Recommender system (RS) analyzes the purchase behavior of existing users and predicts relevant item(s) to a new user. In RS, collaborative filtering (CF) is a popular approach to suggest item(s) that are most similar to the new user’s interests using item/user similarity. Generally, in CF-RS top rated items are recommended. However, in this approach the low and average rated items are neglected which may be liked by the user. As a result, the CF-RS approach is unable to improve the user satisfaction. In this paper, we propose an approach called Serendipitous Recommender System (SRS) to recommend the items which are liked by the users; however, the items need not be top-rated. Generally, the user ratings may not express the true opinions of the users. We observe that the users express their opinions in user-reviews through the emotional words and they may contain surprise emotion. These emotional words are considered to update the user-rating such that the items will be available for recommendations. We use the updated user-ratings for the final recommendations using the user-based and item-based CF approaches. We call such recommended items as serendipity items. This allows us to provide the recommendations that are nearer to the users’ intent. We conducted experiments on real-world datasets, Amazon and Yelp. We evaluated the proposed approach using precision, recall, F1-Score, and unexpectedness metrics. The results show that the proposed approach performed better in recommending surprise items.