transforEmotion: Sentiment Analysis for Text, Image and Video Using Transformers

transforEmotion: Sentiment Analysis for Text, Image and Video Using Transformer Models

Aleksandar Tomašević, University of Novi Sad

In this talk, I will introduce the transforEmotion R package for sentiment analysis and emotion detection using transformer models. The package provides a convenient way to use a number of text and image classification transformer models with zero-shot learning abilities in R. Users can (1) specify the task they want to perform (sentiment analysis, facial expression recognition in images or videos, RAG), (2) select a model from Hugging Face for the specified task, and (3) receive the output of the model as a dataframe in R.
The key component of the package is the seamless management of the Python virtual environment in the backend and the execution of Python scripts using the reticulate package. This enables users proficient in R, but not in Python, to perform the specified tasks and gather model outputs in a convenient format, without directly handling Python distribution and package management on their system. In the final part of the talk, I will outline the vision for future updates of the package and discuss several new use-cases that might be served by future versions of the package, both in academic and industry research.

About the lecturer

Aleksandar Tomašević is an Assistant Professor at the University of Novi Sad, where he obtained his PhD in 2019. His research is at the intersection of computational social science, network science, and applied statistics. His latest work deals with emotion detection in videos, and while working at the University of Virginia, he helped develop the transforEmotion R package. He has published his work in leading data science and psychology journals.

Links:
https://www.linkedin.com/in/atomasevic/
https://twitter.com/atomasevic
https://github.com/atomashevic/transforEmotion