POI Recommendation Towards Smart Tourism

Personalized point-of-interest(POI)recommendation is a challenging but vital task,also serve as an important part in urban tourism.Previous efforts on POI recommendation mainly focus on local users.According to user's activity areas,e.g.,home and workplace,nearby locations have higher probability to be recommended.However,in many practical scenarios such as urban tourism,target users are usually out-of-town travelers which are unfamiliar with target cities.Their preferences are hard to model due to sparse distributed check-ins.How to effectively solve the cross-city POI recommendation then becomes a great challenge.In this paper,we manage to improve the cross-city POI recommenda-tion accuracy for travelers,via finding correlations between different POIs.For cross-city POIs,the influence of travel intent(I),e.g.,business trip and family trip,is studied.For local POIs,we focus on their geographical neighbors(N).In addition,reviews(R)are in-troduced to bridge the gap between distant POIs and make recommendation explainable.Incorporating these three factors into the learning of latent space,a novel matrix factor-ization approach(INRMF)is proposed.Main contributions of this paper are as follows:·POI Feature Mining Model and quantize three kinds of POI features that have potential value.Including travel intent(I),geographical neighbors(N)and reviews(R).By conducting rating level statistics and analysis,we discover the correlations between POI attributes,which may have positive effect on improving POI recom-mendation accuracy.·POI Recommendation towards Urban Tourism Based on Matrix Factorization,we propose a cross-city POI recommendation framework towards smart tourism.This model can effectively solve sparse data issue and improve recommendation accuracy,via incorporating cross-city feature:travel intent and reviews,and local feature:geographical neighbors.·Semantic Feature Optimization We conduct phrase-level sentiment analysis to mine review contents.Applying word-embedding further improves the correlation between different words on semantic level.Based on optimized semantic feature,we propose two kinds of optimized MF model,which improve the utilization of reviews a lot.