Recommender systems have gained significant interest across various fields, including e-commerce and research, in today's fast-expanding information landscape. However, it is becoming more and more difficult to generate reliable and pertinent recommendations for consumers due to the sparsity and cold-start problems. This work attempts to address the sparsity and cold-start problems by merging the model-based collaborative filtering and content-based method. The proposed model considers ratings database along with movie plot information to strengthen the performance of the approach. For model-based approach, the autoencoder technique is utilized on extremely sparse rating matrix to extract the pertinent information of rating database. On the other hand, content-based filtering is incorporated on movie plot information by using the TF-IDF vectorizer to extract significant terms from movie descriptions and subsequently applying cosine similarity for similarity computations among the movies to generate a list of recommendations. The experiments are carried out on the MovieLens1M dataset and accuracy of the proposed methodology is compared with state-of-the-art techniques by analysing Mean Absolute Error (MAE), Mean Squared Error (MSE) and Root Mean Square Error (RMSE). The accuracy of proposed approach has improved by a factor of 20.42%, 13.93% and 20.13% in terms of MAE, MSE and RMSE respectively.