New Release: Reliable Sentiment Analysis Tool for Social Media Research
26 May 2021, by Jochen Hartmann
Drawing on insights and data sets from a large-scale meta-analysis (>200 papers from the field of computer science), Heitmann et al. (2020) release a state-of-the-art sentiment analysis model for social media researchers working with text data. After less than a month, the model has already been downloaded more than 2,000 times.
The model is freely available for download here and can be applied with only 2 lines of code (no expert knowledge required): https://huggingface.co/siebert/sentiment-roberta-large-english
The model is a fine-tuned checkpoint of RoBERTa-large (Liu et al. 2019). It enables reliable binary sentiment analysis for various types of English-language text. For each instance, it predicts either positive (1) or negative (0) sentiment. The model was fine-tuned and evaluated on 15 data sets from diverse text sources to enhance generalization across different types of texts (reviews, tweets, etc.). Consequently, it outperforms models trained on only one type of text (e.g., movie reviews from the popular SST-2 benchmark).
The model as well as the paper by Heitmann et al. (2020) are an extension of the previous work on automated text classification by Hartmann, Huppertz, Schamp, and Heitmann (2019) from the first phase of the DFG research group. The paper by Hartmann et al. (2019) appeared in International Journal of Research in Marketing and was among the finalists for the 2019 best paper award.