Context Information And User Perference Based POI Recommendation Algorithm Research And Implementation

With the rapid development of the Internet,LBSN has more and more applications.As a key application in the LBSN scenario,point of interest recommendation has become a current research hotspot.However,there are still many shortcomings in most of the current research work:1)the existing knowledge embedding methods are difficult to capture the associations of all triples;2)the current method of using contextual information to predict user preferences is insufficient to extract some important information and is difficult Generate accurate user preferences and affect the recommendation results;3)Existing models use check-in records to predict user preferences,and poor recommendation results for users with sparse check-in records.Aiming at the shortcomings of the current research work,this paper proposes algorithm solutions and forms the final recommendation framework.The main tasks include:(1)An improved knowledge representation method is proposed.The loss function of knowledge representation algorithm based on translation model is improved to solve the problem of non-convergence of partial triples of existing translation models.The comparison of the two algorithms on the data set shows that the new algorithm has better performance.(2)Time Intercal and Frequecy based Long Short-Term Memory(TIFLSTM)is proposed,which takes into account the influence of check-in time and check-in times on user preferences.Compared with Long Short-Term Memory(LSTM),a better preference is achieved.forecast result.(3)In view of the lack of user travel data or the lack of travel data,a preference fusion algorithm is proposed to mine the user's preferences from friends with similar interests,which solves the problem that traditional models have difficulty predicting sparse data users.(4)Design and implement the recommendation module according to different recommendation requirements,and test it.