This study examines the cognitive image of Austria as a travel destination through Instagram content posted by travel influencers. The study also investigates how the account type, influencer type, and posting frequency affect user engagement. Machine learning techniques and statistical analysis are used to analyze the data. The study found that influencers contribute to Austria's destination image mainly through content about the Alps, Vienna, and cycling. The study provides insights into successful destination promotion on Instagram through influencer marketing. Micro-influencers who post regularly with relevant content are ideal for DMOs. Meso-influencers and verified accounts receive more likes for less popular themes, while micro-influencers are sufficient for more popular themes. It is also disadvantageous for meso-influencers to be perceived as commercial accounts and not to post as often as emerging influencers.
1 Introduction
Social media are extensively studied in the tourism literature as they provide consumers with a wealth of information for purchase decisions [1] and have become a more important source of information in terms of brand image. User-generated content (UGC) can shape destination image [2, 3] a role previously reserved for traditional channels [4]. It is difficult for destinations to control UGC, which is why influencers act as intermediaries in their social media communications. [5]. Influencers’ testimonials based on personal experiences are perceived as a strong eWOM, leading to increased demand and positive attitudes towards the brand [6].
As travel influencers have the power to shape the image of a destination and influence the experience of a destination and its future development [7, 8] this study 1) analyses the image of Austria as a travel destination and examines it through the content posted by travel influencers on Instagram, and 2) assesses how the account type (business account or verified account), influencer type (micro or meso influencer, depending on the number of followers) and posting frequency (number of posts) influence user engagement.
Anzeige
This study uses data collected on Instagram and applies a mixed methods approach using machine learning techniques and statistical analysis. The methodology includes computer vision analysis using a convolutional neural network (CNN) and multilevel cluster analysis to answer the following three research questions:
(1)
What kind of content was posted by influencers on Instagram about Austria as a travel destination?
(2)
What content is most attractive to users in terms of likes and comments?
(3)
How do the influencers’ profile characteristics (number of followers, number of posts, type of profile account) affect user engagement (likes and comments)?
From an industry perspective, the results should provide valuable insights into how successful promotion of a destination on Instagram by an influencer can be achieved. In addition, the study makes a methodological contribution by demonstrating the benefits of machine learning-assisted image content analysis.
2 Literature Review
Destination image is described as a tourist's collection of ideas, impressions and thoughts about a particular destination [9]. This image is built on the basis of the knowledge a person has about the destination and the emotions that the destination evokes [10‐12]. Visitors nowadays have the possibility to create and spread their own impressions [8, 13, 14]. The image of a destination is thus shaped by the official channels and the tourists [8] which is why some DMOs turn to influencers to shape the destination's image through the influencer [4, 15]. Influencers provide content based on personal experiences, reflecting stories from their daily lives. [1, 6, 8, 16] and thus increase their popularity and influence, which is reflected in the number of likes, shares and comments [6, 8, 17]. According to Veirman et al. [1], the number of followers of influencers has a positive effect on perceived popularity and thus on likeability, while a positive effect on attributed opinion leadership is present but weak. In other words, a high number of likes is not always an indicator of their influencing power. Several studies [8, 16, 18] assume that the perceived match between the influencer and the brand they support plays a crucial role in the credibility and success of the advertising campaign [19]. These studies emphasise the importance of matching the influencer's lifestyle and published content with the brand's image and the brand's target audience with the influencer's target audience for successful advertising and promotion.
Other studies in the tourism industry have focused on the impact of social media influencers on the travel-related behavioural intentions of their followers [20]. Magno and Cassia [20] showed that the influence of travel bloggers on the travel intentions of their followers is highly dependent on the perceived trustworthiness [21] of the blogger and the perceived quality of the information they provide.
Anzeige
Taking into account these previous findings, this article aims to further contribute to the knowledge of how destination endorsement by influencers on Instagram affects destination image and user engagement.
