Collaborative　filtering　algorithms　have　obvious　advantages　in　recommendation　accuracy,　and　Bandit＇s　algorithm　is　a　strategy　to　address　diversity　needs.　The　COFIBA　algorithm　combines　the　collaborative　filtering　algorithm　with　the　Bandit　algorithm　to　provide　a　solution　for　recommending　the　balance　of　diversity　and　accuracy.　However,　COFIBA　does　not　consider　the　influence　of　time　characteristics,　and　COFIBA　is　a　cumulative　regret.　It　is　relatively　slow　to　solve　the　problem　of　diversity.　Therefore,　this　paper　proposes　a　learning-based　model.　On　the　one　hand,　it　introduces　the　openness　characteristics　of　users　to　achieve　diversity　recommendation,　and　relies　on　the　＇exploration-feedback-update＇　strategy　to　adjust　the　user＇s　openness.　At　the　same　time,　the　time　factor　is　incorporated　into　the　COFIBA　algorithm　as　a　feature,　and　the　change　of　user　interest　with　time　is　analyzed　to　ensure　the　accuracy　of　recommendation.　The　experimental　results　show　that　the　combination　algorithm　with　time　and　open　features　has　a　significant　improvement　in　the　diversity　and　accuracy　of　the　results　compared　with　the　COFIBA　algorithm.　©　2019　IEEE.