Geographical Bias on Social Media and Geo-Local Contents System with Mobile Devices

Transcription

1 212 45th Hawaii International Conference on System Sciences Geographical Bias on Social Media and Geo-Local Contents System with Mobile Devices Kazunari Ishida Hiroshima Institute of Technology Abstract This paper examines geographical bias on social media and develops a geo-local contents system with mobile devices. Because of the maturations of GPS devices and application software, vast amount of people generate various types of contents with geographical information. However, information outside of metropolitan areas seems to be scarce because of population, even though, it is necessary to attract tourists to local areas. To clarify the geographical biases, information posting behavior on Foursquare, which is one of the famous location services, was analyzed. Moreover, geographically local contents system was developed with mobile devices in order to support creation of local contents by local community members and to reduce the geographical biases. In order to find what kind of information must be created with the system, information available on the web was collected automatically concerning two areas, i.e., a rural town and a mid-size city. According to the analysis of the collected information, the local community members have to provide more information concerning entertaining places to attract tourists. 1. Introduction A Social media is growing and generating a vast amount of information provided by consumers through the Internet. This paper analyzes geo-tagged information, which is continuously collected from social media such as blogs, Twitter, and Foursquare, in terms of time and place. According to the results, different information posting behaviors are found between business districts and residential areas. In addition, the information is heavily clustered in metropolitan areas. In order to boost information from local areas and to flatten the geographicallybiased distribution of geographical information, geolocal contents systems, which can take geo-tagged pictures and sounds with assistance of geographical map and augmented reality user interfaces, have been developed with mobile devices for people in local communities. Members of a local community, e.g., shopping districts and tourist spots, have strong incentives to provide high quality information to their customers. Hence, the systems are provided to the people in communities so that a huge amount of geolocal content is going to be published on the Internet. 2. Location Information To collect location information posted on Foursquare, Twitter was employed as a sampling mechanism all over the world. The duration of the sampling was from Feb. 17 to Jun 28 in 21. The number of tweets and that of location information were 5,324,981 and 12,361, respectively Geographical Distribution To observe distribution of the information all over the world, it is divided into 24 categories based on every 15 degrees in terms of longitude. Based on the categories, frequent areas are found around America, Asia, and Europe (Fig. 1) (-18:-165) (-15:-135) (-12:-15) (-9:-75) (-6:-45) (-3:-15) (:15) (3:45) (6:75) (9:15) (12:135) (15:165) Figure 1. Geographical Distribution on Longitude /12 $ IEEE DOI 1.119/HICSS

2 (-18:-165) (-15:-135) (-12:-15) (-9:-75) (-6:-45) (-3:-15) (:15) (3:45) (6:75) (9:15) (12:135) (15:165) [75:9] [3:45] [-15:] [-6:-45] [75:9] [6:75] [45:6] [3:45] [15:3] [:15] [-15:] [-3:-15] [-45:-3] [-6:-45] [-75:-6] [-9:-75] Asia 1 (Singapore) (Fig. 5). Europe is categorized into morning peak type (Fig. 6) ::-1:: 1::-2:: 2::-3:: 3::-4:: 4::-5:: 5::-6:: 6::-7:: 7::-8:: 8::-9:: 9::-1:: 1::-11:: 11::-12:: 12::-13:: 13::-14:: 14::-15:: 15::-16:: 16::-17:: 17::-18:: 18::-19:: 19::-2:: 2::-21:: 21::-22:: 22::-23:: 23::-24:: Figure 2. Geographical Distribution on the Earth Figure 3. Time Distribution (GMT) To specify a precise area of the distribution, each category is divided into 12 categories based on every 15 degrees in terms of latitude (Fig. 2). Based on the categories, nine frequent areas are found: western America (lon. [-12:-15], lat. [3:45], time diff. = - 8); central America (lon. [-15:-9], lat. [3:45], time diff. = -7); eastern America 1 (lon. [-9:-75], lat. [3:45], time diff. = -6); eastern America 2 (lon. [- 75:-6], lat. [3:45], time diff. = -5); Europe (lon. [:15], lat. [45:6], time diff. = ); south Asia 1 (lon. [9:15], lat. [:15], time diff. = +6, Singapore); south Asia 2 (lon. [:15], lat. [45:6], time diff. = +7, Indonesia); east Asia 1 (lon. [12:135], lat. [3:45], time diff. = +8, China, South Korea, and Japan); and east Asia 2 (lon. [135:15], lat. [3:45], time diff. = +9, Japan) Time Distribution To observe time distribution of the information throughout the world, it is divided into 24 categories based on global mean time (GMT). Fig. 3 shows that there is almost a constant amount of posting. To clarify the characteristics of hours of each area, the information is divided into 24 categories. In each category, a trajectory of the number of posting is illustrated based on a local time. With respect to the nine areas in Fig. 2, there are three types of peaks on the trajectories--night, noon, and morning. The most frequent type is night peak (Fig. 4). Central America, east America 1, South Asia 2 (Indonesia), east Asia 1 (China, South Korea, and Japan), and east Asia 2 (Japan) are categorized into the night peak type. The noon peak type consists of west America and south ::-1:: 2::-3:: 4::-5:: 6::-7:: 8::-9:: 1::-11:: 12::-13:: 14::-15:: 16::-17:: 18::-19:: 2::-21:: 22::-23:: Figure 4. Peaks around Night ::-1:: 2::-3:: 4::-5:: 6::-7:: 8::-9:: 1::-11:: 12::-13:: 14::-15:: 16::-17:: 18::-19:: 2::-21:: 22::-23:: Figure 5. Peaks around Noon (-15:-9)[3:45]- 7 (-9:-75)[3:45]-6 (-75:-6)[3:45]-5 (15:12)[- 15:]+7 (12:135)[3:45]+ 8 (135:15)[3:45]+ 9 (-12:- 15)[3:45]-8 (9:15)[:15]

