What's TLSA?

TLSA is a free and open web service that offers different tools of sentiment analysis, including transformer- and lexicon-based methods. TLSA helps you to conduct valid sentiment analysis at ease 😃. No programming skill is required. No specialized computer hardware (e.g., GPU) is needed.

Why TLSA?

Research shows that the performance of sentiment analysis varies to a large extent and depend on the specific context (Boukes et al., 2020; van Atteveldt et al., 2021). Thus, it's important to compare different tools 🤔 and select a validated tool for your research project.

Citation

If you use TLSA, please cite our paper 🤗:

• Zhao, X., & Wong, C.-W. (2023). Automated measures of sentiment via transformer- and lexicon-based sentiment analysis (TLSA). Journal of Computational Social Science. https://doi.org/10.1007/s42001-023-00233-8
• Zhao, X., & Wong, C.-W. (2023). Transformer- and lexicon-based sentiment analysis as a web service. In 73rd Annual Conference of International Communication Association (ICA'23), Computational Methods Division. May 25-29, 2023. Toronto, Canada.

BibTex:

@article{zhao2023transformer1,
  author    = {Xinyan Zhao and Chau-Wai Wong},
  title     = {Automated measures of sentiment via transformer- and lexicon-based sentiment analysis ({TLSA})},
  journal   = {Journal of Computational Social Science},
  month     = {Nov.},
  year      = {2023},
  doi       = {10.1007/s42001-023-00233-8}
}

@inproceedings{zhao2023transformer2,
  author    = {Xinyan Zhao and Chau-Wai Wong},
  title     = {Transformer- and lexicon-based sentiment analysis as a web service},
  booktitle = {73rd Annual Conference of International Communication Association (ICA'23)},
  year      = {2023},
  month     = {May},
  address   = {Toronto, Canada},
}

Best Practices

1. Include about 50 pairs of texts and reference labels for stable prediction performance. Use at least 2 human coders and follow standard procedures in content analysis to obtain the reference labels (Negative, Neutral, Positive). For detail, refer to our paper.

2. Use a CSV file with two columns: "text" and "label". A user must fill in the “text” column with data to be analyzed. For validation, user should include the reference labels generated in the first step in the “label” column. Please download our template to format your CSV file.

3. After you upload your CSV file, you will receive a ZIP file with two outputs: one with all predicted sentiment labels and scores and the other one with prediction performance results. Scholars can inspect the results and adopt a valid and reliable tool for sentiment analysis for all data. If none of these tools are satisfactory, scholars can conduct manual content analysis or work with computer scientists to further fine-tune the BERT-based models to achieve better prediction performance.

Disclaimer

There is no conflict of interest. TLSA is licensed under Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International (CC BY-NC-SA 4.0).