《电子商务 E-business》阅读文献：Photo-Based User Profiling for Tourism Recommender Systems

Photo-Based User profiling for Tourism Recommender Systems Helmut Berger, Michaela Denk, Michael Dittenbach', Andreas Pesenhoferl, and Dieter Merkl2 E-Commerce Competence Center-EC3 Donau-City-StraBe 1, A-1220 Wien, Austria andreas. pesenhofer/ @ec3.at Institut fur Softwaretechnik und Interaktive Systeme, Technische Universitat wien FavoritenstraBe 9-11/188, A-1040 Wien, Austria dieter. merkl@ec. tuwien, ac, at Abstract. The World Wide Web has become an important source of in- formation for tourists planning their vacation. So, tourism recommender systems supporting users in their decision making process by suggesting suitable holiday destinations or travel packages based on user profi are an area of vivid research. Since a picture paints a thousand words we have conducted an online survey revealing significant dependencies between tourism-related photographs and tourist types. The results of the survey are packaged in a Web-based tourist profiling tool. It is now possible to generate a user profile in an enjoyable way by simply selecting photos without enduring lengthy form-based self assessments 1 Introduction Photographs bring moments back to life- be they very personal or moments shared by many. Assuming you are interested in relaxing and sunbathing in warm places with lots of sun, sand and ocean, you will have taken, without much doubt, pictures of sunny and sandy beaches. Conversely, if your primary emphasis is to remain active while on vacation, you may engage in your favorite sports and so take snapshots of your magic moments. All these moments, and thus tourism activities, can be categorized according to some typology of tourists and, in turn, it is possible to identify the relationship between these types and tourist activities 4. Grounded on this relationship we postulate the hypothesis that preferences for particular tourism-related photographs can be used to derive a tourist's type, and so, generate a profile of the user's likings. Currently, the process of creating such profiles can be a rather annoying, time-consuming and cumbersome task. however. intelligent services such as tourism recommender systems heavily rely on personal user profiles in addition to explicitly expressed needs and constraints. These systems focus on recommending destinations and product bundles tailored to the users'needs in order to support their decision

Photo-Based User Profiling for Tourism Recommender Systems Helmut Berger1, Michaela Denk1, Michael Dittenbach1, Andreas Pesenhofer1, and Dieter Merkl2 1 E-Commerce Competence Center–EC3, Donau-City-Straße 1, A–1220 Wien, Austria {helmut.berger,michaela.denk,michael.dittenbach, andreas.pesenhofer}@ec3.at 2 Institut f¨ur Softwaretechnik und Interaktive Systeme, Technische Universit¨at Wien, Favoritenstraße 9–11/188, A–1040 Wien, Austria dieter.merkl@ec.tuwien.ac.at Abstract. The World Wide Web has become an important source of in￾formation for tourists planning their vacation. So, tourism recommender systems supporting users in their decision making process by suggesting suitable holiday destinations or travel packages based on user profiles are an area of vivid research. Since a picture paints a thousand words we have conducted an online survey revealing significant dependencies between tourism-related photographs and tourist types. The results of the survey are packaged in a Web-based tourist profiling tool. It is now possible to generate a user profile in an enjoyable way by simply selecting photos without enduring lengthy form-based self assessments. 1 Introduction Photographs bring moments back to life – be they very personal or moments shared by many. Assuming you are interested in relaxing and sunbathing in warm places with lots of sun, sand and ocean, you will have taken, without much doubt, pictures of sunny and sandy beaches. Conversely, if your primary emphasis is to remain active while on vacation, you may engage in your favorite sports and so take snapshots of your magic moments. All these moments, and thus tourism activities, can be categorized according to some typology of tourists and, in turn, it is possible to identify the relationship between these types and tourist activities [4]. Grounded on this relationship we postulate the hypothesis that preferences for particular tourism-related photographs can be used to derive a tourist’s type, and so, generate a profile of the user’s likings. Currently, the process of creating such profiles can be a rather annoying, time-consuming and cumbersome task. However, intelligent services such as tourism recommender systems heavily rely on personal user profiles in addition to explicitly expressed needs and constraints. These systems focus on recommending destinations and product bundles tailored to the users’ needs in order to support their decision G. Psaila and R. Wagner (Eds.): EC-Web 2007, LNCS 4655, pp. 46–55, 2007. c Springer-Verlag Berlin Heidelberg 2007

