Research Of Location Prediction And Service Recommendation Based On Temporal And Spatial Data Mining

With the rapid development of location-based services and social networks, location-based social network(LBSN) platform has been widely concerned, users can chenk in based on the current location and share their checkins with friends on LBSN platform. Meanwhile, the location-based social networks generate a large number of information, which provide big data for the study of the next-place prediction and the personalized recommendation, on the other hand, next-plaxe prediction and personalized recommendation also provides users with convenience.Most of the existing studies only simply based on the user's historical location information, doesn't consider the user's sequential move and sociability in LBSN, resulting in the presence of a single characteristic and the low accuracy, the paper will carry out the research about next-place prediction and recommended service based on temporal and spatial data mining in LBSN. The main work includes:This thesis preprocessed and analyzed the data sets. Through a series of statistical processing methods, after mining in many aspects including space, time and social contact, get the features of user behavior, analyzed the important factors affecting the user's mobility and social contact. In addition, by using the detection of stay points and location clustering algorithm, complete location information abstract process, which provide data support for constructing model and designing similar measures.This thesis proposed the n order Markov model with temporal features(n-TMM), which add temporal features to the mobile Markov Model, then constructed state transition matrix to summarize the user's moving law, taking into account the factors of space and time on the user's mobile behavior. Then the next-place prediction research experiment is conducted on this basis, analyze the effectiveness of this method.This thesis designed a mixed similarity metric which combined location sequence similarity and friend similarity, this metric considers both location sequence and social relationship, then it's applied to the research of friend recommendation and place recommendation. Compared with baseline similarity in experimental evaluation, it shows that this method improves the issue of a single characteristic, achieves better recommendation results.