Relevance Feedback methods generally suffer from topic drift caused by word ambiguities and synonymous uses of words. As a way to alleviate the inherent problem, we propose a novel query phrase expansion approach utilizing semantic annotations in Wikipedia pages, trying to enrich queries with context disambiguating phrases. The idea was implemented for patent search where patents are classified into a hierarchy of categories, and the analyses of the experimental results showed not only the positive roles of phrases and words in retrieving additional relevant documents through query expansion but also their contributions to alleviating the query drift problem. More specifically, our query expansion method was compared against Relevance Model, a state-of-the-art, to show its superiority in terms of MAP on all levels of the classification hierarchy. Furthermore, we investigate the relationship between QE methods with respect to topic drift concept. More precisely, how QE methods behave causing topic drift. To investigate that, we propose a method of showing drifting topics, among other topic classes within queries, and their effect on the retrieval effectiveness. We further compare our work against Relevance Model to show the amount of drift generated by each expansion method.