Photo-Based User Profiling for Tourism Recommender Systems making process [1, 2, 8. QUite frequently, tourism recommender systems need to deal with first-time users, which implies that such systems lack purchase histories and face the cold-start problem 9. Some systems tackle this problem by request ing the user to answer a predefined set of questions. However, these might be misunderstood or simply remain unanswered 5. Such non-adaptive approaches are problematic, since poorly assembled user profiles reduce the quality of rec- ommendations, and consequently, negatively effect the acceptance and success of tourism recommender systems. a different line for user preference elicitation is taken in 7, where profiles for new users are generated based on Likert-scale ratings of products. The new user is required to assess her likings until suffi- cient overlap to profiles of known users can be derived. This is feasible in an application setting where commodities are being sold. In tourism, however, the constraints are different since the products are generally rather expensive and annual leave is limited In order to prove our hypothesis, we have conducted an online survey revealing gnificant dependencies between tourism-related photographs and tourist types Our results show that we can take advantage of this relationship and propose a profiling technique based on photograph selection, which minimizes the efforts for users formalizing their likings and get them as quickly as possible fun part. The results of the survey are further packaged in a Web-based tourist The remainder of this paper is organized as follows In Section 2 we present our online survey and some basic facts regarding the respondents Sections 3 and 4 contain our findings from the survey regarding motivating factors and significant photographs for various tourist types. In Section 5 we show the tourist type profiler developed based on the survey results. Finally, Section 6 gives some conclusions 2 The Online Survey To investigate whether tourist's preferences can be derived from tourism-related photographs we conducted a survey. An online questionnaire was made public in July 2006 on a Web portal. This questionnaire consisted of three parts whereof the first part aimed at obtaining personal and demographic data of the partici pants. These were age group, gender, marital status, number of children, highest level of education, and whether they live in a city or town The second part was designed to capture the tourism preferences of the par- ticipants. They were asked to select from a set of 17 tourist types based on the tourist typology proposed by Yiannakis and Gibson 10. The tourist types were described in terms of statements such as"interested in relaxing and sunbathing in warm places with lots of sun, sand and ocean"or"mostly interested in meet- ing the local people, trying the food and speaking the language"whereof the first description corresponds to the tourist type referred to as the Sun Lover and the latter to the Anthropologist. Note that we refrained from providing the actual la bels of the tourist types presuming that participants might be biased by these Additionally, we have defined four age groups, viz. less than 20, 21 to 40, 41 to 60

Photo-Based User Profiling for Tourism Recommender Systems 47 making process [1,2,8]. Quite frequently, tourism recommender systems need to deal with first-time users, which implies that such systems lack purchase histories and face the cold-start problem [9]. Some systems tackle this problem by request￾ing the user to answer a predefined set of questions. However, these might be misunderstood or simply remain unanswered [5]. Such non-adaptive approaches are problematic, since poorly assembled user profiles reduce the quality of rec￾ommendations, and consequently, negatively effect the acceptance and success of tourism recommender systems. A different line for user preference elicitation is taken in [7], where profiles for new users are generated based on Likert-scale ratings of products. The new user is required to assess her likings until suffi- cient overlap to profiles of known users can be derived. This is feasible in an application setting where commodities are being sold. In tourism, however, the constraints are different since the products are generally rather expensive and annual leave is limited. In order to prove our hypothesis, we have conducted an online survey revealing significant dependencies between tourism-related photographs and tourist types. Our results show that we can take advantage of this relationship and propose a profiling technique based on photograph selection, which minimizes the efforts for users formalizing their likings and get them as quickly as possible to the fun part. The results of the survey are further packaged in a Web-based tourist profiling tool. The remainder of this paper is organized as follows. In Section 2 we present our online survey and some basic facts regarding the respondents. Sections 3 and 4 contain our findings from the survey regarding motivating factors and significant photographsfor various tourist types. In Section 5 we show the tourist type profiler developed based on the survey results. Finally, Section 6 gives some conclusions. 2 The Online Survey To investigate whether tourist’s preferences can be derived from tourism-related photographs we conducted a survey. An online questionnaire was made public in July 2006 on a Web portal. This questionnaire consisted of three parts whereof the first part aimed at obtaining personal and demographic data of the partici￾pants. These were age group, gender, marital status, number of children, highest level of education, and whether they live in a city or town. The second part was designed to capture the tourism preferences of the par￾ticipants. They were asked to select from a set of 17 tourist types based on the tourist typology proposed by Yiannakis and Gibson [10]. The tourist types were described in terms of statements such as “interested in relaxing and sunbathing in warm places with lots of sun, sand and ocean” or “mostly interested in meet￾ing the local people, trying the food and speaking the language” whereof the first description corresponds to the tourist type referred to as the Sun Lover and the latter to the Anthropologist. Note that we refrained from providing the actual la￾bels of the tourist types presuming that participants might be biased by these. Additionally, we have defined four age groups, viz. less than 20, 21 to 40, 41 to 60

