Cross-City POI Recommendation Combining Intention-Aware and Review Sentiment Analysis

Abstract

Cross-city Point of interest (POI) recommendation aims to recommend locations for users who move from a source city to a target city. Most existing cross-city POI recommendation methods consider users’ check-in records from their source city to analyze user preferences, while neglecting the crucial information in the reviews, which is of importance for modeling user preferences. Additionally, users often have different travel intentions when traveling to other cities, how to accurately mine users’ travel intention in the target city is still challenging. To address above issues, a cross-city POI recommendation method called IARSA is proposed, which integrates user intention aware and review sentiment analysis. Firstly, gated graph neural network (GGNN) is employed to obtain user preferences for the source city, and variational deep embedding is employed to mine users’ different travel intensions in the target city. Then the user drift preference representation is generated with user intentions and inherent preference. Next, we analyze the sentiment information of POI reviews, and utilize the Gaussian Mixture Model (GMM) to model reviews in POIs and generate sentiment vectors for POI. The comprehensive representation of the POIs is a combination of the sentiment vector and the geographic vector generated by Graph Convolutional Network (GCN). Finally, relevance scores for POIs in the target city are computed with user drift preference embedding and POI representations. Extensive experiments conducted on three real-world datasets demonstrate that our approach significantly outperforms state-of-the-art baseline methods, achieving superior results in both precision and recall metrics.