Schedule of Events | Search Abstracts | Invited Symposia | Symposia | Poster Sessions | Data Blitz

Dictionary choice matters when selecting tools for sentiment analysis in younger versus older adult populations

Poster Session E - Monday, March 9, 2026, 2:30 – 4:30 pm PDT, Fairview/Kitsilano Ballroom

Khalil Husein¹ (), Claudia Morales Valiente², Myra Fernandes¹; ¹University of Waterloo, ²University of Alberta

People’s writings can provide insight into the mental and affective state of a person. Computational sentiment analysis tools have shown promise in rapidly predicting emotional valence in human narratives. However, these tools have largely been trained and tested on narratives produced by younger adults. It is unclear whether their performance remains consistent in narratives produced by people across the lifespan. Here, younger and older adults were presented with clips from popular songs as memory cues. They wrote short descriptions of personal memories elicited, and self-rated their emotional valence. Classifications from five sentiment dictionaries (General Inquirer, Harvard IV-4, QDAP, TextBlob, VADER) were compared using conditional inference trees trained separately for each age group. VADER, which was originally designed for informal language used in social media, offered better performance in the narrative style used by young adults. In contrast, TextBlob’s use of grammatical context and lexical relations allowed it to better evaluate sentiment from narratives with more formal vocabulary and syntax used by older adults. The divergent performance of sentiment dictionaries across younger and older adults reflects fundamental differences in both the tools themselves and the language used in autobiographical narration across the lifespan. Findings demonstrate that dictionary choice is not neutral, and that analytic strategies must align with the evolving complexities of language across the lifespan.

Topic Area: METHODS: Other