Search

Large collections of geotagged photos publicly accessible through photo-sharing web sites such as Flickr or Panoramio, present an opportunity for the entire Internet community to access the wealth of visual and textual data stored on photos. At the same time, it presents the challenge of how to efficiently and effectively turn this data into information about locations of interest and people's movement preferences.Geotagged photos can be viewed in a dual way: as independent spatio-temporal events and as trajectories of people in the geographical space. These two views imply, however, two different approaches to an analysis that will yield different kinds of valuable knowledge about places as well as people. In this thesis, geotagged photos are therefore primarily regarded in the context of (event-based) movement rather than multimedia content and present several exploratory and analytical techniques corresponding to event-based and trajectory-based movement analysis.Moreover, geotagged photos are contextually rich data that combine geographical information about the places where photos were taken with textual information. This information includes for the most part titles and tags attached to photos by owners of these photos as well as comments that can be written by other users. This unique combination is what makes the geotagged photos special for analysis of people's movement and behavior in contrast to the general purpose spatio-temporal data like GPS-based records. Analysis of spatio-temporal data is challenging since it always requires combining different analytical methods like geocomputation and geographical analytics, algorithms like aggregation and clustering, technologies in which the data is appropriately visualized on the map and scalable with the amount of the data.Taking into consideration the above mentioned aspects, we propose in this thesis a systematic approach to an analysis of people's movement and events using geotagged photos. We make contributions in four main research areas: (1) geovisual analytics - interactive exploration of patterns of people's trajectories and interesting routes that they take during photographing; (2) data mining - development of techniques and algorithms for discovery of attractive areas and finding frequent sequential patterns of people's movement; (3) text mining and computational linguistics - development of approaches for extraction and analysis of comments that people write for photos, and (4) systems engineering - development of a GIS-based tool to facilitate handling of geotagged photos. Moreover, while some of the approaches were targeted towards the imaginary domain expert, some approaches proposed in this thesis were deliberately usercentered in order to demonstrate how the proposed approaches can be used in user-centered scenarios if implemented and delivered by a service provider. Finally, the research demonstrated in this thesis will be useful for practical reasons, such as future research targeting people's preferences for urban locations, landmarks and corresponding travel itineraries. It may also benefit (research) areas like city promotion and advertising, public safety, tourism, and civic minded activities.

In the Internet age, e-commerce provides customers global reach to a wide variety of products and plays a dominant role in business activity and competition. Competition is especially aggressive in the online travel domain where wholesalers, e.g. brokerage companies, contract through their contract managers with thousands of hotel brands and trade hotel products (usually hotel nights) for travel businesses or end customers. In order to conclude a profitable contract, a contract manager should be able to compare all the particulars of the prospective partner hotel with those of the competing hotels in the target city. Given that the number of contract managers is comparatively small compared to the large number of hotels, the possible knowledge base is limited. Thus, the hotel brokerage companies are only able to bargain with a relatively limited number of hotels, and the contract profitability relies heavily on the contract managers’ expertise and communication skills. In this paper we present a price management decision support system (DSS) for hotel brokers that allows analysis of hotel prices using spatial and non-spatial characteristics, estimation of the objective relative hotel prices, and determination of the profitability of the existing or future contracts. We built our system using free and open source tools including geographic information system and data mining frameworks that allow companies with limited money resources or manpower to implement such a prototype. We show the effectiveness of our tool by covering all the major components of the DSS such as data selection and integration, model management and user interface. We demonstrate our tool on the area of Barcelona, Spain using a real data of 168 hotels provided by one of the travel service providers.

Due to the pervasiveness of positioning technology combined with the pro-liferation of social-oriented web sites, community-contributed spatio-temporal data of people's historical positions are available today in large amounts. The analysis of these data is valuable to scientists and can pro-vide important information about people's behavior, their movement, geo-graphical places, and events. In this paper, we develop a conceptual framework and outline a methodology that allows to analyze events and places using geotagged photo collections shared by people from many countries. These data are often semantically annotated by titles and tags that are useful for learning facts about the geographical places and for de-tecting events occurring in these places. The knowledge obtained through our analysis carries an additional benefit. For example, it may also be uti-lized by local authorities, service providers, tourist agencies, in sociologi-cal and anthropological studies or for building user centric applications like tour recommender systems. We provide a conceptual foundation for the analysis of spatio-temporal data of places visited by people worldwide using community contributed geotagged photo collections. First, we define several types of spatio-temporal clusters of people's visits. Second, we dis-cuss methods that can be used for analysis of these clusters. Third, we of-fer an analysis of tourist activities in Switzerland based on a case study.

