A comparative evaluation of off-the-shelf distributed semantic representations for modelling behavioural data.

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Behavioral Science

Background:

• Distributed semantic vectors (word embeddings) are increasingly used to model linguistic meaning.
• Their application in predicting human behavior and data annotations is a growing area of interest.
• A systematic comparison of available resources is needed to guide researchers.

Purpose of the Study:

• To extensively compare off-the-shelf distributed semantic vectors for predicting behavioral outcomes and human data annotations.
• To provide a practical guide on utilizing vector similarity computations for predictive tasks.
• To highlight available datasets and vector representations for this purpose.

Main Methods:

• Comparative analysis of various word embedding models (e.g., Word2Vec, GloVe, FastText).
• Evaluation of prediction accuracy on diverse behavioral and annotation datasets.
• Development of a framework for applying vector similarity measures.

Main Results:

• Demonstration of the efficacy of certain vector representations in predicting human-centric data.
• Identification of best-performing models for specific prediction tasks.
• Quantification of the performance gains achievable through vector similarity computations.

Conclusions:

• Distributed semantic vectors offer a powerful tool for predicting human behavior and annotations.
• The choice of vector representation and similarity metric significantly impacts prediction performance.
• Further research is needed to address current limitations and enhance predictive capabilities.