Emotion detection for written texts: techniques, their limitations and general challenges

Tijdens mijn premaster schreef ik een paper over het herkennen van emoties in geschreven teksten. Daarbij ging ik in op wat de huidige technieken zijn, maar ook wat hierin de beperkingen zijn. Hieronder zie je het abstract.

Emotion detection is a branch of artificial intelligence and focuses on identifying specific emotions such as happiness, sadness, or anger. This paper researches techniques for detecting emotions in written texts by studying literature from 2020 onwards. We zoom in on these different techniques, how they work and what their limitations are. We discuss the lexicon based approach, the rule based approach, machine learning, deep learning, transfer learning, transformer models, hybrid approaches, and the Multi-label Emotion Detection Architecture (MEDA). Apart from discussing limitations of each technique specifically, we discuss general challenges faced across all techniques. We especially look into challenges caused by a lack of datasets, both quantitatively and qualitatively. We also discuss challenges with detecting implicit emotions and sarcasm, as well as looking at the detection of multiple emotions for a single input.

Meer informatie

Tijdens het schrijven van de paper was de opdracht om dit in twee stappen te doen. Tijdens de eerste stap was het de bedoeling om een paper te schrijven van 4-5 pagina's. Tijdens de tweede stap moesten we ontvangen feedback verwerken en de paper inkorten naar drie pagina's.

Downloads

Enkele bestanden die je kan downloaden.

Paper "Emotion detection for written texts: techniques, their limitations and general challenges" (Draft, 4-5 pagina's)

Paper "Emotion detection for written texts: techniques, their limitations and general challenges" (Final version, 3 pagina's)