Information Retrieval And User Data Mining Based On Geographic Information

With the development of Web technology, the web pages contain massive geographic information. Mining geographic information and making use of it in traditional information retrieval could help search engines to understand the intent of the query better and provide more personalized results for users. The new social network such as Twitter has appeared with the rapid developing of Mobile Internet technology and location-based services. In these social network, users can update their comments, check-in their activity and location in real-time. It became possible to learn users’scope of activities, interests and their habits through mining the massive real physical check-in data, which can provide better advertising services and personalized recommendation.The paper has researched the geographic and location information from the following three aspects:First, a document’s place names aware model for geographic information retrieval, which apply geographic information to the traditional information retrieval model. Different from other geographic information retrieval model, this model does not use the minimum bounding rectangle (MBR) to determine the scope of the document and index them in single geographic space, but index one document in several geographic spaces by place names mentioned in document, and then calculate the relevance between the query scope and all of place names of document. When fusing the textual and geographic models we take into account the place name’s geographic level features and the characteristics of the document but not a fixed weight. Experimental results show that the proposed approach can outperform baselines which are based on determining the MBR’s model and pure textual model of VSM.Second, geo-locate users based on their tweet content and social network in Twitter. Because users may leak some local words (such as place name and dialect word) in their update status and they are more concerned about the people living in the surrounding of them, the paper presents a fusion model which combined the textual and social network to predict the user’s location. In textual model, we present two methods to identity the local words: inverse location frequency and remote word filter, meanwhile, we also consider the traditional named entity recognition to identical the place names in tweets; in social network model, we consider the user’s network relationship of following and follower. Experimental results show that the proposed method to predict user location better than baseline methods. Finally, location recommendation in location-based services (LBS) based on user preference and time feature. It is convenient to learn user preference, activity scope and habits through the massive check-in data in LBS, and then we can provide more proper point of interest for user for their daily life. In this paper, we present a fusion recommendation model which takes into account the time feature of user and location’s check-in, the popularity of location, the user preference and the friend information of user. The experiments show that the recommendation model which considers the multiple features of location and user can really improve performance of recommendation in large-scale data of LBS.