Using Social Media and Personality Predictions to Anticipate Startup Success

Abstract: This thesis explores the potential of integrating predicted founder personalities, based on the Big 5 Personality Framework, into Machine Learning (ML) models to enhance the accuracy of early-stage startup success predictions. Leveraging Natural Language Processing (NLP) techniques, we extracted personality insights from founders' tweets, focusing on US startups funded between 2013 and 2015. Our research utilized a range of models, including XGBoost, Random Forest, and Feed-forward Neural Network for personality predictions, and Logistic Regression, XGBoost, and Random Forest for startup success forecasts. Results indicated that most personality-predicting models outperformed the Naive baseline. In success predictions, XGBoost emerged as the top performer, showcasing the highest scores in Macro F1 and AUC for both Series B and Series C funding rounds. While the trait of Neuroticism was highlighted as significant for Series B predictions across models, Series C predictions emphasized the importance of Openness and Agreeableness. Our findings underline the value of integrating predicted personality traits into ML models for startup success forecasts. However, as with all research, our work had inherent limitations and suggested areas for further exploration and improvement.