Can computational sentiment analysis classify autobiographical memories? Comparing VADER and TextBlob

Written narratives can provide insight into the mental and affective state of a person. However, current methods of analysis require extensive manual human coding, limiting the sample size that can feasibly be analyzed. Computational approaches using natural language processing offer promise as a rapid means of examining characteristics of narrative texts. Though used extensively in social media and customer reviews in industry, their classification accuracy relative to human coders remains unclear for psychological data. We compared the congruence of two popular lexicon-based sentiment analysis tools, Valence Aware Dictionary and sEntiment Reasoner (VADER) and TextBlob, with self-reported valence, for 3,309 narratives representing autobiographical memories from two different data sets. Both tools, but especially TextBlob, displayed a bias in rating memories as neutral in valence. Across both data sets, VADER consistently produced more congruent classifications than did TextBlob. Accuracy improved significantly, to moderate levels of congruence (r_MCC = .32), on par with that of human coders reported in other work, when classification was limited to three valence types (negative, neutral, or positive) rather than five. Tool performance remained consistent regardless of data set size. VADER, more so than TextBlob, is a useful means of rapidly determining the broad emotional valence within large volumes of text data that would be unfeasible for manual coding. (PsycInfo Database Record (c) 2026 APA, all rights reserved).