and over 60. Each participant was asked to select those tourist types which she has belonged to in earlier periods of her life, or currently belongs to. For example, a participant aged 47 was requested to select her personal tourism habits when she was younger than 20, between 21 and 40 as well as her current preferenc The third part of the questionnaire comprised 60 photos depicting different ourism-related situations. Participants should identify those photos that best represent their personal tourism habits. In the end, we have gathered data from 426 respondents; their demographic composition in shown Table 1 Table 1. Personal and demographic characteristics of survey sample(n=426) g with long term partner-31l Resident of a ci The tourist typology is given in Table 2 and the descriptions as provided in the questionnaire are shown. Additionally, the absolute and relative frequencies of the respondents tourism preferences are given. Please note that the sum of the relative frequencies exceeds 100%, because most respondents obviously assigned themselves to more than one tourist type. The rank order of tourist types in this table significantly correlates(Pearson's r=0.895, a=0.001) with the results presented in 3 3 On pack and kick in tourism In order to generate a map of the relationships between tourist types and the photographs we carried out a correspondence analysis. Starting from a cross tabulation of photo click frequencies by tourist type, we obtained the correspon- dence analysis map depicted in Figure 1. The results show that the relationship between tourist type and photo can be mapped onto two dimensions that account for 56.44% of the inertia, i.e. a large amount of the total variance is explained by the first two principal axes. In particular, the x-axis can be referred to as the Pack Factor and the y-axis represents the Kick Factor. The Pack Factor identifies the level of collectivity"one can associate with a particular tourist type. Consider, for example, the explorer, which is the left-most tourist type, and the Organized Mass Tourist, the right-most tourist type along the x-axis The Explorer might be identified as a rather solitary individual compared to an Organized Mass Tourist, who is generally accompanied by a larger number of like-minded tourists. This interpretation is corroborated by the findings of a study in which tourist experiences have been identified to vary along an indi- vidualistic/collectivistic continuum [6. The Kick: Factor identifies the "level of excitement"one might associate with a particular tourist activity. The Thrill Seeker, for instance, is by definition interested in risky, exhilarating activities

48 H. Berger et al. and over 60. Each participant was asked to select those tourist types which she has belonged to in earlier periods of her life, or currently belongs to. For example, a participant aged 47 was requested to select her personal tourism habits when she was younger than 20, between 21 and 40 as well as her current preferences. The third part of the questionnaire comprised 60 photos depicting different tourism-related situations. Participants should identify those photos that best represent their personal tourism habits. In the end, we have gathered data from 426 respondents; their demographic composition in shown Table 1. Table 1. Personal and demographic characteristics of survey sample (n=426) Gender Female - 208; Male - 218 Age group 21 to 40 - 200; 41 to 60 - 187; 61 and above - 39 Education Primary - 148; Secondary - 156; University - 122 Marital status Single/separated - 115; married/living with long term partner - 311 Kids no kids - 189; one or more kids - 237 Resident of a city - 188; village/town - 238 The tourist typology is given in Table 2 and the descriptions as provided in the questionnaire are shown. Additionally, the absolute and relative frequencies of the respondents’ tourism preferences are given. Please note that the sum of the relative frequencies exceeds 100%, because most respondents obviously assigned themselves to more than one tourist type. The rank order of tourist types in this table significantly correlates (Pearson’s r = 0.895, α = 0.001) with the results presented in [3]. 3 On Pack and Kick in Tourism In order to generate a map of the relationships between tourist types and the photographs we carried out a correspondence analysis. Starting from a cross tabulation of photo click frequencies by tourist type, we obtained the correspon￾dence analysis map depicted in Figure 1. The results show that the relationship between tourist type and photo can be mapped onto two dimensions that account for 56.44% of the inertia, i.e. a large amount of the total variance is explained by the first two principal axes. In particular, the x-axis can be referred to as the Pack Factor and the y-axis represents the Kick Factor. The Pack Factor identifies the “level of collectivity” one can associate with a particular tourist type. Consider, for example, the Explorer, which is the left-most tourist type, and the Organized Mass Tourist, the right-most tourist type along the x-axis. The Explorer might be identified as a rather solitary individual compared to an Organized Mass Tourist, who is generally accompanied by a larger number of like-minded tourists. This interpretation is corroborated by the findings of a study in which tourist experiences have been identified to vary along an indi￾vidualistic/collectivistic continuum [6]. The Kick Factor identifies the “level of excitement” one might associate with a particular tourist activity. The Thrill Seeker, for instance, is by definition interested in risky, exhilarating activities

