Keywords

1 Introduction

The rapid development of Information & Communication Technologies (ICT), and especially the emergence of the World Wide Web, lead to fundamental changes in the tourism industry, both on the supply and the demand side. Consumers now have ubiquitous access to vast amounts of information at a very low cost, allowing them to compose and compare products and optimize their trips. However, increasing cognitive costs to process the amount and variety of information could lead to the problem of information overload. This shows the necessity of new techniques and tools to analyse, categorize and visualize information in a proper way (Hwang, Gretzel, Xiang, & Fesenmaier, 2006). On the other side, the Web also allows a massive “informatization” of the whole tourism value chain, resulting in many novel value-generating strategies, to satisfy new consumer needs (Werthner & Ricci, 2004; Werthner et al., 2015).

According to a recent study (Ipsos MediaCT, 2014) people rely on online sources such as social media, photo- and video-sites and search engines to get inspired where to go or how to travel. It is shown that 65% of the leisure travellers start researching online before a travel decision. Particularly in this early phase of decision making a considerable amount of people has difficulties to explicitly express their preferences and needs (Zins, 2007). Recommender Systems (RSs) are facilitating this decision-making. Ricci, Rokach, and Shapira (2015) define RSs as “software tools and techniques providing users with suggestions for items a user may wish to utilize”. Particularly, profiling and personalization might help in such cases, where preferences and needs are unknown or hard to express. Especially in tourism this is a big challenge, since tourism products are considered as very complex (i.e., they typically combine accommodation, transportation, activities, food, etc.), mostly intangible and highly associated with emotional experiences (Werthner & Ricci, 2004).

Neidhardt, Seyfang, Schuster and Werthner (2014, 2015) introduced a picture based approach to elicit the preferences of a user and a seven-factor model to capture the respective user’s profile within a travel recommender system. These factors form the basis of a seven-dimensional vector space and are referring to travel behavioural patterns summarized as Sun & Chill-Out, Knowledge & Travel, Independence & History, Culture & Indulgence, Social & Sport, Action & Fun, and Nature & Recreation. Based on a picture-selection process, a user’s profile is determined. This profile comprises a score for each of the factors and thus can be seen as a point in the seven-dimensional vector space. In order to provide recommendations to a user, those items have to be determined that are closest to him/her. Thus, also the items have to be mapped into the vector space, i.e., represented with respect to the travel behavioural patterns. In order to build up a reasonable recommendation base more than 10,000 tourism products were initially mapped manually by experts. Obviously, this approach does not scale.

The aim of this work is to find an automated way of relating tourism products to travel behavioural patterns. In particular, the relationships between the seven-factors and attributes of destinations will be examined in order to map the destinations onto the seven-factors and to group similar destinations for a better understanding and generalization.

The rest of the paper is organized as follows. In Sect. 2 the state of the art is presented, focusing on tourist roles, the seven-factor model, and tourism recommender systems. In Sect. 3, the data sample is described. In Sect. 4 regression analyses based on expert mappings are conducted and discussed. In Sect. 5 a cluster analysis is conducted and presented. In Sect. 6 the main insights are summarized and conclusions are drawn.

2 State-of-the-Art

2.1 Tourist Roles & The Seven-Factor Model

Much research has been conducted in order to identify and categorize tourist roles, describing the relation between a person’s travel behaviour and his or her preferences, interest, and needs. In this context, Gibson and Yiannakis (2002) introduced a well-established classification framework, distinguishing 17 different tourist roles, capturing short-term behaviour. Gretzel, Mitsche, Hwang, and Fesenmaier (2006) demonstrated that tourist roles can be used in order to recommend touristic activities and, in turn, destinations. It has also been shown that tourist roles can be related to personality traits. Delić, Neidhardt and Werthner (2016) are providing significant evidence to the relation between the well-established Big-Five personality traits (Goldberg, 1990) and the 17 tourist roles (Gibson & Yiannakis, 2002). Personality traits tend to be stable over time and can be considered as long-term preferences of a person (Woszczynski, Roth, & Segars, 2002). Taking into account both the Big-Five personality traits (long-term behaviour) and 17 tourist roles (short-term behaviour) Neidhardt et al. (2014, 2015) developed a seven-factor model by conducting factor analysis. Each factor is referring to a travel behavioural pattern, which are summarized in Table 1. These factors are easier to process computationally as well as cognitively compared to the original 22 dimensions. Neidhardt and Werthner (2017) showed that based on different demographic characteristics different user groups can be well distinguished within this model.

