Active Learning for Incremental POI Extraction and Pairing

摘要

The rapid development of the Internet and mobile smart devices has made the electronic map gradually become a good helper in our lives.If we hope the electronic map can provide a quality Location-Based Service, it must be able to help users accurately find nearby POIs (Point of Interest) in the nearby location, including food, clothing, housing, communications etc.The most powerful electronic map today is Google Maps. Many users are used to search for POIs with it. However, not all user-desired POIs can be found on Google Maps. Therefore, we have to expand the sources of POIs, and build a resourceful database of POIs for user queries. As the rise of social networking in recent years, users often share food information and travel experiences on these media. As the same time, businesses are in favor of setting up official pages to increase the visibility of their products.In this paper, we propose a web-based system, which could be roughly divided into the following three parts.The first part is the crawling of address associated snippets.The second part is the POI extraction model. Through the Conditional Random Field (CRF) and Conditional Random Field Sharp (CRF Sharp) as the learning algorithm. The purpose of this algorithm is to find out all the addresses and POI names in snippets.The third part is the POI pair verification model. The verification model is trained by the LibSVM learning algorithm, paired the address and POI name.