Research On User Context-aware Preferences Mining Based On Location Social Semantics

User activity preference mining is an important issue in the study of human mobile behavior patterns in GIS,and it makes significant sense in realizing personalized recommendation,urban planning and management.In recent years,with the rapid development and advancement of sensor technology,mobile communication technology,and Internet technology,large-scale data of individuals and spatial temporal markers have been generated,which has also greatly promoted the widespread application of location-based social tagging.Location-attached social annotation expands geographical positions into social locations,and "sociality" becomes the fifth dimension after time and space,and reflects user behavior patterns,preferences,etc.,providing a new research perspective for user activity preference mining.At present,user preference research mainly focuses on the characteristics of mobile space-time or location inherent attributes,ignoring the relevance between user preference and location social semantics,it is difficult to explore users more dimensions and more granular activity preference.User location comments often directly reflect the users'activity preference.How to mine users' fine-grained and multi-dimensional activity preference related to contexts such as time,space,and location comment in the large-scale user activity trajectory that attached time,space,and location comment has become a new research challenge.Location activity and comment tags are a more granular form of expression than categories.The various aspects of the location tags reflect the users' preference at different levels.Integrating GIS theory and Labeled LDA topic modeling technology,this thesis proposes a user context-aware preference mining method based on location social semantics,focusing on the formal expression of multi-dimensional location social semantics,the spatial clustering of users' activities under the dual constraints of space and semantics and the construction of context-aware preference mining model.It is implemented in the trajectory of the users' moving locations to mine users' fine-grained activity preference related to contexts such as time,space,and location comment.The specific research content and innovations are as follows:(1)Ontology modeling of multi-dimensional location social semantics.Introduce location comments,expand the social semantic dimension of location,design a new location social semantic ontology model,analyze the weighted association of location social semantics,and realize the formal expression of the concepts of multi-dimensional social semantics,the semantic hierarchical and weighted association relationships.(2)Spatial-semantic dual-constraint clustering based on tihe location social semantic ontology.The concept of personal activity area is proposed to express user activity space with similar social semantic features.A spatial-semantic dual-constrained clustering algorithm based on the location social semantic ontology is proposed to descover personal activity areas with close spatial distance and similar social semantics in geographical space.(3)Unified modeling and parameters estimation of correlation between user activity preference,spatiotemporal characteristics and location comments.Define a new context-aware preference concept.The Labeled LD A topic model is extended,and the topical classification label of the Labeled LDA model is constructed based on the location activity and user comment tags of the location social semantic ontology.A context-aware preference generation model is proposed,and the context-aware preference probability distribution related to the users5 time,space,and location comment is tapped.Based on Collapsed Gibbs,a context-aware preference generation model inference algorithm is proposed to estimate the probability distribution parameters of context-aware preference,and then the users' context-aware preference distribution is discovered.Through the experiments of the real dataset in Dianping,we find that the location social semantic ontology constructed in this thesis can accurately reflect the relationship between social semantic concepts.Compared with other baseline approaches in various datasets,the context-aware preference generation model has a higher accuracy of preference recognition and proves the effectiveness of the proposed approach.