Table 1. Seven-factor model
Full size table

2.2 Tourism Recommender Systems

The relationship between tourism and ICT is described as a symbiosis by Gretzel (2011) and travel & tourism have always been major application domains for Web-related services (Werthner & Klein, 1999). As the amount of information on the Web started to rise the call for techniques to cope with information overload began to grow. One answer to that are RSs. There are different types of recommendation techniques and hybrid solutions, but according to Burke and Ramezani (2011) the most appropriate ones in the matter of tourism are either content-based (recommending items similar to the ones the user already liked in the past) and/or knowledge-based (recommending items using domain knowledge about user preferences and needs and utility of items to the user). This work is based on a picture-based approach to RSs (Neidhardt et al. 2014, 2015), which can be considered as a content- and knowledge-based approach. Preferences and needs of a user are determined via a simple picture selection process. In this way, the so-called cold-start problem of learning-based techniques (collaborative, content-based, and demographic) (Burke, 2007) as well as tedious questionnaires for preference elicitation are avoided. Thus, the users are addressed on an emotional, implicit level and do not have to state their preferences explicitly. According to Garcia, Sebastia, and Onaindia (2011) tourism RSs can be distinguished into two types: one focusing on destination selection the other on activities that can be performed at a certain destination. This work belongs to the first mentioned group, in contrast to Neidhardt et al., (2014, 2015), where the focus lies on the Point of Interests (POIs), e.g., activities, events, restaurants, sights. Much research has already been conducted targeting destination recommender systems (Fesenmaier, Werthner, & Wöber, 2006; Borràs, Moreno, & Valls, 2014), but most of them are mainly focused on a distinct region or POIs in a destination. There are few that are focussed on personality traits and motifs of a user (see for example Braunhofer, Elahi, & Ricci, 2014).

3 The Data Sample

The data is provided as a SQL-dump by a German eTourism companyFootnote 1 and consists of more than 30,000 destinations all around the world. Almost all countries are represented in the database, but the majority of destinations are located in the USA, Germany, France, Italy, Spain, Great Britain, Austria, Greece, Switzerland, and Sweden (65%). Beirman (2003) defines tourism destinations as “a country, state, region, city or town which is marketed or markets itself as a place for tourists to visit”. In this work destinations are defined in a similar way, except the range goes from a hamlet (population <100) to a metropolis (population >1M). Destinations are described via destination features, which can be separated into two groups:

  1. 1.

    Motivational ratings in an interval of 0–1, describing the degree of appropriation for a particular motif. Following 27 motivational ratings are listed: nightlife, wellness, shopping, nature & landscape, image & flair, culture, sightseeing, entertainment, mobility, price level, accommodations, gastronomy, beach & swimming, golf, scuba diving, kite & windsurfing, hiking, cycling, horseback riding, winter sports, sports, family, quietness, surfing, sailing, gays, and mountain biking. The motivational ratings are determined considering factors such as infrastructure, climate, geographical nature, user opinions, number of services, image, and marketing.

  2. 2.

    Geographical attributes are presented in binary format, describing the presence or absence of a particular attribute. Following 14 attributes are listed: sea, mountain, lake, island, sandy beach, metropolis, forest, river, desert, old town, pebble beach, sand & pebble beach, hill, swamp, volcano, fjord, flat decaying sand beach, beach promenade, wine-growing, heath, health resort, and winter sports resort.

For this study, 561 destinations were chosen randomly and mapped manually by experts by assigning a score for each factor using the scale 0–0.25–0.50–0.75–1. The higher the score the more suitable the destination for that specific factor. The consulted experts are members of an Austrian eTourism company that has already been using an implementation of the picture based approach introduced by Neidhardt et al. (2014, 2015). Hence, they have expertise in both the seven-factor model and the travel destinations. Three experts analyzed 561 destinations first individually and then determined a final mapping in a joint-discussion. Table 2 shows the resulting distributions.

