Various techniques have been used over the years to implement recommendation systems. In this research, we have analyzed several papers and majority of them have used collaborative and content-based filtering techniques to implement recommender system. To build a recommender system, we require abundant amount of data which comprises of a spectrum of reviews and sentiments from all user domains. Websites like Yelp and TripAdvisor, allow users to post reviews for various businesses, products and services. In this work we have two objectives 1) Recommend restaurants to user based on user reviews in Yelp dataset and 2) Suggest improvements to business based on user reviews. In the first scenario, we will use the comments and ratings available in the Yelp dataset to generate restaurant recommendations and personalize them with user profile data. In the second scenario, we intend to suggest improvements to businesses based on various user reviews and provide them with a ranked list of predefined parameters to help them understand where they stand with respect to their competitors and where they should improve to do better. For both scenarios, we will perform two major steps to achieve our objective 1) Sentiment Analysis and 2) Content Based Recommendation. The first step gives us the - sentiment, quality, subject of discussion relevant to product and in the second step we use the outcomes of first step for personalizing and ranking our results. We came across Gensim and Latent Dirichlet Allocation which seemed to be interesting and was tailored to our requirements. In the yelp dataset, user comments are a mixture of various topics which are extracted by the algorithm (LDA) to provide accurate recommendation for all the users. A prototype of this method provided us with 93% accuracy.