Mining User Check-In Behavior with a Random Walk for Urban Point-of-Interest Recommendations

Abstract

In recent years, research into the mining of user check-in behavior for point-of-interest (POI) recommendations has attracted a lot of attention. Existing studies on this topic mainly treat such recommendations in a traditional manner-that is, they treat POIs as items and check-ins as ratings. However, users usually visit a place for reasons other than to simply say that they have visited. In this article, we propose an approach referred to as Urban POI-Walk (UPOI-Walk), which takes into account a user's social-triggered intentions (SI), preference-triggered intentions (PreI), and popularity-triggered intentions (PopI), to estimate the probability of a user checking-in to a POI. The core idea of UPOI-Walk involves building a HITS-based random walk on the normalized check-in network, thus supporting the prediction of POI properties related to each user's preferences. To achieve this goal, we define several user-POI graphs to capture the key properties of the check-in behavior motivated by user intentions. In our UPOI-Walk approach, we propose a new kind of random walk model-Dynamic HITS-based Random Walk-which comprehensively considers the relevance between POIs and users from different aspects. On the basis of similitude, we make an online recommendation as to the POI the user intends to visit. To the best of our knowledge, this is the first work on urban POI recommendations that considers user check-in behavior motivated by SI, PreI, and PopI in location-based social network data. Through comprehensive experimental evaluations on two real datasets, the proposed UPOI-Walk is shown to deliver excellent performance. . 2014. Mining user check-in behavior with a random walk for urban point-of-interest recommendations.

GeoSoCa

“…Most current works derive the similarities between users from social links and put them into the traditional memory-based or model-based collaborative filtering techniques. For example, some literatures [4,20,23,27,30,31,34,35] seamlessly integrated the similarities of users into the user-based collaborative filtering techniques, while others [2,21,36] employed the user similarities as the regularization terms or weights of latent factor models. In this paper, we contrive a new method to exploit the social correlations between users by aggregating the checkin frequency or rating of friends to POIs and transforming them into relevance scores, based on the estimated social check-in frequency or rating distribution from the historical check-in data of all users.…”

“…Bao et al [1] calculated the category biases of users to compute the similarity of the users for the user-based collaborative filtering method. Besides the category biases of users, Ying et al [27] also derived the category weights of POIs from the tags annotated on the POIs and then estimated the relevance scores between users and POIs based on the inner product of the category biases and weights. Liu et al [15] clustered POIs into groups based on their categories, built a user-category transition matrix instead of user-POI check-in matrix from the historical check-in data of users, and applied the matrix factorization technique to discover the next top-k categories that a user would like to check in.…”

“…Most existing works regard the popularity of a POI as the universal prior preference of users to the POI. Specifically, the study [27] employed the prior preference of users to unvisited POIs as the weight of the edges between the users and POIs in the complete bipartite graph, while other studies utilize the prior preference to adjust the posterior preference that are derived from the geographical information [8,13,14,28]. However, in these studies the prior preference is not personalized for users and thus in practice the benefit from the popularity is considerably limited.…”

“…Our method is different from the existing works [1,8,15,18,27,36] that separately utilize the category and/or popularity information of POIs. (Section 3.4)…”

“…Our method is distinct from the current works [2,4,20,21,23,27,30,31,34,35,36] that derive the similarities between users in terms of their social links and then integrate them into the traditional collaborative filtering techniques. (Section 3.3)…”