3 ::-1:: 2::-3:: 4::-5:: 6::-7:: 8::-9:: 1::-11:: 12::-13:: 14::-15:: 16::-17:: 18::-19:: 2::-21:: 22::-23:: (:15)[45:6] Figure 6. Peaks around Morning Figure 7. Address Distribution 2.3. Address Distribution In order to clarify quality and pervasiveness of location information, another dataset was extracted from the entire dataset, which has been collected continuously for this research. The sub dataset contains points of interest (POI) in a country that were submitted on Twitter through Foursquare. The POI can be categorized into the administrative divisions of the country, based on the geospatial information provided by the authority of the country. The dataset contains 3817 POIs and the duration is from Feb. 27, 21 to Jan. 3, 211. Fig. 7 shows extremely biased POI distribution among the administrative divisions. The most crowded region has 1,594 POIs, which is approximately 41.8 percent of all POIs. The region includes 444 POIs concerning the railway station, which is approximately 28 percent. The dataset tells us that the crowded region tends to have a high percentage of POIs related to public transportation. The dataset is also evaluated in terms of integrity of address information written by individuals. The information is checked with the proper address information provided by the geospatial information authority of the country. Based on the observation of each POI, POIs of public places, such as a train station, have the correct address information; however, other types of POI, such as a privately owned restaurant, have incorrect or no address information. According to the result, the correct information comprises approximately 44 percent. To sum these results, POI information generated by individuals on social media has a huge regional bias and lacks correct address information. 3. Information Flow in Local Community In order to reduce a huge bias of location information and incorrect information, the circular information flow in local community is defined in this section (Fig. 8). On one hand, ordinal location information services, provided by well-known Internet companies, focus on each individual or small informal group, without attention to a local community member such as a tourist association, who has high quality information and knowledge concerning tourist spots. The result of activities on the services leads to a huge bias and incorrect information, as shown in section II C and Fig. 7. On the other hand, the circular information flow in Fig. 8 contains local community members in addition to individual or small tourist groups. The members in this kind of community have a strong incentive or motivation to generate geo-local information to advertise their shopping districts and tourist spots for attracting customers. Hence, the systems are provided to the people in communities so that high quality geo-local contents are going to be published on the Internet. The information flow consists of three phases. In the first phase, employing the mobile device, community members of a local tourism association take pictures of their shops and tourist spots with voices and sounds to advertise these places. The device has all the fundamental information, i.e., name, address, telephone number, longitude, latitude, and so forth. A user of the device can easily browse the information on two types of user interfaces, i.e., geographical map or augmented reality (AR). Based 1792

