Without the need for the third party and key sharing, the dummy-based privacy protection scheme enables the users to obtain precise query results while protecting their location privacy. However, when the adversary has certain background knowledge, e.g., the spatiotemporal reachability information, the location semantics, the users' historic query statistics, the probability of dummies being inferred will rise and the degree of privacy protection will be reduced. To solve this problem, a personalized dummy generation method based on spatiotemporal correlations and location semantics is proposed. Dummies are first generated based on the continuous reachability with previous request locations, and then filtered through the check of location semantic similarity and finally filtered by accessibility to user's historic query statistics. Experiments based on real datasets show that the proposed dummy generation method can effectively reduce the risk of privacy disclosure compared with current two dummy generation methods, especially when the adversary has related background knowledge.