Latent Semantic Indexing (LSI)

So far, most of the publicly available Page/Image/Video search engines (e.g., Google, Yahoo, MSN) adopts query-list paradigm, that is, when a user submits a query, the search engine returns a ranking list of pages/images/videos. The higher on the list, the more relevant to the query the page/image/video is supposed to be. This approach works efficiently for well-defined narrow queries, when the query is too general or even ambiguous, e.g., “jaguar”, “apple”, the users may have to sift through the list to locate the interest. For example, for query “apple”, pages/images/videos of apple fruit are intermixed with pages/images/videos of apple ipod, apple logo, apple pie, and so on