracy (as measured by the "correct" emotion being the top 3 returned by the model) is approximately 72%.

Name		Name	Last commit message	Last commit date
Latest commit History 34 Commits
AffectiveText.test		AffectiveText.test
AffectiveText.trial		AffectiveText.trial
neuronet		neuronet
rnn		rnn
templates		templates
test		test
train		train
.gitignore		.gitignore
HeadlineSentiment.py		HeadlineSentiment.py
NaiveBayes.py		NaiveBayes.py
NewsMain.py		NewsMain.py
README.md		README.md
constants.py		constants.py
database.py		database.py
emotion_data.csv		emotion_data.csv
emotion_data_test.csv		emotion_data_test.csv
emotion_data_train.csv		emotion_data_train.csv
get_news.py		get_news.py
headlines		headlines
init.py		init.py
launch_gunicorn.sh		launch_gunicorn.sh
news_server.py		news_server.py
requirements.txt		requirements.txt
to_csv.py		to_csv.py
utils.py		utils.py