Photo-Based User Profiling for Tourism Recommender Syster Table 2. Tourist types, their descriptions and distributions statistics urist type Sun lover hing in warm place Active Spor that provide emotional highs. Contrary, the Escapist I enjoys taking it easy away from the stresses and pressures of the home environment The generated layout of photos is to a high degree in-line with the alignment of the tourist types. For example, photos 22(alpine ski touring) and 37(alpine skiing) are highly associated with Active Sports whereas photos 46(whitewater rafting), 52(sky diving), 56(bungee jumping) and 59(windsurfing) correspond to the Thrill Seeker. The Action Seeker is represented by photos such as 3, 21 and 29 all of which are party sujets. The photo layout also reflects the characteristics of the axes. For example, photo 27 shows the highest level of individualism; in fact it depicts a solitary hitch hiker. On the contrary, photo 14 represents a typical packaged tour enjoyed by a group of bus tourists. In terms of the Kick Factor, photos 1(car rental area at airport)and 55(group listening to tour guide) show a moderate level of excitement whereas photos 52 and 56 depict risky and exhilarating activities. Note that photo 30(audience with an Indian Bhagwan) was selected by 1l respondents only and, thus, is regarded as a statistical outlier. The correspondence map is divided into four quadrants each of which reflect ng peculiarities of a set of tourist types. The lower left quadrant, for example, describes a high level of individualism and rather tranquil activities. As a result this quadrant contains tourist types such as the Anthropologist, Archaeologist

Photo-Based User Profiling for Tourism Recommender Systems 49 Table 2. Tourist types, their descriptions and distributions statistics Tourist type Description Freq. % Anthropologist Mostly interested in meeting the local people, trying the food and speaking the language 334 78.40 Escapist I Enjoys taking it easy away from the stresses and pressures of home environment 320 75.12 Archaeologist Primarily interested in archaeological sites and ruins; enjoys studying history of ancient civilizations 265 62.21 Sun Lover Interested in relaxing and sunbathing in warm places with lots of sun, sand and ocean 263 61.74 Independent Mass , Visits regular tourist attractions but avoids Tourist I, (IMT I) packaged vacations and organized tours 223 52.35 High Class Travels first class, stays in the best hotels, goes to shows and enjoys fine dining 207 48.59 Independent Mass Plans own destination and hotel reservations Tourist II, (IMT II) and often plays it by ear (spontaneous) 196 46.01 Escapist II Gets away from it all by escaping to peaceful, deserted or out of the way places 174 40.85 Organized Mass Tourist, Mostly interested in organized vacations, packaged tours, (OMT) taking pictures/buying lots of souvenirs 163 38.26 Active Sports Primary emphasis while on vacation is to remain active engaging in favorite sports 158 37.09 Seeker Seeker of spiritual and/or personal knowledge to better understand self and meaning of life 136 31.92 Explorer Prefers adventure travel, exploring out of the way places and enjoys challenge in getting there 132 30.99 Educational Tourist, Participates in planned study tours and seminars (Edu-Tourist) to acquire new skills and knowledge 127 29.81 Jet Setter Vacations in elite, world class resorts, goes to exclusive night clubs, and socializes with celebrities 104 24.41 Action Seeker Mostly interested in partying, going to night clubs and meeting people for uncomplicated romantic experiences 86 20.19 Thrill Seeker Interested in risky, exhilarating activities which provide emotional highs for the participant 61 14.32 Drifter Drifts from place to place living a hippie-style existence 55 12.91 that provide emotional highs. Contrary, the Escapist I enjoys taking it easy, away from the stresses and pressures of the home environment. The generated layout of photos is to a high degree in-line with the alignment of the tourist types. For example, photos 22 (alpine ski touring) and 37 (alpine skiing) are highly associated with Active Sports whereas photos 46 (whitewater rafting), 52 (sky diving), 56 (bungee jumping) and 59 (windsurfing) correspond to the Thrill Seeker. The Action Seeker is represented by photos such as 3, 21 and 29 all of which are party sujets. The photo layout also reflects the characteristics of the axes. For example, photo 27 shows the highest level of individualism; in fact it depicts a solitary hitch hiker. On the contrary, photo 14 represents a typical packaged tour enjoyed by a group of bus tourists. In terms of the Kick Factor, photos 1 (car rental area at airport) and 55 (group listening to tour guide) show a moderate level of excitement whereas photos 52 and 56 depict risky and exhilarating activities. Note that photo 30 (audience with an Indian Bhagwan) was selected by 11 respondents only and, thus, is regarded as a statistical outlier. The correspondence map is divided into four quadrants each of which reflect￾ing peculiarities of a set of tourist types. The lower left quadrant, for example, describes a high level of individualism and rather tranquil activities. As a result, this quadrant contains tourist types such as the Anthropologist, Archaeologist as

