Abstract:Point-of-Interest (POI) recommendation using preference mining based on spatial data ascertained through Location-based Social Networks (LBSNs) is a critical personalized recommendation task. An efficient recommendation system requires an enhanced representation of users' preferences. We propose High-order Spatial Connectivity Minning an enhancement over Neural Graph Collaborative Filtering (HSCM-NGCF) that solves the issue of over-smoothing in NGCF when the node embeddings are indistinguishable. The process involves three novel steps. First, the sequence propagation analysis is embedded in the interaction propagation layer to capture the high-order spatial connectivity in the interaction graph. Second, to emphasize and learn the relevant patterns, an adaptive weight mechanism is added that adjusts the relative importance of different edges to prioritize the edges with higher weights. Third, HSCM-NGCF can handle sparsity and noise better because it incorporates the heterogeneous structure of the input data into the embedding learning process, which helps to capture the complex relationships between different types of nodes by modifying the user-item embedding. It overcomes another demerit of NGCF wherein dissimilar items are often placed within the same set. It successfully encapsulates the local and global patterns in the user's check-in data by modeling the short-range dependencies from the immediate neighborhood and long-range dependencies from the distant nodes via sequential propagation. HSCM-NGCF captures the high-order collaborative signals embedded into the user-POI check-ins. The three essential elements of HSCM-NGCF are (i) the interaction layer that utilizes the spatial connectivity for adaptive edges to estimate the weights of edges, (ii) the aggregation mechanism that aggregates the message vectors from the previous layer, and (iii) the prediction layer that provides the final recommendation of POIs using both observed and unobserved POIs for the user. The approach has been tested over two datasets, Gowalla and Foursquare, and the results have been juxtaposed to several state-of-the-art approaches. We achieved an accuracy of 41.52 % with the Gowalla dataset and 38.13% with the Foursquare dataset.

ReGFFM : A New Graph Model Designed with Spatial Information for Potential Social Connection Mining

Community Mining From Multi-Relational Networks

Detecting User Community in Sparse Domain via Cross-Graph Pairwise Learning

Friendship prediction model based on factor graphs integrating geographical location

Multi-Graph Heterogeneous Interaction Fusion for Social Recommendation

An Adaptive Federated Relevance Framework for Spatial Temporal Graph Learning

Exploring multi-relational spatial interaction imputation with distance-decay effects

Mining Location-based Social Networks: A Predictive Perspective.

A Constraints Fusion-induced Symmetric Nonnegative Matrix Factorization Approach for Community Detection

REBUILD: Graph Embedding Based Method for User Social Role Identity on Mobile Communication Network.

Research on Resolution Limit of Community Detection in Location-Based Social Networks

Leveraging graph neural networks for point-of-interest recommendations

Social Relationship Link Inference Based on Graph Convolutional Networks

Actively Learning to Infer Social Ties

High-order spatial connectivity mining over neural graph collaborative filtering for POI recommendation in location-based social networks

Graph Convolutional Networks on User Mobility Heterogeneous Graphs for Social Relationship Inference.

STGSN — A Spatial–Temporal Graph Neural Network framework for time-evolving social networks

A Social Recommendation Model Based on Basic Spatial Mapping and Bilateral Generative Adversarial Networks

Meta-path Enhanced Lightweight Graph Neural Network for Social Recommendation

Predicting User Activity Intensity Using Geographic Interactions Based on Social Media Check-In Data

Mining Hidden Community in Heterogeneous Social Networks