4 on the fundamental information and the user interfaces, a user with the device can organize the information in terms of shops and spots. The collected pictures, voices, and sounds are transferred from the mobile device to a computer through a Wi- Fi network. The transferred information is edited by the members of the local community to provide local information on mobile devices for tourists in the second phase. system, which is composed of two parts--a mobile device and PC server applications (Fig. 9). The smart phone application provides functions to take pictures, voice memos, and traces with geographical information sensed by the GPS module in the phone. It also provides local map and address browsing functions without a wired and wireless connection because it has map and address databases on the device. The databases are generated by means of a PC server application and downloaded through a Wi-Fi communication channel between a PC server and a smart phone. A PC server provides editing and managing functions concerning map, POI, and traces. The edited geo-local information, including POIs, is also downloaded on smart phones to find places to enjoy travel. The application on a PC server has functions to publish geo-local information on the Internet to advertize shopping districts and tourist spots. Figure 8. Circular Information Flow in Local Community Collecting contents Navigation Browsing map and address at current position POI Data: recording pictures, voices, texts with geographical information Track Data: Recording streets Retrieving local map and POI data from server Shopping district and tourist spot Editing and Managing POI Data Editing and Managing Track Data In the second phase, tourists have a device at the hotel where they are staying. They can search restaurants or other tourist spots on the device and browse the location and other information on a geographical map or AR user interface. After arriving at these spots, they may enter voice comments with pictures if they like. When they arrive at their hotel, the device is returned. The device automatically uploads the information concerning the tourists comments to the specialized computer at the hotel through a Wi-Fi connection. The device also automatically downloads updated fundamental information organized by the local community members. In the third phase, community members analyze the information provided by tourists to find tourists needs. They try to find excellent spots and improve required spots in terms of tourists point of view. Based on the provided information, they improve their services and start the first phase again to update the fundamental information. 4. Geo-Local Contents System To facilitate the circular information flow in local community, I developed a geo-local information Bluetooth connection Sharing maps, POI data, and track data Wi-Fi Router Broadcasting local place contents (Internet connection required) Figure 9. Overview PC Server Internet 4.1. Geo-Local Information System on Mobile Device An application software on a mobile device has been developed in order to gather geo-local contents and browse map and address information without any map application API, e.g., Google Maps API, because the software tends to be deployed outside of Internet enabled areas, such as rural beautiful beaches or mountains for sightseeing. Various maps on the Internet can be employed in order to create map databases of the software. Fig. 1 shows browsing a map on a mobile device, generated from OpenStreetMap (OSM). 1793

5 The software displays a map so that the current user position is centered on it. The map can be moved by dragging on the touch screen. On the top of the screen, current address information and zoom level of the map are displayed. On the bottom of the screen, there are five buttons from right to left: zoom-in and zoom-out of map, toggle of POI, enabling GPS, and taking pictures and voice memos with location information. Figure 11. Collecting Contents Figure 1. Browsing Offline Map and POI Figure 12. AR Interface When toggling POI, POIs are displayed as purple dots on the map in Fig. 1. When one of the POI is focused on the center of the map, a thumbnail of the picture and additional text information will be displayed on the screen. When we touch the thumbnail, we can browse detailed information and add voice comments on the POI. Moreover, when enabling GPS, your current position is always displayed on the center of the map. Hence, when you go to one of the POIs, the information is automatically displayed for your navigation. When you push the camera button, a camera function is invoked in order to take a picture with current location information. After taking a picture, you can add voice memos if needed. Voice memos can be convenient to memorize information on a small mobile device without a keyboard (Fig. 11). The taken pictures are also browsed on live view camera, i.e. AR interface, based on the point where the pictures are taken (Fig. 12). When one of the pictures is tapped, the detail information of each picture will be displayed as shown in Fig Editing and Publishing System on PC Server Geo-local contents gathered by the software on mobile devices are transferred to a PC server in order to edit and manage the contents based on map interface for arrangement of the contents without any map application API, e.g., Google Maps API, because the software tends to be deployed outside of Internet enabled areas. Various maps on the Internet can be employed in order to create map databases of the software. Fig. 13 shows browsing a map with a web browser, generated from OpenStreetMap (OSM). The system is able to be deployed on LAMP or WAMP stack on a PC server and is operated through a web browser. Fig. 13 shows the Hiroshima Peace Memorial Museum and park. When you click the picture link on the left top of the map in the figure, you can browse pictures around the center of the map. Fig. 14 shows an example of a list of pictures around the museum and park. The information concerning each picture consists of the picture itself, date, longitude, latitude, altitude, and description. 1794

