A User Feedback-based Personalized Spatial Keyword Semantic Query Approach

With the widespread application of mobile network and GPS,it becomes easier to obtain spatial dimension information,and more and more Point of Interests(POIs)appear on the Web,which contains location information(generally represented by longitude and latitude)and textual information.Location-based service systems(such as Ctrip,Meituan,DiDi chuxing,etc.),according to the user's location and query keywords,provide users with query results that are close to the location and match the text.Therefore,the research of spatial keyword query has become a hot spot in the field of database query.Existing spatial keyword query methods usually evaluate text relevance according to the frequency of occurrence of query keywords in spatial object text information,while do not considerthe users' preference for different query keywords.Besides,only text matching is supported without considering the semantic relevance between query keywords and spatial object text information.To solve the above problems,this paper proposes a user feedbackbased personalized spatial keyword semantic query approach.This method is divided into two stages,the offline processing and the online query processing.In the offline processing stage,Gibbs algorithm is used to estimate the thematic probability distribution of spatial object text information,and then LDA model is used to extend the spatial data set semantically.During the online query processing stage,for the initial query conditions of users,the IR-tree hybrid index structure is first used to obtain candidate query results from the expanded spatial database.Then,the user explicitly marked the relevant query results(i.e.relevant feedback)in the candidate set according to his/her preferences.According to the user's feedback information,Rocchio algorithm is leveraged to update the user's initial query conditions,so that the new query conditions are closer to the user's actual needs and preferences.The updated query condition is then retrieved to obtain a new candidate set and the feedback process is repeated until the query result is satisfactory to the user or the stop threshold is reached.Experimental results show that the proposed approach can effectively capture users' implicit/explicit preferences and reflect semantic relevance,which to some extent improves the personalized degree and accuracy of spatial keyword query results.The paper includes 14 figures,9 tables and 58 references.