Twitter users express their opinions in different domains such as politics, products, movie reviews, and sports. Opinions are expressed about particular topics, entities or subjects of a certain domain. Analysing the sentiment underlying these opinions has important applications in product marketing and other areas. However, existing research work on sentiment analysis does not consider that expressions of opinion are domain-dependent and can also evolve over time. We propose to investigate algorithms for domain-focused sentiment analysis in time-evolving social media streams to improve accuracy.