In　this　paper,　we　propose　a　new　approach　to　predict　users＇　interest　eigenvalues　based　on　multi-Markov　chain　model,　which　provides　a　better　personalized　service　for　the　users　timely.　We　first　collect　a　dataset　from　Sina　Weibo　that　includes　4613　users　and　more　than　16　million　messages;　Then,　preprocess　data　set　to　obtain　users＇　interest　eigenvalues.　After　that,　divide　users　into　several　categories　and　establish　multi-Markov　chain　to　predict　users＇　interest　eigenvalues.　Our　experiments　show　that　using　multi-Markov　model　to　predict　users＇　interest　eigenvalues　is　feasible　and　efficient,　and　could　predicting　both　long-term　and　short-term　user　interests　based　on　a　suitable　selection　of　the　initial　state　distribution,　lambda.