The trend toward the growing multilinguality of the Internet requires text summarization techniques that work equally well in multiple languages. Only some of the automated summarization methods proposed in the literature, however, can be defined as “languageindependent”, as they are not based on any morphological analysis of the summarized text. In this paper, we perform an in-depth comparative analysis of language-independent sentence scoring methods for extractive single-document summarization. We evaluate 15 published summarization methods proposed in the literature and 16 methods introduced in (Litvak et al., 2010). The evaluation is performed on English and Hebrew corpora. The results suggest that the performance ranking of the compared methods is quite similar in both languages. The top ten bilingual scoring methods include six methods introduced in (Litvak et al., 2010).

This article presents a geovisual analytics approach to discovering people's preferences for landmarks and movement patterns from photos posted on the Flickr website. The approach combines an exploratory spatio-temporal analysis of geographic coordinates and dates representing locations and time of taking photos with basic thematic information available through the Google Maps Web mapping service, and interpretation of the analyzed area. The article describes data aggregation and filtering techniques to reduce the size of the dataset and focuses on information addressing research questions. The results of analysis for the Seattle metropolitan area help to distinguish between sites that are occasionally popular among the photographers and can be considered as potential attractions from sites that are regularly visited and already known as city landmarks. The analysis of photographers' movements across the metropolitan area shows that most photographers' itineraries are short and highly localized.

The key component of Social Networking Sites (SNS), gaining increasing popularity among Internet users, is the user profile, which plays a role of a self-advertisement in the aggregated form. While computer scientists investigate privacy implications of information disclosure, social scientists test or generate social or behavioral hypotheses based on the information provided by users in their profiles. Statistical analysis of the SNS phenomenon often is performed using only a very small sample of information extracted from a particular SNS or by interviewing students from a particular university. In this paper, we apply classification algorithm to a large-scale SNS dataset obtained from more than 10 million public profiles with 50 different attributes extracted from one of the largest dating sites in the Russian segment of the Internet. In particular we build gender classification models for the residents of the most active countries, and investigate the particular differences between genders in one country and the differences between the same-genders in different countries. The preliminary results are reported in this paper. To the best of our knowledge, this is the first attempt to conduct a large-scale analysis of SNS profiles and compare gender differences on a country level.

Various frameworks are being developed to support geospatial data processing and exploration. However, these frameworks are insufficient due to their inability to support rapid development and reuse of existing geospatial operations. To cope with the difficulties that arise, more and more Web-based solutions are being implemented incorporating Web services and Open Geospatial Consortium (OGC)a standards for data interchange and distributed geo-processing. Although these solutions often target specific problems, they rarely can be extended by a third party. As a result, analysts are frequently required to implement additional software in order to complete their exploration process. In this paper, we present a unified, extensible framework that can solve generic spatio-temporal analysis tasks. Specifically, we present the prototype of a new framework that allows quick hypothesis definition and testing. The proposed framework, which we term GEO-SPADE, uses Google Earth as a primary visualization platform and data interchange system. Pluggable components can easily be integrated into the framework. We demonstrate the feasibility of the GEO-SPADE framework by following the scenario of analysis of travel sequences near Funchal, Madeira Island.

User generated content in the form of customer reviews, feedbacks and comments plays an important role in all types of Internet services and activities like news, shopping, forums and blogs. Therefore, the analysis of user opinions is potentially beneficial for the understanding of user attitudes or the improvement of various Internet services. In this paper, we propose a practical unsupervised approach to improve user experience when exploring photo collections by using opinions and sentiments expressed in user comments on the uploaded photos. While most existing techniques concentrate on binary (negative or positive) opinion orientation, we use a real-valued scale for modeling opinion and sentiment strengths. We extract two types of sentiments: opinions that relate to the photo quality and general sentiments targeted towards objects depicted on the photo. Our approach combines linguistic features for part of speech tagging, traditional statistical methods for modeling word importance in the photo comment corpora (in a realvalued scale), and a predefined sentiment lexicon for detecting negative and positive opinion orientation. In addition, a semiautomatic photo feature detection method is applied and a set of syntactic patterns is introduced to resolve opinion references. We implemented a prototype system that incorporates the proposed approach and evaluates it on several regions in the World using real data extracted from Flickr.

Space- and time-referenced data published on the Web by general people can be viewed in a dual way: as independent spatio-temporal events and as trajectories of people in the geographical space. These two views suppose different approaches to the analysis, which can yield different kinds of valuable knowledge about places and about people. We present several analysis methods corresponding to these two views. The methods are suited to the large amounts of the data.