<p>As social media platforms with check-in features continue to grow in popularity, next POI (point-of-interest) recommendation has emerged as a prominent area of research. As a result, most of studies leverage spatiotemporal information of POIs visited by users for next POI recommendation. However, the current task of recommending next POIs based on users’ long-term and short-term preferences mainly focuses on the impacts of serial factors of POIs, ignoring the effects of regional information in long-term preferences as well as the correlations of temporal information with other multi-dimensional auxiliary information in short-term preferences. Therefore, we propose a geographic region-based long-term and short-term preferences learning model (GLSPL), which comprehensively considers the multi-dimensional auxiliary information in users’ check-in sequences, thereby providing more precise recommendation. Specifically, for long-term preference learning, our proposed model uses the Geohash algorithm to convert the latitude and longitude of POIs into Geohash IDs. This enables us to encode spatial regions at multiple granularities and capture inherent spatial correlations between adjacent POIs—addressing the limitation of prior work that treats POIs as isolated spatial points. By doing so, we fully incorporate the impact of POI regional information on users’ long-term preferences. For short-term preference learning, the proposed GLSPL model integrates temporal information with other auxiliary information to model their correlations, so as to capture users’ short-term preferences more accurately. Then, we adopt a personalized combination unit to derive users’ comprehensive preferences by combining long-term and short-term preferences. Finally, we evaluated our model using two real-world public datasets. The trial results clearly show that our proposed model is better than the state-of-the-art methods in the field of next POI recommendation.</p>