Recommender　systems　can　be　used　to　provide　users　with　personalized　recommendation　information.　They　always　rely　on　collaborating　filtering　(CF),　since　it　is　an　effective　way　to　establish　connections　between　products　and　users.　Most　of　the　CF　methods　are　based　on　neighborhood　models,　which　calculate　the　similarities　between　users　and　products.　Moreover,　some　improvements　that　model　neighborhood　relations　by　minimizing　a　cost　function　are　made　to　predict　a　better　result,　since　latent　factor　models　can　provide　more　aspects　of　the　data,　and　offer　more　accurate　results　than　neighborhood　models.　Past　models　were　limited　by　using　simple　cosine　similarity　only,　and　they　did　not　consider　that　the　change　of　interests.　Moreover,　age　groups　may　have　a　significant　effect　on　the　final　results.　In　this　paper,　to　solve　the　problem,　a　new　comprehensive　item　similarity　based　on　information　entropy　is　proposed.　We　introduce　a　time　and　age　weight　to　alleviate　the　influence　on　the　interest　change　of　different　age　groups.　Some　experiments　were　made　to　test　the　methods　on　the　Movielens　dataset,　and　encouraging　results　were　obtained.