Track: Machine Learning
Different organizations have started to provide automation in evaluating applicants during their interviews. Due to globalization, bilingualism has become essential in assessing individuals. Past studies have focused on analyzing the features and developing models for evaluating applicants who spoke only one language – English – during their interviews. Hence, this paper aims to analyze the lexical and prosodic features of interviews spoken using both English and Tagalog in evaluating interview performances based on three criteria: Leadership Ability, Communication Skills, and Candidate Enthusiasm. After consulting with domain experts, 60 mock interview recordings were gathered. The researchers used Google Cloud Services API for transcription and translation, PRAAT to extract prosodic features, and LIWC to extract lexical features. The researchers gathered different algorithms from past studies and selected the most suitable ones based on the characteristics of the datasets. From the results of model training, the researchers concluded that the optimal algorithms were LDA for Leadership Ability with 94% accuracy and 95% F1-score, SVM for Communication Skills with 88.89% accuracy and 82% F1-score, and Random Forest for Candidate Enthusiasm with 87.5% accuracy and 89.2% F1-score. Furthermore, the researchers concluded that words categorized under Psychological Processes were the most prevalent for lexical features, while features under Frequencies and Formants were the most prevalent for prosodic features
Job Interviews, Bilingualism, Natural Language Processing, Multi-modal Behaviors, and Frequencies and Formants.