A Lexical and Sentiment Analysis of TikTok Hooks and Their Influence on Digital Engagement

This study investigates the relationship between the linguistic and sentimental characteristics of TikTok hooks, which are defined as the brief opening seconds of short-form video content, and subsequent digital engagement levels. The research aims to determine whether the frequency and contextual use of specific lexical features, as well as the emotional tone present in hooks, are associated with key interaction metrics on TikTok, such as views, likes, and comments. To achieve this objective, a dataset of 4983 hooks was compiled by scraping publicly available TikTok content published between January 2020 and December 2022. Audio-to-text transcription facilitated the extraction of linguistic information, which was subsequently analysed via three sentiment analysis methods: Sentimentr, AFINN, and Syuzhet. These methods were employed to capture contextual polarity, affective intensity, and emotional variation. Statistical analyses included Spearman’s correlation tests to assess the relationship between lexical frequencies and engagement metrics, as well as Wilcoxon’s nonparametric tests to compare sentiment profiles between hooks with above- and below-median engagement levels. The findings indicate no statistically significant correlation between the selected lexical features and any of the interaction metrics, suggesting that simple word frequency is not a reliable predictor of user engagement. However, significant differences were observed in sentiment profiles, particularly for the number of views, implying that emotionally charged hooks may be more effective at capturing initial audience attention and may benefit from early algorithmic amplification. These results are consistent with theoretical perspectives from the attention economy, which emphasize the role of emotional salience in shaping user behaviour during the initial moments of content exposure. Several methodological limitations are acknowledged, including potential inaccuracies in audio-to-text transcription, the exclusion of multimodal features such as visual and audio cues, and reliance on lexicon-based sentiment tools that may not fully capture nuanced emotional expression. Future research should incorporate multimodal analysis, semantic modelling, and experimental designs to more comprehensively examine how linguistic, visual, auditory, and algorithmic factors collectively influence engagement patterns on short-form video platforms.