This paper describes a novel keyword extraction algorithm Position Weight (PW) that utilizes linguistic features to represent the importance of the word position in a document. Topical terms and their previous-term and next-term co-occurrence collections are extracted. To measure the degree of correlation between a topical term and its co-occurrence terms, three methods are employed including Term Frequency Inverse Term Frequency (TFITF), Position Weight Inverse Position Weight (PWIPW), and CHI-Square (?2). The co-occurrence terms that have the highest degree of correlation and exceed a co-occurrence frequency threshold are combined together with the original topical term to form a final keyword. With the linear computational complexity of the algorithm, the vector space of documents in a large corpus or boundless web can be quickly represented by sets of keywords, which makes it possible to retrieve large-scale information fast and effectively.