Remerciez-le!

Remerciez @Admin pour avoir partagé cet document gratuitement, de la manière la plus simple, en partageant sur les réseaux sociaux.

Exploiting POI-Speciﬁc Geographical Inﬂuence for Point-of-Interest Recommendati

Exploiting POI-Speciﬁc Geographical Inﬂuence for Point-of-Interest Recommendation Hao Wang1,2, Huawei Shen1,2, Wentao Ouyang1, Xueqi Cheng1,2 1Institute of Computing Technology, Chinese Academy of Sciences, Beijing, China 2University of Chinese Academy of Sciences, Beijing, China wanghao@software.ict.ac.cn, {shenhuawei,ouyangwt,cxq}@ict.ac.cn Abstract Point-of-Interest (POI) recommendation, i.e., rec- ommending unvisited POIs for users, is a funda- mental problem for location-based social networks. POI recommendation distinguishes itself from tra- ditional item recommendation, e.g., movie rec- ommendation, via geographical inﬂuence among POIs. Existing methods model the geographical inﬂuence between two POIs as the probability or propensity that the two POIs are co-visited by the same user given their physical distance. These methods assume that geographical inﬂuence be- tween POIs is determined by their physical dis- tance, failing to capture the asymmetry of geo- graphical inﬂuence and the high variation of ge- ographical inﬂuence across POIs. In this paper, we exploit POI-speciﬁc geographical inﬂuence to improve POI recommendation. We model the geographical inﬂuence between two POIs using three factors: the geo-inﬂuence of POI, the geo- susceptibility of POI, and their physical distance. Geo-inﬂuence captures POI’s capacity at exerting geographical inﬂuence to other POIs, and geo- susceptibility reﬂects POI’s propensity of being ge- ographically inﬂuenced by other POIs. Experi- mental results on two real-world datasets demon- strate that POI-speciﬁc geographical inﬂuence sig- niﬁcantly improves the performance of POI recom- mendation. 1 Introduction Location-based social networks (LBSNs), such as Foursquare and Gowalla, are increasingly popular, bridging the gap be- tween the physical world and online social networking ser- vices [Xiao et al., 2010; Sun et al., 2017]. In LBSNs, users share their locations and content associated with location information, facilitating the understanding of users’ prefer- ence and behavior [Bao et al., 2012; Liu and Xiong, 2013; Gao et al., 2015; Wang et al., 2015a]. Point-of-Interest (POI) recommendation, i.e., recommending for users unvis- ited POIs (e.g., restaurants, shopping malls, and theaters) ac- cording to users’ check-in records, gains great research in- terest in the last few years [Li et al., 2016; He et al., 2016; Zhang et al., 2016; Li et al., 2017]. One of the most prominent features for POI recommen- dation is that locations of POIs and target user are critical factors for recommendation. For example, in Gowalla and Foursquare, 90% of users’ consecutive check-ins are within the distance less than 50km [Liu et al., 2017]. Therefore, besides modeling users’ preference from the interaction be- tween users and POIs, as done in traditional item recom- mendation, researchers devote to exploiting the geographical proximity or geographical inﬂuence among POIs to improve the performance of POI recommendation [Ye et al., 2011; Lian et al., 2014; Xie et al., 2016]. Existing methods that exploit geographical inﬂuence for POI recommendation roughly falls into two paradigms. The ﬁrst kind of methods leverages the geographical proximity to improve the learning of users’ preference, assuming that POIs in close proximity to each other share similar user pref- erences [Liu et al., 2014; Li et al., 2015; Xie et al., 2016; Feng et al., 2017]. For these methods, geographical prox- imity is used as a kind of spatial regularization for users’ preferences. The second kind of methods explicitly mod- els the geographical inﬂuence among POIs as the probabil- ity or propensity that the two POIs are co-visited by the same user given their physical distance [Ye et al., 2011; Cheng et al., 2012; Zhang and Chow, 2013; Lian et al., 2014; Saleem et al., 2017]. Various forms of functions, e.g., power law function and Gaussian distribution, are employed to cap- ture the co-visited probability distribution of POIs with re- spect to their physical distance. Although the aforemen- tioned methods gain some success at leveraging geographi- cal inﬂuence, they are incapable to capture the high varia- tion of geographical inﬂuence across POIs. For example, as shown in Figure 1, 10 randomly-selected POIs in Foursquare dataset exhibit quite different geographical inﬂuence, indicat- ing that geographical inﬂuence cannot be well captured solely by physical distance and thus geographical inﬂuence should be POI-speciﬁc. In this paper, we exploit POI-speciﬁc geographical inﬂu- ence to improve POI recommendation. We model the POI- speciﬁc geographical inﬂuence between two POIs using three factors: the geo-inﬂuence of POI, the geo-susceptibility of POI, and their physical distance. Geo-inﬂuence captures POI’s capacity to spread its visitors to other POIs, and geo- Proceedings