Techniques and applications for sentiment analysis

2 minute read

Feldman, R. (2013). Techniques and applications for sentiment analysis. Communications of the ACM, 56(4), 82—89. doi:10.1145/2436256.2436274

The article provides an overview over the following five problems in the field of sentiment analysis:

  • document level sentiment analysis,
  • sentence level sentiment analysis,
  • aspect-based sentiment analysis,
  • comparative sentiment analysis, and
  • sentiment lexicon acquisition

Selected methods

Document level sentiment analysis either uses supervised or unsupervised approaches to determine the sentiment. Another interesting method determines the PMI of phrases to positive (excellent) and negative (poor) words to compute the phrase˜s polartiy.

Sentence level sentiment analysis often further split the sentences into homogeneous phrases, if that is required. Research also suggest that it is beneficial to handle special kind of sentences (e.g. conditional sentences, sarcasm, ...) differently [1]. Such methods must be combined with approaches for detecting these sentence classes, e.g. Davidov et al. [2] for sarcams.

Aspect-based sentiment analysis which is also known as feature-based sentiment analysis considers that products have different aspects (attributes, properties) which in turn may receive different opinions. The extraction of implicit aspects (e.g. žit is quite heavy rather than žthe battery life is extensive) is still an interesting research topic.

Comparative sentiment analysis draws upon sentences that compare sentiment targets. Jundal and Liu [3] have shown that a small number of words such as the ones listed below covers up to 98% of all comparisons.

  • Comparative: more, less, ¦ word endings with -er (lighter, smaller, ¦)
  • Superlative: most, last, ¦ word endings with -est (lightest, finest, ¦)
  • Additional phrases: favor, exceed, outperform, prefer, than, superior, inferior, number one, up against, ...)
Sentiment lexicon acquisition is another interesting research area that uses methods such as WordNet distance to the terms good and bad or double propagation to discover sentiment terms.

Selected applications

  1. Google product search
  2. - how does a word feels on Twitter
  4. - analyzes tweets that contain stock symbols


[1] R. Narayanan, B. Liu, and A. Choudhary, Sentiment analysis of conditional sentences, in Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing: Volume 1 - Volume 1, Stroudsburg, PA, USA, 2009, pp. 180—189.

[2] D. Davidov, O. Tsur, and A. Rappoport, Semi-supervised recognition of sarcastic sentences in Twitter and Amazon, in Proceedings of the Fourteenth Conference on Computational Natural Language Learning, Stroudsburg, PA, USA, 2010, pp. 107—116.

[3] N. Jindal and B. Liu, Identifying comparative sentences in text documents, in Proceedings of the 29th annual international ACM SIGIR conference on Research and development in information retrieval, New York, NY, USA, 2006, pp. 244—251.