Abstract
POI recommendation is a key task in tourism information systems. However, in contrast to conventional point of interest (POI) recommender systems, the available data is extremely sparse; most tourist visit a few sightseeing spots once and most of these spots have no check-in data from new tourists. Most conventional systems rank sightseeing spots based on their popularity, reputations, and category-based similarities with users' preferences. They do not clarify what users can experience in these spots, which makes it difficult to meet diverse tourism needs. To this end, in this work, we propose a mechanism to recommend POIs to tourists. Our mechanism include two components: one is a probabilistic model that reveals the user behaviors in tourism; the other is a pseudo rating mechanism to handle the cold-start issue in POIs recommendations. We carried out extensive experiments with two datasets collected from Flickr. The experimental results demonstrate that our methods are superior to the state-of-the-art methods in both the recommendation performances (precision, recall and F-measure) and fairness. The experimental results also validate the robustness of the proposed methods, i.e., our methods can handle well the issue of data sparsity.
Introduction
Point-of-Interest (POI) recommendation in tourism faces unique challenges compared to daily-life recommendation systems. While standard systems predict routine visits based on frequent history, tourists typically visit a location once and often arrive in a new city with no local check-in history. This leads to extreme data sparsity and the "cold-start" problem.
Furthermore, existing systems often suffer from popularity bias, recommending famous landmarks regardless of individual preferences. This paper proposes shifting the focus from specific locations to the underlying user experience—the activities (e.g., "eating," "sightseeing") that drive tourist behavior.
Methodology
Basic User Experience Model (B-UEM)
The core of the proposed framework is the Basic User Experience Model (B-UEM), a probabilistic generative model. B-UEM assumes that a latent factor, the "User Experience" ($g_d$), generates four observable factors: the tourist ($u_d$), the activities or words ($w_{d,i}$), the POI ($l_d$), and the time of visit ($t_d$).
By modeling the "experience" as a latent variable, the system can group users and locations with similar activity patterns, effectively mitigating the impact of sparse data. The notation used in the model is detailed in the table below.
Pseudo-Rating Mechanism
To address the cold-start problem for new tourists, the authors introduce a pseudo-rating mechanism. When a new user ($u'$) enters the system, they are presented with a small set of images or keywords to "rate." The system then uses a Bayesian approach to infer the user's likely experience group distribution and recommends POIs that align with those preferences.
Experimental Evaluation
Datasets and Processing
The models were evaluated using two real-world datasets from Flickr:
- YFCC100M Kyoto: 783 tourists and ~1,500 spots.
- User Visits: A multi-city dataset (8 cities) with over 260,000 images.
To extract objective behavior words, the researchers employed automated scene detection (Places and Attribute methods) on geo-tagged images rather than relying on subjective manual tags.
Key Results
The proposed B-UEM and its variants (S-UEM, ST-UEM) were compared against baselines like GeoMF++ and SAE-NAD.
- Accuracy: B-UEM achieved significantly higher Precision, Recall, and F-measure. For instance, B-UEM-al reached a Precision@5 of ~0.137, more than doubling the performance of GeoMF (0.056).
- Cold-Start Performance: The pseudo-rating mechanism enabled an F-measure of ~0.321 for entirely new users, a scenario where traditional methods typically fail.
- Fairness: By focusing on experiences rather than just raw popularity, the system improved recommendation fairness (measured by the Gini index), distributing traffic across a wider variety of POIs.
Conclusion
This research demonstrates that modeling latent user experiences is a powerful approach for tourism POI recommendation. By decoupling the experience from specific location history, the B-UEM framework overcomes data sparsity and provides personalized, fair recommendations even for first-time visitors. Future work will focus on reducing the cognitive load of the pseudo-rating mechanism and deepening the semantic analysis of user activities.
Summary updated