Abstract: In check-in records with corresponding locations,each record only contains the attributes of name and location,i.e.,longitude and latitude.Traditional name deduplicating algorithms deduplicate names by matching attributes between two entities or computing similarity between names of the two entities,and thus neglect the particularity of locations.In order to improve the quality of name deduplicating in spatial records,this paper proposed an expectation maximization based name deduplicating algorithm.Firstly,we proposed a text name model containing core and background words,and gave an expectation maximization algorithm for computing parameters of the model.Secondly,we introduced location into the text name model,partitioned the whole world into tiles,computed the distributions of core and background words in each tile,and proposed a text name model including location.Finally,we used the location text name model to deduplicate names in location records,and presented corresponding name deduplicating algorithm.The experiments show that,our proposed algorithm can better recognize core word in a name than related works,and thus performs better while deduplicating name in location records.

Key words: Check-in,Location,Expectation maximization,Name deduplicating