Sentiments are fundamental to human communication. However, dealing with sentiment analysis is challenging for researchers. Achieving high accuracy in the extraction of sentiments from data often proves to be difficult and complex. While lexicons can only relate individual words and phrases to sentiment scores, machine learning methods are usually too complex to interpret, yet they promise higher accuracy, i.e., fewer misclassifications. But which method highly affects the accuracy of sentiment classification and how? The new publication by Jochen Hartmann, Mark Heitmann, Christian Siebert, and Christina Schamp in the International Journal of Research Marketing, "More than a Feeling: Accuracy and Application of Sentiment," aims to remedy this problem. They compare the three most applied methods of sentiment analysis, transfer learning, machine learning, and lexicon, by performing a meta-analysis that carefully analyzes the different sentiment methods from over 272 data sets. As a result, the authors provide a pre-trained sentiment analysis model (called SiEBERT) with open source scripts that can be applied as easily as an off-the-shelf lexicon.

You can access the results here.