眉 Fig 1. Correspondence map of the relationship between tourist types and tourism- related photographs. The x-axis represents the level of collectivity and the y-axis the level of excitement well as the Escapist I that were quite frequently chosen by the respondents. The rather compact arrangement of these tourist types reflects their very close rela- tionship and, hence, it is difficult to distinguish between them. The upper-left quadrant comprises the Explorer, Active Sports and Drifter tourist types that show a rather high level of individualism as well as excitement

50 H. Berger et al. HighClass SunLover IMT_II OMT Escapist_II Seeker IMT_I Explorer JetSetter ActionSeeker ThrillSeeker Drifter Anthropologist Archaeologist -0.15 -0.10 -0.05 0.00 0.05 0.10 0.15 -0.05 0.00 0.05 0.10 0.15 0.20 Individual Pack Factor Group Moderate Kick Factor High 30 56 52 46 16 59 21 38 45 4 60 22 ActiveSports 57 37 51 58 12 7 23 9 54 39 18 25 50 32 49 10 41 53 11 29 3 8 5 1 55 14 24 31 47 33 15 27 36Escapist_I EduTourist Fig. 1. Correspondence map of the relationship between tourist types and tourism￾related photographs. The x–axis represents the level of collectivity and the y–axis the level of excitement. well as the Escapist I that were quite frequently chosen by the respondents. The rather compact arrangement of these tourist types reflects their very close rela￾tionship and, hence, it is difficult to distinguish between them. The upper-left quadrant comprises the Explorer, Active Sports and Drifter tourist types that show a rather high level of individualism as well as excitement

Photo-Based User Profiling for Tourism Recommender Systems 51 Seven tourist types can be found in the upper-right qu uadrant of the map. The types range from Educational Tourist to Thrill Seeker and from Escapist Il to High Class. The differences between some of these types seem to be rather small taking their close position in the map into account. a possible interpretation is that the Seeker (".searching for spiritual and/or personal knowledge.. ")and the Educational Tourist(".searching for new skills and knowledge. ")share some common ground or are performed in a sense at the same time. The lower- right quadrant contains two tourist types, namely the Sun Lover and the Or- ganized Mass Tourist. The degree of individuality attributed to these tourist types is rather low since packaged tours can be regarded as their dominating characteristic. Nevertheless, there seems to be considerable difference in terms of individuality between the Sun Lover and the Organized Mass Tourist taking the distance of their alignment in the map into account. The Kick Factor asso- ciated with these tourist types is rather moderate highlighting their desire for relaxation and hassle-free tourism experiences 4 Significant Photos The significance of individual photos to distinguish between tourist types is an- lyzed by means of logistic regression. In particular, the photos with positive, Anthropologist".IMT I Fig. 2. Affirmative example photos for particular tourist types

Photo-Based User Profiling for Tourism Recommender Systems 51 Seven tourist types can be found in the upper-right quadrant of the map. The types range from Educational Tourist to Thrill Seeker and from Escapist II to High Class. The differences between some of these types seem to be rather small taking their close position in the map into account. A possible interpretation is that the Seeker (“...searching for spiritual and/or personal knowledge...”) and the Educational Tourist (“...searching for new skills and knowledge...”) share some common ground or are performed in a sense at the same time. The lower￾right quadrant contains two tourist types, namely the Sun Lover and the Or￾ganized Mass Tourist. The degree of individuality attributed to these tourist types is rather low since packaged tours can be regarded as their dominating characteristic. Nevertheless, there seems to be considerable difference in terms of individuality between the Sun Lover and the Organized Mass Tourist taking the distance of their alignment in the map into account. The Kick Factor asso￾ciated with these tourist types is rather moderate highlighting their desire for relaxation and hassle-free tourism experiences. 4 Significant Photos The significance of individual photos to distinguish between tourist types is an￾alyzed by means of logistic regression. In particular, the photos with positive, Fig. 2. Affirmative example photos for particular tourist types

