Thumbs Up or Thumbs Down? Semantic Orientation Applied to Unsupervised Classification of Reviews
By Peter D. Turney. 2002
5,849 citations recorded by the source corpus.
This paper presents a simple unsupervised learning algorithm for classifying reviews as recommended (thumbs up) or not recommended (thumbs down). The classification of a review is predicted by the average semantic orientation of the phrases in the review that contain adjectives or adverbs. A phrase has a positive semantic orientation when it has good associations (e.g., "subtle nuances") and a negative semantic orientation when it has bad associations (e.g., "very cavalier"). In this paper, the semantic orientation of a phrase is calculated as the mutual information between the given phrase and the word "excellent" minus the mutual information between the given phrase and the word "poor". A review is classified as recommended if the average semantic orientation of its phrases is positive. The algorithm achieves an average accuracy of 74% when evaluated on 410 reviews from Epinions, sampled from four different domains (reviews of automobiles, banks, movies, and travel destinations). The accuracy ranges from 84% for automobile reviews to 66% for movie reviews.
Subfields: General Health Professions, Philosophy
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In General Health Professions
- Michael A. Karpowicz
- A. Hassan
- Juuso Kalevi Kristian Eronen
- José Camacho-Collados
- Jason Baldridge
- Shai Ben-David
- Simon Corston-Oliver
- Amjad Abu-Jbara
In Philosophy
- Michael A. Karpowicz
- A. Hassan
- Juuso Kalevi Kristian Eronen
- José Camacho-Collados
- Jason Baldridge
- Cristian Danescu-Niculescu-Mizil
- Shai Ben-David
- Simon Corston-Oliver
Citations in the corpus
- SHINE: Signed Heterogeneous Information Network Embedding for Sentiment Link Prediction (2017, background): “where PMI is the point-wise mutual information [25] defined as PMI (x ,y) = log p(x,y) p(x )p(y) , pos and neд are the tweets of positive and”
- Recognizing Contextual Polarity in Phrase-Level Sentiment Analysis (2005, background): “Much work on sentiment analysis classifies documents by their overall sentiment, for example determining whether a review is positive or negative (e.g., (Turney, 2002; Dave et al., 2003; Pang and Lee, 2004; Beineke et al., 2004)).”
- Learning Extraction Patterns for Subjective Expressions (2003, methodology): “…words (e.g., Levin’s desire verbs (1993)), and some empirical methods in NLP have automatically identified adjectives, verbs, and N-grams that are statistically associated with subjective language (e.g., (Turney, 2002; Hatzi-vassiloglou and McKeown, 1997; Wiebe, 2000; Wiebe et al., 2001)).”
- Learning Attitudes and Attributes from Multi-aspect Reviews (2012, background): “Reviews consisting of plain-text feedback and a single numeric score have proved a valuable source of data in many applications, including product recommendation [2], feature discovery [27], review summarization [16], and sentiment analysis [33], among others.”
- Unsupervised named-entity extraction from the Web: An experimental study (2005, methodology): “One of K NOW I T A LL ’s main contributions is adapting Turney’s PMI-IR algorithm [42, 43, 44] to serve as validation for information extraction.”
- Empath: Understanding Topic Signals in Large-Scale Text (2016, background): “For example, researchers have learned sentiment models from the relationships between words [14], classified the polarity of reviews in an unsupervised fashion [43], discovered patterns of narrative in text [4], and (more recently) used neural networks to model word meanings in a vector space [23, 2”
- Mining and summarizing customer reviews (2004, methodology): “Turney’s work in [ 42 ] applies a specific unsupervised learning technique based on the mutual information between document phrases and the words “excellent” and “poor”, where the mutual information is computed using statistics gathered by a search engine.”
- Biographies, Bollywood, Boom-boxes and Blenders: Domain Adaptation for Sentiment Classification (2007, background): “While we do not have a direct comparison, we note that Turney (2002) performs worse on movie reviews than on his other datasets, the same type of data as the polarity dataset.”
- A machine learning approach to sentiment analysis in multilingual Web texts (2009, methodology): “Turney (2002) searches the texts on the World Wide Web to determine the semantic orientation of adjectives and nouns by using the “close to” operator, comparing the number of hits of the word as a close neighbor of respectively “excellent” or “poor”.”
- SENTIMENT CLASSIFICATION of MOVIE REVIEWS USING CONTEXTUAL VALENCE SHIFTERS (2006, methodology): “Results ranged from 84% for automobile reviews to as low as 66% for movie reviews (Turney 2002).”
- Sentiment analysis in Facebook and its application to e-learning (2014, methodology): “The results obtained in those works served as the basis for classifying entire documents, considering that the average semantic orientation of the words in a review may be an indicator of whether the text is positive or negative (Turney, 2002).”
- Pulse: Mining Customer Opinions from Free Text (2005, methodology): “We have begun experimenting with a variation of a technique for bootstrapping from seed words with known orientation [ 1 ,2] with promising initial results [15].”
- Opinion spam and analysis (2008, methodology): “Sentiment classification techniques [20, 23 ] may be used to automatically assign a rating to a review solely based on its review content.”