6 You can search pictures with a keyword based on the description of each picture. Fig. 14 also shows an example of a search result with a keyword meaning fountain. You can also adjust the longitude and latitude of the POI because they might not be precise due to the spatial condition for the GPS device. The description you edit will be displayed on mobile devices as POI. Figure 15. Browsing and Editting POI Figure 13. Browsing Map Figure 14. Listing and Searching POI When you push an edit button of a picture, you can edit the description of the picture. Fig. 15 shows an example of browsing and editing POI. The picture and a map where the picture was taken are displayed with a text area to edit the description of the POI. 5. Current Social Media as Information Source Based on the systems, the circular information flow in Fig. 8 will be implemented in a local community. An experiment is planned to deploy the developed systems with a local tourism association. The association maintains a tourist spot with a beautiful beach in a rural town on the west side of Japan. The members of the association cooperate by collecting and organizing local information. To begin the data collection and organization, we have to determine what kind of information is on the web and whether enough exists concerning the rural town. Our aim is to add valid and useful information instead of replacing existing information. To find out what information exists, we need to collect data concerning the rural town, called Area-1. Another tourist area is then selected for comparison. This area, called Area-2, is in a mid-size city and has two world heritage sites. The distance between the two tourist areas is approximately 6 km. In the dataset of section II-C, the numbers of POI s for Area-1 and Area-2 are 7 and 26, respectively. On one hand, POIs of Area-1 consist of two restaurants, a library, a museum, an amusement place, a corporation, and a driver license center. On the other hand, POIs of Area-2 consist of nine public transportation stations, seven shops, five restaurants, 1795

7 a company, a convention center, a bridge, a power substation, and a seaport. In other words, Area-1 has no POI of public transportation; however, Area-2 has 9 POIs, which is approximately 34.6% of POIs in Area-2. As mentioned in II-C, crowded regions tend to have a high percentage of POIs related to public transportation. According to statistics published by local and central governments [5] [6] [7], Area-1 and Area-2 are described in terms of three aspects: total land space (958 km2 : 1,395 km : 59.28), number of tourists (1,988,594 : 4,635, 3.2 : 69.98), and population (9,82 : 1,269, : 93.33), respectively. The differences are summarized in Fig. 16. There is a significant difference between the two areas, especially in terms of population. Around the two areas, six categories of tourist spot information are collected from the web. These categories are: (1) restaurant, (2) shop, (3) entertaining place, (4) accommodation, (5) gas station, and (6) police. The number of spots in Area-1 and Area-2 are 215 and 442, respectively. Figure 17 summarizes the percentages of information among six categories for the two areas. According to the figure, Area-1 has many restaurants compared to Area-2. On the flipside, Area-2 has many entertaining places compared to Area-1. Based on the test of independence, the dependence between the areas and categories is rejected on a critical point of.1. The result of the test suggests that Area-1 has plenty of restaurants; however, it needs to facilitate entertaining places, compared to Area-2. According to the result, the local community members have to provide more information concerning entertaining places to attract tourists Total land space Number of tourists Population Figure 16. Comparison between Two Areas Area-1 Area Restaurant Shop 6. Conclusion Figure 17. Distribution of Spot Categories Geographical information, provided by social media such as Foursquare, is quite sparse in local areas, even though it is heavily clustered in metropolitan areas. In order to generate a high content in local areas, the circular information flow in a local community is proposed in this paper. To facilitate the flow, a geo-local contents system is developed with mobile devices for people in local communities. Members of a local community, e.g., shopping districts and tourist spots, have strong incentives to provide high quality information to their customers. Hence, the system for editing, publishing, and managing local content has been developed for these communities so that they will be published on the Internet. To begin collecting and organizing local content, information available on the web is collected automatically concerning two areas, i.e., a rural town and a mid-size city. According to the analysis of the collected information, the local community members have to provide more information concerning entertaining places to attract tourists. 7. References Entertaining place Accommodation Gas station Police Area-1 Area-2 [1]Foursquare [2]Geoospatial Information Authority of Japan, Maps & Geospatial Information, Ministry of Land, Infrastructure, Transport and Tourism, [3]OpenStreetMap [4]Mordechai (Muki) Haklay, Patrick Weber, "OpenStreetMap: User-Generated Street Maps," IEEE Pervasive Computing, vol. 7, no. 4, pp , October-December, 28. [5]National Statistics in Japan, [6]Tourism Statistics in Shimane, [7]Tourism Statistics in Hiroshima,