Table 2. Factor distribution of the expert mapping
Full size table

For example, in case of the factor Sun & Chill-Out 39.93% of the destinations scores were with 0, 19.07% with 0.25, 13.9% with 0.5, 8.74% with 0.75, and 18.36% with 1. A similar distribution can also be observed for factors Knowledge & Travel, Culture & Indulgence, and Action & Fun, where the majority of destinations either score with 0 or 0.25. On the other hand, for the factors Independence & History and Social & Sports most of the destinations score with 0.5 and only very few with 0 or 1. The only factor where the differences are not that pronounced is Nature & Recreation.

4 Mapping of Destination Features to Seven Factors

The aim of the work is not only to project destinations into the seven-dimensional vector space of travel behavioural patterns using their features, but more importantly to understand the relationship between the seven-factors and destination features. In James, Witten, Hastie, and Tibshirani (2013a) it is suggested to choose linear models over more complex ones if inference and interpretability is the goal. Taking this into account, a multiple linear regression model (James, Witten, Hastie, & Tibshirani, 2013b) with step-wise variable selection (James, Witten, Hastie, & Tibshirani, 2013c) is applied. All seven factors are considered as independent from each other, since they are obtained from factor analysis. Therefore, they can be treated separately by fitting a model for each travel behavioural pattern, which takes the features of a destination as input and returns the factor score (0–1) as output. The data sample is split into a training and test set in a ratio of 80/20. Model performance is assessed by R2 (proportion of variance explained) and root mean square error (standard deviation of the residuals/prediction errors) of training (RMSEtrain) and test set (RMSEtest).

Table 3 shows the performance of each model. There is a statistically significant relationship between each travel behavioural pattern and destination features (used in the corresponding models), with p < 0.001 (***). RMSEtrain and RMSEtest are close, indicating that the resulting models will perform similar out of sample. Overall, all travel behavioural patterns are well described by the resulting models, except Social & Sports, where only 19% of the variance is explained. This is caused by the uneven distribution of the expert mapping of Social & Sports, where 53.83% of the destination scored with 0.5 and only 1.78% scored with 0 and 4.10% with 1 respectively. Nevertheless, there is significant evidence of a relation between destination features and the factor Social & Sports. Sun & Chill-Out and Action-Fun have the best fitted models, where 70 and 76% of the variance can be explained.

Table 3. Performance of the resulting multiple linear regression models
Full size table

The resulting multiple linear regression models comprise both motivational ratings and geographical attributes. After the variable selection 18 out of 27 motivational ratings and 7 out of 14 geographical attributes in total are used. Table 4 summarizes the outcomes of the regression analysis. Motivational ratings sightseeing, culture, entertainment, family, quietness, gastronomy, and image & flair appear in more than one model. Also, geographical attributes sea, health resort, and winter sports resort are in several models.

Table 4. Results of the regression analysis
Full size table

Model A— Sun & Chill - Out. The geographical attributes sea, health resort, and lake and motivational ratings beach & swim and nature & landscape are significantly, positively related to this factor. Those features can be interpreted as indicators for relaxation and indirectly also for sun. On the other side, motivational ratings sightseeing and nightlife have a significant negative impact, which can be associated with crowded places and mass tourism.

Model B— Knowledge & Travel. The motivational ratings sightseeing, culture, and entertainment are significantly, positively related to this factor. Those features can be seen as the main motivation of travellers with the ambition to learn something on their trip. On the contrary, motivational rating family and geographical attribute sea are significantly, negatively associated with this factor. The negative impact of motivational rating family can be explained by the fact that most destinations in the data sample, which have an increased motivational rating family (>0.5), show typical characteristics of recreational travel destinations. Namely, they are quietly situated and in the nature. The majority of destinations in the data sample, which are located at the sea, have sea resort characteristics, where cultural and educational interests are short comings. This explains the negative sign of the geographical attribute sea.

Model C— Independence & History. The motivational ratings culture, sightseeing, and gastronomy are significantly, positively related to this factor. Those features can be seen as the main motivation of travellers with interests in history and tradition. Whereas motivational rating quietness and geographical attributes sea and winter sports resort are significantly, negatively related. The negative impact of the geographical attribute winter sports resort is not surprising. The majority of destinations in the database, which are located in a winter sport resort, are mostly dedicated to winter sports and après-ski only. Thus, interests of an independent traveller with passion for history and tradition is short coming. The negative impact of the geographical attribute sea can be interpreted in the same way as for the factor Knowledge & Travel (see Model B—Knowledge & Travel). The motivational rating quietness is an indicator for low population and tourism density, where high rated places are considered more appropriate for recreational travellers than for people with interest in history and tradition. By this means, the negative sign can be explained.

