NRC-Canada: Sentiment Analysis System for Tweets
(Contact: Saif M. Mohammad, email: firstname.lastname@example.org)
The NRC-Canada system (Kiritchenko et al. 2014) ranked first in three sentiment shared tasks: SemEval-2013 Task 2 (Mohammad, Kiritchenko, and Zhu, 2013), SemEval-2014 Task 9 (Zhu, Kiritchenko, and Mohammad, 2014), and SemEval- 2014 Task 4 (Kiritchenko, Zhu, and Mohammad, 2014). Many of the same features used in NRC-Canada were also used in a stance-detection system that outperformed submissions from all 19 teams that participated in SemEval-2016 Task 6 (Mohammad et al., 2017).
SENTIMENT ANALYSIS OF TWEETS:
NRC-Canada: Building the State-of-the-Art in Sentiment Analysis of Tweets, Saif M. Mohammad, Svetlana Kiritchenko, and Xiaodan Zhu, In Proceedings of the seventh international workshop on Semantic Evaluation Exercises (SemEval-2013), June 2013, Atlanta, USA.
Paper (pdf) BibTeX System Description and Downloads Poster Slides
Sentiment Analysis of Short Informal Texts. Svetlana Kiritchenko, Xiaodan Zhu and Saif Mohammad. Journal of Artificial Intelligence Research, volume 50, pages 723-762, August 2014.
Paper (pdf) BibTeX
NRC-Canada-2014: Recent Improvements in Sentiment Analysis of Tweets, Xiaodan Zhu, Svetlana Kiritchenko, and Saif M. Mohammad. In Proceedings of the eighth international workshop on Semantic Evaluation Exercises (SemEval-2014), August 2014, Dublin, Ireland.
Paper (pdf) BibTeX
Official Rankings: Our team (NRC-Canada) ranked first in five of the ten subtask-domain combinations. About 40 teams participated.
ASPECT-BASED SENTIMENT ANALYSIS (ABSA):
NRC-Canada-2014: Detecting Aspects and Sentiment in Customer Reviews, Svetlana Kiritchenko, Xiaodan Zhu, Colin Cherry, and Saif M. Mohammad. In Proceedings of the eighth international workshop on Semantic Evaluation Exercises (SemEval-2014), August 2014, Dublin, Ireland.
Paper (pdf) BibTeX Poster
Official Rankings: Our team (NRC-Canada) ranked first in three of the six subtasks. About 30 teams participated.
Semeval-2016 Task 6: Detecting Stance in Tweets. Saif M. Mohammad, Svetlana Kiritchenko, Parinaz Sobhani, Xiaodan Zhu, and Colin Cherry. In Proceedings of the International Workshop on Semantic Evaluation (SemEval ’16). June 2016. San Diego, California.
Paper (pdf) BibTeX Presentation Task Website
Stance and Sentiment in Tweets. Saif M. Mohammad, Parinaz Sobhani, and Svetlana Kiritchenko. Special Section of the ACM Transactions on Internet Technology on Argumentation in Social Media, 2017, 17(3).
Paper (pdf) BibTeX Data and Visualization
The NRC-Canada system code used is copyright of the Crown, and hence not currently available for download. However, we encourage the use of software packages that implement parts of the NRC-Canada system such as those listed below:
Direct download and use of code available:
- The AffectiveTweets Package: Felipe Bravo-Marquez implemented the AffectiveTweets for the Weka machine learning workbench that provides a collection of filters for extracting state-of-the-art features from tweets for sentiment classification/regression and other related tasks. The package is especially useful to generate feature vectors from a large number of affect lexicons. The vector can then be concatenated to other features vectors (say dense-distributed representations of the text) to improve perfomance.
- Webis: Software used in a SemEval-2015 shared task.
Papers that mention re-implementing the NRC-Canada system (code may be obtained by requesing the authors directly):
If you know of other re-implementations, please let us know (email: email@example.com).
A large number of manually created and automatically generated lexicons were utlized. They are available here. You may also be interested in the The AffectiveTweets Package to generate feature vectors from a large number of affect lexicons.
- all-caps: the number of words with all characters in upper case;
- clusters: presence/absence of tokens from each of the 1000 clusters (provided by Carnegie Mellon University's Twitter NLP tool);
- elongated words: the number of words with one character repeated more than 2 times, e.g. 'soooo';
- presence/absence of positive and negative emoticons at any position in the tweet;
- whether the last token is a positive or negative emoticon;
- hashtags: the number of hashtags;
- negation: the number of negated contexts. A negated context also affects the ngram and lexicon features: each word and associated with it polarity in a negated context become negated (e.g., 'not perfect' becomes 'not perfect_NEG', 'POLARITY_positive' becomes 'POLARITY_positive_NEG');
- POS: the number of occurrences for each part-of-speech tag;
- the number of contiguous sequences of exclamation marks, question marks, and both exclamation and question marks;
- whether the last token contains exclamation or question mark;
- sentiment lexicons: automatically created lexicons (NRC Hashtag Sentiment Lexicon, Sentiment140 Lexicon), manually created sentiment lexicons (NRC Emotion Lexicon, MPQA, Bing Liu Lexicon). For each lexicon and each polarity we calculated:
- total count of tokens in the tweet with score greater than 0;
- the sum of the scores for all tokens in the tweet;
- the maximal score;
- the non-zero score of the last token in the tweet;
The lexicon features were created for all tokens in the tweet, for each part-of-speech tag, for hashtags, and for all-caps tokens.
- word ngrams, character ngrams.
- word embeddings (these were used in the stance system)
Updated: June, 2017