Tweet recommendation using Clustered Bert and Word2vec Models

This work presents a tweet recommendation algorithm based on Clustered Bert and Clustered word2vec models. The paper proposes a hybrid recommender system using content and collaborative filtering techniques. In the first phase, the content of the tweets is captured and processed via clustered Bert and word2vec models. These models work on clustering the tweet embeddings generated from Bert and Word2vec models giving out tweet clusters as an output. In the second phase, Collaborative filtering algorithms is then used to recommend tweets to a user based on the historic tweet cluster of interest of the user. The work uses Agglomerative and Kmeans clustering algorithms to cluster similar tweets together. Results show that the recommendations generated from Clustered Bert model have a better performance than Clustered Word2vec model.