Evaluating Bert and GPT-2 Models for Personalised Linkedin Post Recommendation
Prerna Singh, Bhawna Bhutoria Jain, Kirti Sinha · 2023
Social networking platforms have become essential tools for individuals and organisations to connect, communicate, and collaborate in today’s digital age. LinkedIn is a professional social networking platform facilitating career development and job searching. LinkedIn’s traditional post recommendation system limits users to seeing only those posts that have been engaged upon by their following or first-degree connections, limiting the user’s perspective. The proposed system leverages the content posted by the user to provide personalised content delivery on LinkedIn, potentially enhancing user engagement and satisfaction. This research study proposes and examines the performance of three content-based recommender models developed with Machine Learning (ML), Generative Pre-Trained Transformer (GPT-2), and Bidirectional Encoder Representations from Transformers (BERT). In terms of capturing the similarity between user-generated and recommended posts, BERT outperformed the other models, achieving the highest similarity score of 97.13%, compared to GPT-2 (96.27%) and basic ML (95.69%).