A Method For POI Status Changes Detection Based On Weibo Data

As an important digital infrastructure for location based service,POI(Point of Interest)has been put forward higher requirements regarding the data quality and freshness,under timeliness demand of location based service for public.However,due to the problems of complex and diverse types,rapid status changes and high cost of data update,the contradiction between the POI freshness and high timeliness of service demand is increasingly prominent.Therefore,this paper introduces a method for detecting POI status changes based on Weibo data,which detects POI status changes from the social media data,with big volume,high timeliness and rich semantic information.It can make the POI fresher,which is of great research significance and practical application values.The proposed method combines the name,location and current state description from Weibo data to complete the POI state changes detection,and the main tasks are as follows:(1)Aiming at the limited recognition entity types and the lack of recognition ability of POI types with complex and flexible word-building in current geographic named entity recognition algorithms,this paper proposes a method for POI name recognition by combining conditional random field algorithm and rules.To improve the accuracy of POI name recognition,this method selects five feature factors from aspects of POI word-building features,Weibo language expressions and regional characteristics,establishes word dictionary constraints,and designs corresponding CRF feature templates for features fusion,so as to improve the accuracy of POI name recognition.By the experiment verification,the proposed method can achieve good results in the recognition of complex commercial POI names.(2)Aiming at the problem that the Weibo check-in positions deviate from the POI real positions,this paper introduces a method to deduce POI positions by combining text address and check-in coordinates.This method constructs an address database to extract the unstructured address information from Weibo,and then selects the correct text address from extraction results based on text relationship,address hierarchy and distance calculation.Finally,POI location coordinates are deduced by combining check-in coordinates under the constraint of coordinate and address consistency,which can improve the reliability of obtained coordinates from Weibo.(3)Aiming at the problem of insufficient change detection of POI current status in previous studies,this paper introduces a method to support multiple types of POI status changes detection.This method backtracks the Weibo recognized POI names,and further extracts and optimizes the status words through the three-level status word database.Finally,it maps the status words to the corresponding change types.Through case study,the accuracy of change detection results is quantitatively evaluated,verifying the effectiveness of the proposed method.