Sentiment Analysis of Short Informal Texts

By Svetlana Kiritchenko, Xiao-Dan Zhu, Saif M. Mohammad. Journal of Artificial Intelligence Research, 2014

896 citations recorded by the source corpus.

We describe a state-of-the-art sentiment analysis system that detects (a) the sentiment of short informal textual messages such as tweets and SMS (message-level task) and (b) the sentiment of a word or a phrase within a message (term-level task). The system is based on a supervised statistical text classification approach leveraging a variety of surface-form, semantic, and sentiment features. The sentiment features are primarily derived from novel high-coverage tweet-specific sentiment lexicons. These lexicons are automatically generated from tweets with sentiment-word hashtags and from tweets with emoticons. To adequately capture the sentiment of words in negated contexts, a separate sentiment lexicon is generated for negated words. The system ranked first in the SemEval-2013 shared task 'Sentiment Analysis in Twitter' (Task 2), obtaining an F-score of 69.02 in the message-level task and 88.93 in the term-level task. Post-competition improvements boost the performance to an F-score of 70.45 (message-level task) and 89.50 (term-level task). The system also obtains state-of-the-art performance on two additional datasets: the SemEval-2013 SMS test set and a corpus of movie review excerpts. The ablation experiments demonstrate that the use of the automatically generated lexicons results in performance gains of up to 6.5 absolute percentage points.

Subfields: Artificial Intelligence, Information Systems

View Paper: https://www.semanticscholar.org/paper/e2ae833ec0bf4a0435d7feea94a48c027e06a0e1

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Artificial Intelligence

In Information Systems

Citations in the corpus

Stance and Sentiment in Tweets (2016, methodology): “10 Positive and negative language tend to have sufficient amount of commonality regardless of topic of discussion, and hence sentiment analysis systems traditionally learn a single model from all of the training data [Liu 2015; Kiritchenko et al. 2014a; Rosenthal et al. 2015].”
Revisiting Word Embedding for Contrasting Meaning (2015, background): “, sentiment contrast (Kiritchenko et al., 2014b; Zhu et al., 2014a), in a dramatic and complicated manner, which has been considered in learning sentiment contrast (Kiritchenko et al.”
Hierarchical Attention Networks for Document Classification (2016, background): “Text Features are constructed according to (Kiritchenko et al., 2014), including word and character n-grams, sentiment lexicon features etc.”
Collaborative Multi-domain Sentiment Classification (2015, methodology): “The sentiment lexicon used to extract prior general sentiment knowledge for Amazon dataset is SentiWordNet3 [1], and the sentiment lexicon used for Sanders dataset is the Sentiment140 lexicon4 [11], which is a Twitter-specific sentiment lexicon.”
Sentiment Analysis of Review Datasets Using Naive Bayes and K-NN Classifier (2016, background): “Information Engineering and Electronic Business, 2016, 4, 54-62 Analysis is of three different types: Document level, Sentence level and Entity level [11].”
How Translation Alters Sentiment (2016, methodology): “In the process of developing these experiments to study how translation impacts sentiment, we created a new dialectal Arabic sentiment lexicon and a state-of-the-art Arabic sentiment analysis system by porting NRC-Canada’s competition winning system (Mohammad, Kiritchenko, & Zhu, 2013; Kiritchenko, ”
NRC-Canada-2014: Detecting Aspects and Sentiment in Customer Reviews (2014, methodology): “We built two lexicons, Yelp Restaurant Sentiment AffLex and Yelp Restaurant Sentiment NegLex, as described in (Kiritchenko et al., 2014).”
Learning Semantic Representations of Users and Products for Document Level Sentiment Classification (2015, background): “, 2010) and sentiment lexicon features (Kiritchenko et al., 2014).”
A Multilayer Perceptron based Ensemble Technique for Fine-grained Financial Sentiment Analysis (2017, background): “Sentiment, Sentiment140, Sentiment140 Context) lexicons (Kiritchenko et al., 2014; Mohammad et al., 2013) which associate a positive or negative score to a token.”
Sentiment Embeddings with Applications to Sentiment Analysis (2016, methodology): “applying them as features for Twitter sentiment classification in a state-of-the-art supervised learning pipeline [64], [67].”
A Practical Guide to Sentiment Annotation: Challenges and Solutions (2016, methodology): “Methods for obtaining real-valued sentiment scores are also not covered in this paper; we refer the reader to Kiritchenko and Mohammad (2016a) and Kiritchenko et al. (2014) for the use of best–worst scaling to obtain reliable real-valued sentiment associations.”
Multitask Learning for Fine-Grained Twitter Sentiment Analysis (2017, methodology): “In tasks such as sentiment classiﬁcation, additional features like membership of words in sentiment lexicons or counts of elongated/capitalized words can be used to enrich the representation of tweets before the clas-siﬁcation step (Kiritchenko et al., 2014).”
Sentiment Analysis: An Overview from Linguistics (2016, neutral): “Most recent applications have involved forms of blogging and microblogging, such as Twitter or Facebook messages (Kiritchenko et al 2014, Mohammad et al 2013, Ortigosa et al 2014, Thelwall et al to appear, Vilares et al 2015), including the Hedonometer project, an attempt to measure happiness in…”
Like It or Not (2016, methodology): “An et al. [2014] used the chi-squared measure, which is a common statistical test for feature selection, whereas Kiritchenko et al. [2014] used a Pointwise Mutual Information (PMI) measure to identify terms that reveal sentiment.”
DataStories at SemEval-2017 Task 4: Deep LSTM with Attention for Message-level and Topic-based Sentiment Analysis (2017, methodology): “In previous runs of the Task, sentiment analysis was usually tackled using hand-crafted features and/or sentiment lexicons (Mohammad et al., 2013; Kiritchenko et al., 2014; Palogiannidi et al., 2016), feeding them to classiﬁers such as Naive Bayes or Support Vector Machines (SVM).”
SemEval-2018 Task 1: Affect in Tweets (2018, methodology): “Observe that several of the NRC emotion and sentiment lexi-cons as well as AFINN and Bing Liu Lexicon were widely used (Mohammad and Turney, 2013; Mo-hammad, 2018b; Kiritchenko et al., 2014; Nielsen, 2011; Hu and Liu, 2004).”