Improving Rating and Relevance with Point-of-Interest Recommender System
This work addresses the challenge of expensive user feedback collection in POI recommender systems for location-based social networks, though it appears incremental by building on existing retrieval-based methods.
The paper tackled the problem of improving point-of-interest (POI) recommendations by integrating multiple feedback sources, such as user-item interactions and auxiliary contextual information, into a deep neural network model, resulting in significant improvements on a large-scale dataset.
The recommendation of points of interest (POIs) is essential in location-based social networks. It makes it easier for users and locations to share information. Recently, researchers tend to recommend POIs by treating them as large-scale retrieval systems that require a large amount of training data representing query-item relevance. However, gathering user feedback in retrieval systems is an expensive task. Existing POI recommender systems make recommendations based on user and item (location) interactions solely. However, there are numerous sources of feedback to consider. For example, when the user visits a POI, what is the POI is about and such. Integrating all these different types of feedback is essential when developing a POI recommender. In this paper, we propose using user and item information and auxiliary information to improve the recommendation modelling in a retrieval system. We develop a deep neural network architecture to model query-item relevance in the presence of both collaborative and content information. We also improve the quality of the learned representations of queries and items by including the contextual information from the user feedback data. The application of these learned representations to a large-scale dataset resulted in significant improvements.