Evaluation of Different Word Embeddings to Create Personality Models in Spanish
| Autor: | Felipe Orlando López-Pabón, Juan Rafael Orozco-Arroyave |
|---|---|
| Rok vydání: | 2021 |
| Předmět: |
business.industry
Psychological research media_common.quotation_subject computer.software_genre Focus (linguistics) Job performance Personality Artificial intelligence Big Five personality traits Personality Assessment Inventory business Psychology computer Reliability (statistics) Natural language processing Word (computer architecture) media_common |
| Zdroj: | Communications in Computer and Information Science ISBN: 9783030867010 WEA |
| Popis: | Research in psychology has shown that personality directly influences the way people think, feel and communicate. It also has consequences on behavior and indirectly affects work effectiveness and job performance. Automatic personality assessment has gained attention in the last decade, and one of the most common models in psychology for automatic personality analysis is the Big Five model, also called as OCEAN model. Different works that study personality traits are based on English texts; conversely, very few studies focus on creating Spanish models. This paper proposes a methodology for the automatic modeling of personality in Spanish texts. Transliterations of videos from YouTube are translated to Spanish to create and evaluate the models. Classical word embeddings are considered, including Wor2Vec, GloVe, BERT, and BETO. Classification and regression experiments are performed to predict the labels of the five traits in the OCEAN model. The results show that 3 out of the five traits can be predicted with high reliability. Additionally, embeddings created with transformer-based models (i.e., BERT and BETO) yield the highest accuracies. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|