Model D—Culture & Indulgence. The motivational ratings culture, sightseeing, gastronomy and image & flair are significantly, positively related to this factor. Those ratings can be interpreted as the main motivation of a culture and history interested high class tourist, who is also a connoisseur of good food and wine. On the other hand, the motivational rating family and the geographical attribute sea have a significant negative impact on the factor, which has the same cause as in Knowledge & Travel (see Model B—Knowledge & Travel).

Model E—Social & Sports. The motivational ratings sports and hiking are significantly, positively related to this factor, which is obvious. Whereas, the motivational ratings entertainment and sightseeing and the geographical attribute old town are significantly, negatively related to the factor. Those features can be seen as indicators for crowded places and mass tourism.

Model F—Action & Fun. The geographical attributes metropolis, sea, and winter sports resort and motivational ratings winter sports, sailing, entertainment, mobility, sightseeing, mountain biking, and golf are significantly, positively related to this factor. Metropolis, sea, entertainment, mobility, and sightseeing can be seen as indicators for vibrant places with party and fun. Whereas winter sports, mountain biking, and winter sports resort can be interpreted as motivators for thrill seeking and après-ski loving tourists. Finally, golf and sailing can be interpreted as indicators for exclusiveness and VIP. On the other hand, the motivational ratings family and quietness have a negative impact on the factor Action & Fun, which is not surprising and self-explanatory.

Model G—Nature & Recreation. The motivational rating quietness and geographical attributes mountains and health resort are significantly, positively related to this factor. Those features can be seen as the main interests of tourist, whose goal is recreation, finding silence and peace, and escaping from everyday life. On the other hand, the motivational ratings gastronomy, sightseeing, shopping, and image & flair have a significant negative impact on this factor. Those features can be interpreted as indicators for crowded places and city life.

5 Clustering

Identifying conceptually meaningful groups of destinations with shared common characteristics will help to further understand the data and its structure, which may contribute to a more generalized solution. Prototype-based, partitional clustering techniques are considered, where the most prominent ones are k-means and k-medoids. Since the data comprises binary attributes, centroids are not meaningful. Therefore, k-medoids is applied. A medoid corresponds per definition to an actual data point, which is considered as the most representative point for the cluster (Pang-Ning, Steinbach, & Kumar, 2006). Specifically, Partitioning Around Medoids (PAM) (Kaufman & Rousseeuw, 1990), the most common k-medoids algorithm, is used. Since the data consists of two different data types, i.e., binary (geographical attributes) and continuous (motivational ratings), the Gower distance (appropriate for mixed data types) (Gower, 1971) is used as distance metric. In order to find an appropriate number of clusters, the internal evaluation metric silhouette width (Rousseeuw, 1987) is used for assessment. Based on the silhouette width 4, 6 and 8 cluster solutions are considered, but for the sake of interpretability a 6-cluster solution is chosen. In Table 5 average factor scores and corresponding standard deviations (SD) for each cluster are listed.

Table 5. Average factor scores (and SD) in different clusters
Full size table

Destinations in C1 are more or less suitable for Social & Sports and Nature & Recreation. All other factors have a low average score. Liederbach (Hessen, Germany) is in the centre of C1 and the most representative observation. It is a small, calm suburb with almost no tourism destination characteristics, but just a few offers of sport activities.

Destinations in C2 are very well suited for Knowledge & Travel, Independence & History, Culture & Indulgence and Action & Fun. However, tourists with high Sun & Chill-Out or Nature & Recreation score will avoid destinations in C2. The most representative data point of C2 is Brussels, the capital of Belgium. The centre of C2 is a vibrant metropole, with many opportunities for nightlife, party, culture, sightseeing, and gastronomy. Recreation, relaxation and peace are alien to the medoid of C2.

