Sentiment Analysis

sentimentThe Goal: Distinguish Positive and Negative Texts

In Sentiment Analysis, you want to classify a given text as positive or negative. For simple sentences, this is quite straightforward: Even the most trivial algorithms find out that “The weather is beautiful” is positive. But the tasks becomes more challenging (and interesting) for more complex texts, such as “This is a great product! It broke after two days!”.

Alternative Terms: Sentiment Detection, Opinion Mining

Our Contribution

We approach sentiment analysis from two perspective:

1. Develop a new algorithm that allows to “learn” sentiment analysis for arbitrary languages: We use automatically crawled product reviews to train a machine learning algorithm for sentiment analysis. This approach can be applied to any language where a sufficient amount of product reviews is available.

2. Analyze and improve accuracy of existing sentiment analysis tools: We evaluated 10 commercial sentiment analysis tools and saw that their average accuracy is about 60%. Using machine learning algorithms, we could combine these tools to a meta-classifier that outperformed each single tool.

Achievements

We participated in SemEval 2014, an international competition on semantic analysis. We submitted two systems for Subtask 9B, which targeted sentiment analysis on tweets. Our systems reached rank 8 and 12, respectively, out of 50 submissions.

Research Team

Contact

Do you want to know the “best” sentiment analysis tool for your purpose? Do you have an interesting idea how to apply sentiment detection in your business? Then please feel free to contact us!

Email:   ciel <insert @ here> zhaw.ch
Phone:  +41 58 934 72 39

Publications

  1. JOINT_FORCES: Unite Competing Sentiment Classifiers with Random Forest. Oliver Dürr, Fatih Uzdilli and Mark Cieliebak. SemEval 2014.
  2. Swiss-Chocolate: Sentiment Detection using Sparse SVMs and Part-Of-Speech n-Grams. Martin Jaggi, Fatih Uzdilli, and Mark Cieliebak. SemEval 2014.
  3. Meta-Classifiers Easily Improve Commercial Sentiment Detection Tools. Mark Cieliebak, Oliver Dürr, and Fatih Uzdilli. LREC 2014.
  4. Potential and Limitations of Commercial Sentiment Detection Tools. By Mark Cieliebak, Oliver Dürr and Fatih Uzdilli. ESSEM 2013.
    See Section “Data” for the data we used.
  5. Social Media: Wissen, was Kunden wollen. By Mark Cieliebak and Daniel Krebser. In: Erfolgsfaktor Emotionalisierung, by Brian Rüeger and Frank Hannich (eds), 2010.

Data

Evaluation of Commercial Sentiment Detection Tools