Actve Sports Fig 3. Counterexample photos for particular tourist types significant coefficients in the regression model are regarded as affirmative exam- ples for a particular tourist type. Conversely, photos with negative, significant coefficients are counterexamples. Following this approach, we obtain the map- pings of photos to tourist types as given in Figure 2 and Figure 3. We indicate the significance levels with asterisks: ***(a=0.001 )and **(a=0.01) Regarding the affirmative examples (cf. Figure 2), we obtain impressive re sults for characterizing the following tourist types: Anthropologist(photo 0 a group of indigenous musicians), Archaeologist (photo 34- the remnants of an ancient Greek temple), Sun Lover(photo 25-the beach), High Class(photo 24 the entrance hall of a stylish hotel: photo 31-a posh bar), Organized Mass Tourist (photo 14- group of bus tourists), Active Sports(photo 39-cyclists) Action Seeker(photo 29-a party), and Thrill Seeker(photo 46-whitewater rafting: photo 52- sky diving). However, we also recognized the rather unex- pected phenomenon that photo 38, showing the Burj al-Arab hotel in Dubai, is representative for six tourist types. This particular photo was selected by 163 participants. So, we assume that a fairly large number of participants regarded photo 38 as their emblematic vacation dream rather than their vacation prac- ce. For the counterexamples(cf. Figure 3), we refer to photo 54 depicting a street musician. The selection of this photo significantly excludes the member- ship to the archaeologist. Photo 13, showing a tranquil scenery with boat, is a perfect example against the typical Active Sports tourist. Finally, we want to mention that only for a small number of tourist types we were unable to identify mportant photos, viz. Seeker, Explorer and Drifter. 5 Photo- Based profiling We have designed and implemented a Web-based tourist type profiling tool as shown in Figure 4 using the statistical model established by the logistic re- gression. Eight sets of photographs are located in the top row and users may switch from one set to the next by clicking the respective hyperlink. Users may drag photographs they identify with into the lower-left area. If they change their mind, photos can be dragged to the area bordering right, the "Wastebin", to ex- clude them from their selection. Note, however, photos moved to the"Wastebin Giveitagoathttp://ispaces.ec3.at/tourismpRofiler/index.html

52 H. Berger et al. Fig. 3. Counterexample photos for particular tourist types significant coefficients in the regression model are regarded as affirmative exam￾ples for a particular tourist type. Conversely, photos with negative, significant coefficients are counterexamples. Following this approach, we obtain the map￾pings of photos to tourist types as given in Figure 2 and Figure 3. We indicate the significance levels with asterisks: *** (α = 0.001) and ** (α = 0.01). Regarding the affirmative examples (cf. Figure 2), we obtain impressive re￾sults for characterizing the following tourist types: Anthropologist (photo 02 - a group of indigenous musicians), Archaeologist (photo 34 - the remnants of an ancient Greek temple), Sun Lover (photo 25 - the beach), High Class (photo 24 - the entrance hall of a stylish hotel; photo 31 - a posh bar), Organized Mass Tourist (photo 14 - group of bus tourists), Active Sports (photo 39 - cyclists), Action Seeker (photo 29 - a party), and Thrill Seeker (photo 46 - whitewater rafting; photo 52 - sky diving). However, we also recognized the rather unex￾pected phenomenon that photo 38, showing the Burj al-Arab hotel in Dubai, is representative for six tourist types. This particular photo was selected by 163 participants. So, we assume that a fairly large number of participants regarded photo 38 as their emblematic vacation dream rather than their vacation prac￾tice. For the counterexamples (cf. Figure 3), we refer to photo 54 depicting a street musician. The selection of this photo significantly excludes the member￾ship to the Archaeologist. Photo 13, showing a tranquil scenery with boat, is a perfect example against the typical Active Sports tourist. Finally, we want to mention that only for a small number of tourist types we were unable to identify important photos, viz. Seeker, Explorer and Drifter. 5 Photo-Based Profiling We have designed and implemented a Web-based tourist type profiling tool1 as shown in Figure 4 using the statistical model established by the logistic re￾gression. Eight sets of photographs are located in the top row and users may switch from one set to the next by clicking the respective hyperlink. Users may drag photographs they identify with into the lower-left area. If they change their mind, photos can be dragged to the area bordering right, the “Wastebin”, to ex￾clude them from their selection. Note, however, photos moved to the “Wastebin” 1 Give it a go at http://ispaces.ec3.at/TourismProfiler/index.html

Photo-Based User Profiling for Tourism Recommender Systems :@⊙全6hmp/s0ess3 at/TourismProfiler/index.html p(G·cog Photo-Based Tourist Profiler M234A显 要 Fig 4. Web-based tourist type profiler eee Your Personal Tourist Profile H.5.ea0http:/ispaces.ec3.at/toUrismprofileR/yOurproFilejsp7quenv(ig.coogle Your Personal Tourist Profile This tourist proflle was generated for A28 len-ko@web.de onal poie w be sent to this aldos) extraordinary good Send it Fig. 5. Result page from the web-based tourist type profiler

