Current　search　engines　have　two　problems,　losing　useful　information　and　including　useless　information.　These　two　problems　are　aroused　by　the　keyword　matching　retrieval　model,　which　is　adopted　by　almost　all　search　engines.　We　introduce　the　conception　of　category　attribute　of　a　word.　According　to　the　category　attribute　of　a　word,　the　useless　results　can　he　removed　from　the　search　results　and　the　retrieval　efficiency　will　he　improved.　A　latent　semantic　analysis　based　method　of　getting　the　category　attribute　of　the　word　is　presented　in　this　paper,　which　is　proved　to　be　effective　by　experiment.　Latent　semantic　analysis　is　a　method　that　can　discover　the　underlying　semantic　relation　between　words　and　documents.　Singular　value　decomposition　is　used　in　latent　semantic　analysis　to　analyze　the　words　and　documents　and　get　the　semantic　relation　finally.