اکتشاف عکس های برچسب دار-جغرافیایی از طریق روش های داده کاوی

• Propose a points-of-interest associative pattern mining framework.• Extensive case studies with Flickr data in the area of Queensland, Australia.• First time analyzing geo-tagged photos in this study region.With the development of web technique and social network sites human now can produce information, share with others online easily. Photo-sharing website, Flickr, stores huge number of photos where people upload and share their pictures. This research proposes a framework that is used to extract associative points-of-interest patterns from geo-tagged photos in Queensland, Australia, a popular tourist destination hosting the great Barrier Reef and tropical rain forest. This framework combines two popular data mining techniques: clustering for points-of-interest detection, and association rules mining for associative points-of-interest patterns. We report interesting experimental results and discuss findings.

Web 2.0 and Web 3.0 technologies serve as a Web-as-participation-architecture with which users are encouraged to add values. With this Web structure, users are able to share user-generated social medias anytime and anywhere, and interactively communicate others. As web-based and mobile-based technologies advance, social medias are increasingly collected beyond the capability of human analysis (Lee & Torpelund-Bruin, 2011). Social networks combine the traditional blog, BBS, e-mail, instant messaging and other forms, and also add a variety of supporting applications. A key element of the technology is that it allows people to create, share, collaborate and communicate. The nature of this technology makes it easy for people to create and publish or communicate their work to either a selected group of people or to a much wider audience, or to the world. Photo sharing platform is one kind of social network platforms. Benefitting from the development of Web album, people can easily share their photos on websites. Flickr (http://www.flickr.com) is one of the most popular photo sharing websites which is a great resource for photography enthusiasts and increasingly for travelers. Flickr has recently launched its own service for adding latitude and longitude information to a picture and provides the tool that allows a user to display pictures on online maps like OpenStreetMap (http://www.openstreetmap.org). In addition, many photos are geo-tagged automatically using GPS logs or location aware devices. Therefore, the location and time data associated with photos and other related text tags can be considered as useful geographically annotated materials on the Web. Eventually, they generate huge amount of tourist trails and detailed trajectories of what sites and in what order tourists visit. To extract tourists’ photo-taking pattern is of significant importance to obtain the place the tourists visit and take photos for tourism related organizations. Some research has been conducted with Flick datasets (Crandall et al., 2009, Kennedy et al., 2007, Kisilevich et al., 2010, Rattenbury et al., 2007a, Rattenbury et al., 2007b, Yang et al., 2011 and Zheng et al., 2012), however most work focuses on finding attractive areas points-of-interest (PoI) (Crandall et al., 2009, Kisilevich et al., 2010 and Zheng et al., 2012) and folksonomy based social tagging (Kennedy et al., 2007, Rattenbury et al., 2007a and Rattenbury et al., 2007b). Zheng et al. (2012) investigate regions of attractions that are similar to PoI, and use them for route analysis. Kisilevich et al. (2010) modify DBSCAN to find out PoI from Flickr photos. It is adaptive and flexible but only limited to PoI mining. Crandall et al. (2009) use classification methods to analyze Flickr geo-tagged photos. Kennedy et al. (2007) use the concept of representative tags and tag-driven approach to extract place and event semantics. Rattenbury et al., 2007a and Rattenbury et al., 2007b produce similar research and investigate ways to extract place and event semantics from folksonomy. On the other hand, some recent studies pay more attentions to route recommendation systems (Kurashima et al., 2010, Lu et al., 2010, Okuyama and Yanai, 2011 and Shi et al., 2011). In these studies, geo-tagged photos are modeled as a sequence of location points, and travel sequences are then found. Shi et al. (2011) combine user-landmark preference and category-based landmark similarity to provide personalized landmark recommendation. Lu et al. (2010) study one similar work where users are able to specify personal preference in the travel route planning. Kurashima et al. (2010) integrate topic models into Markov models to provide travel route recommendations whilst Okuyama et al. (2011) extract a travel plan using trip models represented by the order sequences of tourist places. These approaches do not reveal associative PoI patterns, but mainly focus on travel route recommendations. None of the previous work reveals associative PoI patterns exposing positive PoI relations. This paper concentrates on the identification of PoI and associative relationship mining among PoI. In this paper, we focus more on structural and practical aspects of Flickr mining rather than technical and algorithmic aspects of it. Main contributions of the paper are in two folds. First, this paper proposes a mining framework for PoI associations. It first finds PoI using clustering and applies association rules mining to detect associative PoI patterns. Second, we analyze geo-tagged photos from Flickr for Queensland Australia, the second largest state hosting the Great Barrier Reef and world heritage rainforest. We report interesting experimental results and discuss findings. The rest of paper is organized as follows. Section 2 briefly outlines preliminaries on clustering and association rules mining. Section 3 introduces our framework for associative PoI mining. Section 4 reports PoI clustering results and Section 5 further explores PoI association mining. Section 6 concludes with final remarks.

Geospatial concentrations convey significant geospatial implications. For instance, crime hot spots are main targets for police patrols, and disease outbreaks are key focuses for epidemiologists. Similarly, people’s photo-taking PoI are main interesting places for local businesses, policy makers, and travellers. Understanding people’ photo-taking behavior is of great importance as more people take photos and upload to photo-sharing websites to communicate with their online social friends while they are travelling. We propose a PoI clustering and associative pattern mining framework for geo-tagged photos from Flickr with a combination of two popular data mining techniques: clustering and association rules mining. This study particularly focuses on Queensland, one of the hottest tourist destinations in Australia. We undertook various case studies in the study region, and found a large set of interesting patterns. These findings demonstrate the usefulness and applicability of the proposed framework. Extensions of the study region to entire Australia are an immediate future study. Adding temporal dimension to the dataset will reveal spatio-temporal patterns. Since geo-tagged photos capture trails of travellers, mining travel patterns including sequence analysis and trajectory analysis is the next step to investigate. Adding folksonomy (social tagging) data to geo-tagged photos provides a richer set of information. Combining all these geospatial, temporal, and textual information together to mine various types of patterns is a challenging future task.