A Research And Implementation On A Method For POI Recommendation Based On Topic Model

As the rapid advance and improvement of mobile devices,GPS and Web2.0 technology,location-based social networks(LBSNs)attracts a mass of users to share sign-in information,including location information,user reviews and experience share.Because of the quick progression in cyber era,and the explosive growth of information,big profits have been made by traditional recommendation which solved the problem of filtering personal information.Similarly,point-of-interests(POIs)recommendation plays an important role in LBSNs.It not only helps users to find new POIs,but also helps POI service providers to issue advertisements about POI location in network platform.Most of current POI recommendation technologies only used the users' sign-in information in LBSNs.They learn the users' preference on POIs by the sign-in frequencies.Although geo matrix factorization(GeoMF)uses matrix factorization(MF)to pattern factors of geographic location,GeoMF ignores the impact of review texts on users' preference extent on POIs.So we add geographic information in MF to combine sign-in times and geographic information.In addition,we consider review texts as potential sign-in times to better reflect the sign-in times on one POI.We use the data from living lives as experimental data which showed that this POI recommendation pattern significantly improves performance and accuracy of recommendation.