Most word segmentation methods employed in Chinese Information Retrieval systems are based on a static dictionary or a model trained against a manually segmented corpus. These general segmentation approaches may not be optimal because they disregard information within semantic units. We propose a novel method for improving word-based Chinese IR, which performs segmentation according to the tightness of phrases. In order to evaluate the effectiveness of our method, we employ a new test

more »

... of 203 queries, which include a broad distribution of phrases with different tightness values. The results of our experiments indicate that our method improves IR performance as compared with a general word segmentation approach. The experiments also demonstrate the need for the development of better evaluation corpora.