Beschreibung
Abstract: Despite growing evidence that relying solely on the positive or negative tone in text is not sufficient to understand consumer behavior, the use of sentiment in marketing research & practice remains widespread. Various ways to extract more fine-grained emotions from text exist, but they all have their drawbacks: While human raters are often too resource-intensive, lexical approaches face challenges regarding incomplete vocabulary and the handling of informal language. Although recent machine learning-based approaches partly address these issues, these models often require expert knowledge, programming skills, human-annotated training data, and extensive computing resources. To overcome these issues, we propose a machine learning approach that utilizes emojis in an intermediate step and extracts more nuanced emotions from text without the need for human-annotated training data: We first predict a vast array of emojis based on the surrounding text, and then reduce the predicted emojis to an established set of eight basic emotions. Using human raters as the ground truth, we benchmark our approach against state-of-the-art affect detection methods and demonstrate its superior performance.| Zeitraum | 12 Jan. 2023 |
|---|---|
| Ereignistitel | European Quant Marketing Seminar |
| Veranstaltungstyp | Konferenz |