The document proposes a method called TwiNews to detect local user interests in Twitter to re-rank news search results. It uses LDA topic modeling on geotagged tweets to determine the local topical distribution, and then calculates cosine similarity between news articles and local topics to re-rank news for a given query based on local user interests. An A/B test with turkers from California and New York showed the localized ranking improved relevance by 3-7% over the baseline.