This study proposes methods to automatically extract and recommend popular and relevant URLs from Twitter to support serendipitous learning for software developers. The researchers collected tweets from seed Twitter users and extracted URLs, calculating 14 features for each. URLs were labeled for relevance and a supervised learning-to-rank model and unsupervised Borda count approach were used to recommend URLs. The supervised approach achieved better performance with an NDCG of 0.832. Future work includes automatically categorizing URLs and building a full recommendation system.