In　recent　years,　the　amount　of　data　generated　by　online　platforms　has　grown　exponentially,　making　it　challenging　for　users　to　process　large　amounts　of　information.　As　a　result,　recommendation　systems　greatly　improve　user　experience　and　retention　by　efficiently　matching　the　content　or　products　that　users　are　interested　in.　However,　most　existing　models　only　consider　long-term　preferences,　ignoring　the　dynamic　users＇　preferences.　To　address　this　issue,　this　paper　proposes　a　novel　recommendation　method　that　leverages　heterogeneous　information　networks　to　capture　both　long-term　and　short-term　preferences.　Our　model　involves　two　components:　a　heterogeneous　graph　neural　network　that　captures　the　long-term　preferences　of　users　and　an　attention　mechanism　that　focuses　on　short-term　preferences.　The　heterogeneous　graph　neural　network　extracts　relevant　information　from　the　input　sessions　to　learn　feature　representations,　and　the　attention　mechanism　weights　the　features　according　to　each　user＇s　current　interests.　We　evaluate　our　method　on　two　public　datasets,　and　our　results　show　that　our　model　outperforms　existing　approaches　in　terms　of　accuracy　and　robustness.　In　conclusion,　our　proposed　method　provides　a　useful　framework　for　developing　more　efficient　and　proactive　recommendation　systems　that　can　adapt　to　users＇　ever-changing　preferences.　The　incorporation　of　both　long-term　and　shortterm　preferences　allows　our　model　to　provide　more　accurate　and　personalized　recommendations,　improving　the　user　experience.