Despite all success of recommender systems, these systems face natural problems such as cold start and data sparseness. To face these issues, in this work a novel hybrid recommender is proposed that leverages the strengths of both collaborative and content-based methods together with sentiment analysis. The proposed system improves recommendation accuracy, alleviates cold start and sparsity issues, predicts future interactions, and adds a new context layer to recommendations, by effectively integrating demographic data, user clustering, Machine Learning (ML) models, sentiment analysis, and improved link prediction strategy. Proposed system was evaluated using the popular MovieLens dataset, and test results indicate the system significantly improves handling new users. The proposed system achieved a decline of 5.2% in Mean Absolute Error (MAE) and 6.8% in Root Mean Square Error (RMSE), compared to baseline methods. Moreover, results indicate that the proposed recommender performs significantly better under user and item cold start conditions, with a reduction of 29.6% and 27.4% in MAE, respectively. Lastly, integrating sentiment analysis and an improved link prediction approach provides a boost of 7% in Precision and 10.8% in Recall. Therefore, such hybrid approaches can be successful in alleviating cold start and sparsity problems, and enhancing overall system performance.