Social media revolution has encouraged millions of users to share their thoughts about various brands and provided services and products. Therefore, social media is a valuable source for companies to discover consumers? feedback about items. However, discovering the most critical feedback from vast volumes of short and informal texts, i.e., microblogs, is highly challenging. This thesis investigates employing deep learning and Natural Language Processing (NLP) techniques on microblog to discover users? latent feedback. Therefore, we proposed a two-stages approach that uses users? tweets related to specific items. The first phase of our approach involves classifying tweets based on their expressed sentiment using deep learning models. After that, we apply text summarization techniques to extract concise and informative summaries representing the main topics of positive and negative tweets. Once the summary is generated for each sentiment orientation, we need to measure how informative it is. This is achieved by automatically evaluating the generated summary against the original tweets using Latent Sematic Analysis (LSA). Overall, the results of our experiments indicated that the proposed approach has merit in extracting informative feedback about items from large-volume and grammatically-unstructured microblogs. The contribution of this thesis lies in presenting an approach to discover consumers? latent feedback from tweets by applying sentiment analysis and text summarization.