Research On Implicit Location Estimation Method Of User Generated Content In Social Network

With the rise of social networks,users generate a large amount of user-generated content on social networks,where location-based user-generated content can provide services for personalized applications such as personalized recommendations,event detection,and public opinion monitoring.However,only a very small number of user-generated content in social networks are marked with location,and the location estimation methods that have generated content for users are mostly coarse-grained at the city level,which is not conducive to providing users with more accurate personalized services.To this end,this thesis proposes a method for speculating the fine-grained location implied by user-generated content in social networks.The method includes the following three parts.First solve the entity identification problem in user generated content.This thesis proposes an entity recognition model based on knowledge base and natural language processing technology.It uses natural language processing technology to identify sentence components,identifies entities in different sentence components according to the knowledge base,and obtains an entity for the misspelled entity by stemming.To some extent,it solves the problems of spoken words,spelling mistakes,irregular writing,etc.of user-generated content in social networks.Second,solve the filtering problem of user-generated content.User-generated content in social networks is very broad,and not all user-generated content is suitable for location speculation.Therefore,this thesis proposes a user-generated content filtering model that determines whether user-generated content is available for location speculation based on the gazetteer and entity location correlation functions.Finally,an entity location model is constructed for user generated content to infer the implicit location.The model takes the data in the social network based on geographical location as the data set,extracts the entities in the content corresponding to each geographical location in the data set,and constructs the entity location model between entities and locations.The candidate location of the user-generated content is estimated by the entity location model,and the candidate locations are sorted,and the optimal location is taken as the speculative location.Experiments on three data sets show that the proposed method can effectively identify the hidden location in user generated content.Compared with the existing methods,the average error distance of the guessed results is increased by 18%,10%and 15%respectively,which shows the effectiveness of the proposed method.