Destinations in C3 are very well suited for Nature & Recreation and they also show an increased average score in factor Social & Sports. Also noteworthy is that people with high Action & Fun score, will avoid destinations in C3 at all. The centre of the cluster is Schönberg am Kamp (Lower Austria). It is a small, peaceful town in the nature good for hiking, cycling, escaping from everyday life, and recreational tourism.

Destinations in C4 are more or less suitable for Knowledge & Travel, Independence & History, Social & Sports, and Nature & Recreation. Overall, there is no dominant factor. The medoid of C4 is Todtnau (Baden-Württemberg, Germany). It is a rural town, where 60% is covered with forest. Todtnau is good for hiking, cycling, mountain biking, and other sports, but also for recreational tourism. Nevertheless, it also offers cultural activities, gastronomy, nightlife and entertainment in a moderate level. Noteworthy to mention is that the effect of both recreation and sports indicators and culture, history, tradition and entertainment indicators are dampening each other. Hence, there is no pronounced differences in the factor scores and none of them are either satisfied or unsatisfied.

Destination in C5 have on average a high score in Sun & Chill-Out, an increased score in Social & Sports and Action & Fun, and a moderate level of Knowledge & Travel, Independence & History, and Culture & Indulgence. Nature & Recreation is the only factor, which can be considered as low. The centre of C5 is Grand Baie on the island of Mauritius. It is considered as a centre of beach tourism on the northern coast of the island. Besides beach tourism there are plenty of opportunities for nightlife, entertainment, gastronomy, shopping, and water sports.

Destinations in C6 are a perfect match for Sun & Chill-Out. Additionally, Nature & Recreation and Social & Sports have an increased score on average. The medoid of C6 is Anaxos, a small, intimate, and tranquil seaside resort, which is located on the island of Lesbos (Greece). Beside sun, beach, and recreation there are opportunities for (water) sports and get-together with locals.

In summary, it can be said that there is an underlying natural structure of the data. Thus, six conceptually meaningful groups of destinations could be identified. For a better understanding, these groups or clusters can be simplified and summarized as follows: C1—passionless suburb, C2—energetic city, C3—peaceful village, C4—ordinary town, C5—vibrant beach resort, C6—tranquil seaside resort. The identified groups can help to better understand the (dis)similarities among the destinations and may also contribute to a more generalized solution of mapping without the need of prior information in contrast to the supervised method in the previous section. Nevertheless, considering the outcomes of the regression analysis one can clearly see relationships between the used destination features and the most determinant features of the cluster solution. Especially, destination features that appear in more than one regression model are the most decisive characteristics of cluster cohesion (separation) in the presented model.

6 Conclusions

Primarily, this works aim is to identify and explain associations between destination attributes and the seven-factor model to enable an automated mapping of destinations onto the seven factors. To do so, a multiple linear regression analysis with step wise variable selection was conducted. Seven models were established, one for each of the seven factors. The resulting models are providing strong evidence that there is a significant relation between selected destination features and the factors. Overall, all travel behavioural patterns are well described (52–76% of the variance) by the resulting models, except Social & Sports, where only 19% of the variance can be explained. This is caused by an uneven distribution of the sample. A statically sounder sample will be targeted in future work. A linear model has the benefit of interpretability, but might suffer in performance. Additionally, the chosen variable selection model reveals core attributes, but is greedy (relevant information might get lost). In this case a performance evaluation of different methods (principal component regression, ridge regression etc.) is planned. Furthermore, a cluster analysis was conducted in order to determine if there is an underlying natural structure of the data sample. Six conceptually meaningful groups were identified (passionless suburb, energetic city, peaceful village, ordinary town, vibrant beach resort, tranquil seaside resort). Those clusters can foster a better understanding of the (dis)similarities among destinations and also of the relation between destination features and the seven-factor model. Further, these clusters can be used for more accurate recommendations or can be targeted directly by a RSs. In future work, it is also planned to recommend accommodations in a second step. Thus, an analysis revealing the relation between hotel features and the seven factors is needed. However, this aim immediately shows a disadvantage of the followed approach, namely data source dependency. To counter this problem there is an ongoing study, which aims to build up a comprehensive data model of tourism products. This data model will serve as an “intermediary” layer between the respective data source and the seven-factor model and can therefore be used to harmonize heterogeneous sources of data (e.g., by mapping different sources of destination data onto this layer).