Photo-Based User Profiling for Tourism Recommender Systems 53 Fig. 4. Web-based tourist type profiler Fig. 5. Result page from the web-based tourist type profiler

54 H. Berger et al. re not regarded as negative examples in the regression model. Whenever the photo selection changes, the degrees of affiliation to particular tourist types is calculated instantaneously and displayed on the right-hand side of the Web page Both, the tourist types speaking significantly for and against a persons tourism habits are shown and listed according to the descending degree of affiliation. As an example consider the photo selection depicted in Figure 4. These photo preferences indicate a strong affiliation with the tourist type "Sun Lover", fol lowed by“ Active Sports”,“ Action Seeker”and“ Independent Mass Tourist".Ad- ditionally, this user profile resembles a person being no"Escapist Ir, "Seeker Jet Setter”or“ Archeologist” The results of the tourist type profiling are finally presented to the user as shown in Figure 5. The users may fill in their personal details and rate the quality of the profiling 6 Conclusion In this paper, we presented the findings of an online survey conducted to inves- tigate whether tourists habits can be derived from tourism-related photographs in order to facilitate the process of user profile creation. The results of this survey show a significant relationship between different tourist types and the preference for particular visual impressions conveyed by photographs. For most tourist types, we have determined representative photos, which, in turn, allow the assignment of tourist types to persons based on their selection of a set of photos. Considering the relationship of tourist types and tourist activities stated in 4, we arrive at a mapping between tourism-related photographs and tourist activities. The concept is showcased by means of a Web-based tourist type pro- filing tool using the statistical model established by the logistic regression. It is now possible to make the traditional process of registration and profile genera- tion more enjoyable by letting the user select from a couple of photos that reflect her tourism habits, and then infer her according tourist types In a next step towards the photo-based tourism recommender system we aim at associating the different tourist types with tourism products. Thus, a user will obtain a result of tourism product recommendations based on the preferences for particular photographs. This approach, and most importantly the product to type associations, will be evaluated by means of a comprehensive user study References 1. Delgado, J, Davidson, R. Knowledge bases and user profiling in travel and hospi- tality recommender systems. In: Proceedings of the 9th International Conference on Information Technologies in Tourism(ENTER'02), Innsbruck, Austria, pp. 1-16 Springer, Heidelberg(2002) 2. Fesenmaier. D.R., Ricci, F, Schaumlechner, E, Wober, K, Zanella, C: DI- ETORECS: Travel advisory for multiple decision styles. In: Frew, A.J., Hitz, M O'Connor, P.(eds Proceedings of the 10th International Conference on Infor mation Technologies in Tourism(ENTER'03), Helsinki, Finland, January 29-3 2003, pp. 232-241. Springer, Heidelberg(2003)

54 H. Berger et al. are not regarded as negative examples in the regression model. Whenever the photo selection changes, the degrees of affiliation to particular tourist types is calculated instantaneously and displayed on the right-hand side of the Web page. Both, the tourist types speaking significantly for and against a person’s tourism habits are shown and listed according to the descending degree of affiliation. As an example consider the photo selection depicted in Figure 4. These photo preferences indicate a strong affiliation with the tourist type “Sun Lover”, fol￾lowed by “Active Sports”, “Action Seeker” and “Independent Mass Tourist”. Ad￾ditionally, this user profile resembles a person being no “Escapist II”, “Seeker”, ”Jet Setter” or “Archeologist”. The results of the tourist type profiling are finally presented to the user as shown in Figure 5. The users may fill in their personal details and rate the quality of the profiling. 6 Conclusion In this paper, we presented the findings of an online survey conducted to inves￾tigate whether tourist’s habits can be derived from tourism-related photographs in order to facilitate the process of user profile creation. The results of this survey show a significant relationship between different tourist types and the preference for particular visual impressions conveyed by photographs. For most tourist types, we have determined representative photos, which, in turn, allow the assignment of tourist types to persons based on their selection of a set of photos. Considering the relationship of tourist types and tourist activities stated in [4], we arrive at a mapping between tourism-related photographs and tourist activities. The concept is showcased by means of a Web-based tourist type pro- filing tool using the statistical model established by the logistic regression. It is now possible to make the traditional process of registration and profile genera￾tion more enjoyable by letting the user select from a couple of photos that reflect her tourism habits, and then infer her according tourist types. In a next step towards the photo-based tourism recommender system we aim at associating the different tourist types with tourism products. Thus, a user will obtain a result of tourism product recommendations based on the preferences for particular photographs. This approach, and most importantly the product to type associations, will be evaluated by means of a comprehensive user study. References 1. Delgado, J., Davidson, R.: Knowledge bases and user profiling in travel and hospi￾tality recommender systems. In: Proceedings of the 9th International Conference on Information Technologies in Tourism (ENTER’02), Innsbruck, Austria, pp. 1–16. Springer, Heidelberg (2002) 2. Fesenmaier, D.R., Ricci, F., Schaumlechner, E., W¨ober, K., Zanella, C.: DI￾ETORECS: Travel advisory for multiple decision styles. In: Frew, A.J., Hitz, M., O’Connor, P. (eds.) Proceedings of the 10th International Conference on Infor￾mation Technologies in Tourism (ENTER’03), Helsinki, Finland, January 29–31, 2003, pp. 232–241. Springer, Heidelberg (2003)

