| Geo-tagged photos of users on social media site such as Flickr provide plentiful location-based data. This data provide a wealth of information about user behavior and their potential is increasing, as it becomes ever-more common for images to be associated with location information in the form of geo-tags.Photos and videos that constitute a huge proportion of the Web information, and are added or exchanged every second, have provided new research opportunities and challenges for multimedia, data mining, and geographic related research and applications. This multi-media data such as photos not only contain textual information like tags, title, notes and description but are also tagged with temporal context, i.e., time at which the photo was taken, and spatial context, i.e., the location (in terms of latitude and longitude) where the photo was taken.In last few years, based on the simple assumption that tourist attractions (landmarks or tourist locations) are those places that are often photographed, it has been a hot research topic to explore common wisdom in photo sharing community to discover landmarks and for travel recommendations. Collection of tourists’geo-tagged photos are assumed as a sequence of visited locations to build travelling histories of users and different methods are proposed to find the popular locations or representative travel sequences to address travel-ling related queries. Furthermore, this vast amount of data provides a unique opportunity to explore ways in which users engage and perceive geographical areas that helps to under-stand the attitude, attention, and interest of a person or community in these geographical areas.For a tourist, before traveling to an unfamiliar city, the most important preparation is planning the trip. Without any prior knowledge, tourist must either rely on travel books, personal travel blogs, or a combination of online resources and services such as travel guides, and human intelligence to piece together an itinerary. It is difficult, time consuming and painstaking to find out the locations worth to visit and to figure out the order in which they are to be visited. The enormous amount of users generated and published contents in the form of social media that exhibit their travelling experiences provide a great opportunity to build a recommendation system for travelling assistance with three features,(1) collective wisdom,(2) personalization and (3) context-awareness.Collective wisdom:A social approach (e.g., ask people who know about the area to be ex-plored) adopted by inexperience travellers in a new area, can provide more up-to-date and accurate information but it takes time for travellers to digest and put together collected information for use. Considering users supplied geo-tagged photos as source of social trav-elling experiences, we can explore collective (social) wisdom to (1) compile tourist locations in a city by grouping photos using their associated geo-tags and (2) to determine usage patterns of locations among tourists in different contexts.Personalization:Making a simple assumption that users have a specific travel preferences and therefore visit locations that have similar features and taking a photo at a visited location is a sign that the user likes that location, we can get users specific travel preferences by building users similarity model from their travel histories (exposed by their contributed photos on sharing sites), and use it to recommend personalized tourist locations to plan trips in different and unknown region.Context-awareness:Tourists preferences in terms of visiting a location or multiple locations in a certain sequence could be affected by their current spatial, temporal, and environmental contexts. From geo-tagged photos, temporal information (photo taken time) and spatial information (geo-tags that describe the locations where photos were taken) can be used to estimate the usage patterns of tourist locations in different temporal contexts. Moreover, various online weather Web services provide not only the current weather condition of a particular geographical area but also offer its historical weather data. Therefore, current weather conditions provided by these services can be used to augment the query with current weather context, and historical weather data can be used to filter the relevant tourist locations to address the weather context-driven query.The objective of this thesis is to study the data mining techniques in the social media setting to investigate the problems of travel recommendations. Specifically, we mine useful knowledge from social media and available web resources for the analysis of attractive area and making recommendations to tourists for traveling. Specific contributions of this research effort are described below in the order they are presented and discussed in the thesis:1. Analysis of Attractive areas:A simple but scalable method is proposed for the analysis of attractive areas using geo-tagged photos. For the analysis of attractive areas, we illustrate (a) how to group photos from user supplied photos collections using their associated geo-tags to find tourist locations, and (b) how to aggregate clustered photos’textual information and enrich with supplementary information provided from Web services (i.e., Google Places) to provide more semantic meaning to aggregated locations. Furthermore, to summarize the locations aggregated from photos and to derive the dynamics of users’interests to these locations; temporal tags annotated to photos are exploited to infer users’visits for profiling locations. Profile of each location provides the information about the users who have visited that location and the history of contexts (i.e., weather and temporal) in which the location has been visited. Note that, we identify the temporal context of a visit by exploiting the time-stamps of photos that were taken during that visit and use this visit time to obtain the weather context of the visit from historical weather dataset retrieved from online weather resources. We show how to synergize disjointed contexts and sparse social contents together with online information sources to enrich primitive contexts and contents with higher levels of semantic meanings, i.e., profiling locations. A conceptual foundation is laid down for the analysis of spatio-temporal data of places (tourist locations) obtained from community contributed geo-tagged photo collections. We use it to provide location-aware tourist information. It can also be utilized by local authorities, service providers, and tourist agencies for building user centric applications and to provide location based services.2. Recommendations of interesting tourist locations:Interestingness (significance) of tourist locations are mined by leveraging the collective wisdom of people from com-munity contributed geo-tagged photos collection to provide a set tourist locations to user that are interesting (significant) and match the user’s current context given a city that is new to that user. Our motivation for this work is twofold:First, a user may benefit from recommendations of tourist locations that are significant to visit when she is traveling to a city for the first time. Second, a promising solution to provide recommendations for a user that are best fit in his current contexts (i.e., weather, temporal). A probability-based approach is used to filter the tourist locations in the target city to meet the active user’s current context.’Popularity’,’significance’or ’interestingness’are subjective terms. Different people have different ideas of defining them. We define a reasonable function, based on user-expertise model, i.e., weighing of tourist locations’significance through the number of user visits to specific location categories, to score tourist location for ranking.3. Personalized travel recommendations:We propose architecture of a system for context-aware personalized landmark recommendation based on geo-tagged photos. The method we propose is designed to be deployed in an application scenario that lever-ages the collective wisdom of people from community contributed geo-tagged photos collection to provide a set of tourist locations that match the user’s interests and current context given a city that is new to that user. We obtain user specific travel preferences from his travel history in one city and use these to recommend tourist locations in another city. Proposed method uses the popularity of tourist locations in different contexts as profile matching criteria to filter the locations according to users’current context and then rank the location in collaborative filtering manner for personalized recommendations.4. Recommendations of interesting travel sequences:Given the locations in a city, a tourist might need to know the interesting (significant) travel sequences among these locations in order to understand an unfamiliar city and plan a trip to visit it. The photos, together with their time-and geo-references, become the digital footprints of photo takers and implicitly document theirs spatio-temporal movements. Therefore, wealth of these enriched online photos can be leverage to retrieve the locations visited by each photographer to build their location histories and the order in which they visited these locations to discover their travel sequences. We suggest an approach to extract semantically annotated significant travel sequences from these collections of geo-tagged photos. The proposed method is able to consider the users’current context while making the tourist sequences recommendations. |
PDF Full Text Request