Personalized POI Recommendation Integrating Indoor Location Big Data And Knowledge Graph

In urban life,humans spend 87% of their time in indoor spaces.With the increasing number of large-scale indoor places and the rapid growth of intelligent information services,indoor items,information,resources are diversified and complex,information overload is increasingly serious,people are difficult to select the information they need,so it is of great significance to build an indoor recommendation system to provide personalized services for people.With the rapid development of smart phones and location technology,Location Based Service(LBS)is widely used in daily life around the world.making it possible to solve this problem.By analyzing the characteristics of massive positioning data and complex indoor spatial environment,this thesis uses the distributed computing framework to efficiently mine the correlation between user trajectory data and indoor Point of Interest(POI),and establishes user semantic trajectories.Integrate social media data and place model to construct an indoor POI knowledge graph,and introduce the knowledge graph into the recommendation system as auxiliary information to build a personalized preference model to complete personalized recommendation.The main research contents of this thesis are as follows::(1)Spark-based indoor semantic trajectory extraction.Based on the Spark distributed computing framework,this thesis firstly performs outlier detection and median filtering on the indoor moving object positioning data to filter the noise,and secondly,by setting the time and space thresholds,the trajectory data is extracted and semantically enhanced.Indoor Semantic Trajectory Extraction.(2)Construction of indoor POI knowledge graph integrating social media data and place model.Based on the concept of the place model,this thesis firstly constructs the ontology of the indoor POI knowledge graph from the three aspects of the functional attribute of the place,the spatial location of the place and the emotion of the place,and constructs the indoor POI knowledge graph model layer;The structured and unstructured knowledge corresponding to POIs obtained from social media pages is used as the data layer of the knowledge graph;on this basis,the fine-grained places of indoor POIs are extracted from the attribute features of POIs and the fine-grained sentiment analysis model.The emotion model further supplements the entity relationship of the indoor POI knowledge graph;thus establishing the indoor POI knowledge graph.(3)A personalized POI recommendation model that integrates indoor space features and knowledge graphs.This thesis firstly uses Voronoi diagram and sliding window algorithm to comprehensively consider the space and relationship between indoor POIs to generate POI heterogeneous graphs,and secondly uses graph embedding algorithm Node2 vec to sample POI spatial proximity graphs and POI motion flow graphs at the same time.Generate random sequences and input them into the skip-gram model for joint embedding.As the final indoor space vector of POI,calculate the cosine similarity between POI vectors to generate the indoor POI spatial similarity matrix,and finally integrate the indoor spatial similarity matrix into the interest diffusion.In the POI-RippleNet framework,the decision-making process of people in indoor space is restored,and the personalized recommendation of POI is completed.And it outperforms other benchmark algorithms on real datasets.