Photo-Based User Profiling for Tourism Recommender Systems 3. Gibson, H, Yiannakis, A: Tourist roles needs and the lifecourse. Annals of Tourism Research 29(2), 358-383(2002) 4. Gretzel, U. Mitsche, N, Hwang, Y -H, Fesenmaier, D R. Tell me who you are and I will tell you where to go: Use of travel personalities in destination recommendation systems. Information Technology and Tourism 7(1), 3-12(2004) 5. Jannach, D,, Kreutler, G. Personalized user preference elicitation for e-services In Proceedings of the IEEE International Conference on e-Technology, e-Commerce and e-Service(IEEE05), Hong Kong, China, March 29-April 1, 2005, pp. 604-611 IEEE Computer Society Press, Los Alamitos(2005) 6. Mehmetoglu, M. A typology of tourists from a different angle. International Jour- nal of Hospitality &z Tourism Administration 5 (3), 69-90(2004) Rashid, A M, Albert, I, Cosley, D, Lam, S K, McNee, S.M., Konstan, J.A., Riedl, J. Getting to know you: Learning new user preferences in recommender systems. In: Proceedings of the 7th International Conference on Intelligent User Interfaces (IUI 02), pp. 127-134. ACM Press, New York(2002) 8. Ricci, F: Travel recommender systems. IEEE Intelligent Systems 17 (6),55-5 2002 9. Schein, A.I., Alexandrin Popescul, A, Ungar, L H, Pennock, D M. Methods and metrics for cold-start recommendations. In: Proceedings of the 25th Annual In- ternational ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR 02), Tampere, Finland, pp. 253-260. ACM Press, New York 10. Yiannakis, A, Gibson, H. Roles tourists play. Annals of Tourism Research 19(2)

Photo-Based User Profiling for Tourism Recommender Systems 55 3. Gibson, H., Yiannakis, A.: Tourist roles – needs and the lifecourse. Annals of Tourism Research 29(2), 358–383 (2002) 4. Gretzel, U., Mitsche, N., Hwang, Y.-H., Fesenmaier, D.R.: Tell me who you are and I will tell you where to go: Use of travel personalities in destination recommendation systems. Information Technology and Tourism 7(1), 3–12 (2004) 5. Jannach, D., Kreutler, G.: Personalized user preference elicitation for e-services. In: Proceedings of the IEEE International Conference on e-Technology, e-Commerce and e-Service (IEEE’05), Hong Kong, China, March 29-April 1, 2005, pp. 604–611. IEEE Computer Society Press, Los Alamitos (2005) 6. Mehmetoglu, M.: A typology of tourists from a different angle. International Jour￾nal of Hospitality & Tourism Administration 5(3), 69–90 (2004) 7. Rashid, A.M., Albert, I., Cosley, D., Lam, S.K., McNee, S.M., Konstan, J.A., Riedl, J.: Getting to know you: Learning new user preferences in recommender systems. In: Proceedings of the 7th International Conference on Intelligent User Interfaces (IUI ’02), pp. 127–134. ACM Press, New York (2002) 8. Ricci, F.: Travel recommender systems. IEEE Intelligent Systems 17(6), 55–57 (2002) 9. Schein, A.I., Alexandrin Popescul, A., Ungar, L.H., Pennock, D.M.: Methods and metrics for cold-start recommendations. In: Proceedings of the 25th Annual In￾ternational ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR ’02), Tampere, Finland, pp. 253–260. ACM Press, New York (2002) 10. Yiannakis, A., Gibson, H.: Roles tourists play. Annals of Tourism Research 19(2), 287–303 (1992)