of the Twenty-Seventh International Joint Conference on Artiﬁcial Intelligence (IJCAI-18) 3877 0 10 20 30 40 50 60 distance / (0.5 km) 0 2 4 6 8 POI id (0-10) 0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 Figure 1: Heat map of the check-in correlation over distance of 10 randomly sampled POIs on the Foursquare dataset. Take 0.5 km as one bin, and for each bin, we count the average correlation between each selected POI and POIs falling into the bin. We normalize these values by the largest one. susceptibility reﬂects POI’s propensity of receiving visitors from other POIs. For example, subway stations generally have high geo-inﬂuence and restaurants usually have high geo-susceptibility. Here, geo-inﬂuence and geo-susceptibility are two low-dimensional vectors, and the geographical inﬂu- ence between two POIs is represented by the inner product of the geo-inﬂuence vector of one POI and the geo-susceptibility vector of the other POI. Our model for POI-speciﬁc geographical inﬂuence has two unique beneﬁts: (1) Geographical inﬂuence between POIs is asymmetric, offering high ﬂexibility to capture the high vari- ability of geographical inﬂuence across POIs. (2) Instead of directly modeling the POI-speciﬁc geographical inﬂuence us- ing a POI interaction matrix, our model represent geograph- ical inﬂuence by two low-dimensional vectors for each POI, signiﬁcantly reducing the number of free parameters [Wang et al., 2015b]. Thus, our model is appropriate for POI recom- mendation which suffers from severe data sparsity issue. Finally, we integrate POI-speciﬁc geographical inﬂuence into a standard model that captures users’ preference, forming a new POI recommendation method. We train our model us- ing users’ check-in records and validate the recommendation performance by applying the model to “predict” the POIs that they are likely to visit in the near future. We conduct exten- sive experiments on two real-world datasets from Foursquare and Gowalla to illustrate the effectiveness of our model. Ex- perimental results demonstrate that POI-speciﬁc geographi- cal inﬂuence signiﬁcantly improves the performance of POI recommendation, outperforming state-of-the-art POI recom- mendation methods. 2 Related Work In this section, we give a brief review about POI recom- mendation. POI recommendation recommends for users un- visited POIs according to users’ check-in records. Consid- ering users’ check-ins are implicit feedback, existing meth- ods model check-ins either by ﬁtting scores converted from check-in counts [Lian et al., 2014] or by optimizing a pair- wise ranking of users’ preferences to POIs [Li et al., 2015; 2016; Zhao et al., 2017]. Due to the sparsity of users’ check-ins, only exploiting check-in counts often suffer from poor performance. Aux- iliary information can be incorporated to alleviate this sit- uation. For example, geographical inﬂuence is one of the most important factors and it does not exist on the online recommendation sense. Existing methods of modeling ge- ographical inﬂuence can be grouped into two categories, i.e., global methods [Ye et al., 2011; Cheng et al., 2012; Zhang and Chow, 2013; Lian et al., 2014] and regional meth- ods [Liu et al., 2014; Li et al., 2015; Xie et al., 2016; Feng et al., 2017]. Global methods model the relation between POIs’ cooc- curence and their geographical coordinates. Ye et al. [2011] and Lian et al. [2014] respectively use a power-law distribu- tion and a Gaussian distribution to characterize geographi- cal inﬂuence over distance. [Cheng et al., 2012; Zhang and Chow, 2013] capture the scatter plot of each user’s check- ins (e.g., the longitude and latitude) by a ﬁxed distribution. Regional methods consider that POIs in a same geographi- cal region share similar attraction to users. [Xie et al., 2016; Feng et al., 2017; Zhao et al., 2017] use representation-based learning method and restrict POIs in the same region share similar representations. [Liu et al., 2014; Li et al., 2015] di- rectly calculate the attraction of a target POI by considering the attraction of its geographical neighbors. However, global methods and regional methods provide two coarse grained representations of geographical inﬂuence, which ignore the POI-speciﬁc attributes. We address the problem in this paper. In addition, many studies have explored other information to facilitate POI recommendation performance, such as social relationship [Tang et al., 2013], temporal factors [Yuan et al., 2013] and category [Zhang and Chow, 2015], etc. 3 Preliminary We denote with U and I the set of users and the set of POIs respectively. For a user u and a POI i, we denote with cui as the number of times that user u visited POI i, and wui is a scaled version of cui. All POIs that user u visited form his/her check-in history, denoted as Hu. For each POI i, its location is denoted as longitude loni and latitude lati. We use dij to represent the physical distance between POI i and POI uploads/Management/ exploiting-poi-specific-geographical-influence-for-point-of-interest-recommendation.pdf