3 Methodology
In this study, a mixed methods approach was used to identify quantitative relationships between user engagement and influencer profile characteristics and to conduct a qualitative content analysis of influencers’ cognitive image of travel destinations. Instagram, the most popular social media platform among influencers and particularly suitable for promoting travel destinations [6, 8, 22] served as the online data source.
The entire research process was divided into three steps: (1) data collection, (2) machine learning model with content analysis and (3) statistical analysis.
3.1 Data Collection
The data collection process was based on the data mining tool PhantomBuster, which imitates human navigation on social media to respect anonymity and the general terms of service [23]. The travel influencer profiles were identified using the influencer marketing platform StarNgage and PhantomBuster. Only profiles with at least 1,000 followers were used, based on the observation of Femenia-Serra and Gretzel [4] regarding trending collaborations of DMOs and micro-influencers (1K-3K followers). The sample was narrowed down to 50 influencer profiles that included both verified (public figures or celebrities) and unverified users, and their profile information and posts were extracted using PhantomBuster. The data included the type of user account (e.g. number of followers, verified/unverified) and post information (e.g. number of likes, number of comments, location). Only those posts that included Austria as a location were used, resulting in 5,522 posts. Finally, for the content analysis, the images were extracted from the posts.
3.2 Data Analysis
In the first phase, image embedding was performed using Google's deep learning model Inception v3 to compute the feature vectors for each image [24‐26]. Then, a hierarchical clustering of the high-dimensional image vectors was performed using the cosine distance and Ward's method to identify content-based clusters [27]. The number of clusters was determined in an iterative process. In each iteration step, cluster content analysis was performed until the maximum number of clusters with a satisfactory categorisation result was reached. Next, t-SNE dimensionality reduction technique was adopted to visualize the clusters in a scatter plot [28].
The second phase involved a statistical analysis using stepwise linear regression to determine whether the account type (business, verified), the total number of posts a user makes and the number of followers a user has have an impact on engagement (likes and comments).
Furthermore, an ANOVA with a Bonferroni Post-hoc test was performed to see, if the clusters differ significantly from each other.
4 Results
4.1 Analysis of the Image Content
The preliminary analysis of the clusters showed that 19% of the images could be very homogeneously characterised as “alpine landscape” (see Fig. 1).
Fig. 1.
Preliminary clustering from the dendrogram
Fig. 2.
The nine final clusters
×
×
To gain deeper insights, a further second-level analysis was conducted, which resulted in a total of nine clusters (see Fig. 2). The description of the clusters is shown in Table 1.
4.2 Statistical Analysis
In the second phase of the analysis, descriptive data was created for the influencers’ Instagram accounts (see Table 2). Of the 50 influencers included in the study, half (n = 25) were labelled as business accounts and six accounts were verified.
Table 1.
Content analysis of the clusters.
Table 2.
Influencer sample
Influencer accounts in total
50
Verified accounts
6
Accounts not verified
44
Business accounts
25
Non-business accounts
25
When looking at the influencer classification according to the number of their followers [15] 32% of micro-influencers (up to 10K followers) and 68% of meso-influencers (10k-1M followers) were included in the sample (see Table 3).
Table 3.
Influencers by number of followers
Number of followers
Frequency
Percentage
1K - 9K
16
32
10K - 999K
34
68
Total
50
100
A stepwise linear regression analysis was performed to investigate the impact of account type (business, verified), total number of posts and number of followers on engagement (likes and comments). In our analysis, we checked for the assumptions of multiple regression and no violations were identified. First, the impact on the number of likes as the dependent variable was measured with two regression models (see Table 4). The first model included the independent variables business account and number of posts. For the second model, additional variables verified account and number of followers were added. Both models were significant (p = 0.000), explaining only 4% of the variance in the first model (adjusted R2 = 0.040) but 54% of the variance in the second model (adjusted R2 = 0.542). The second model shows that the number of followers had a strong positive effect (beta coefficient = 0.705), the verified profile had a weak positive effect (beta coefficient = 0.228), the number of posts had a weak negative effect (beta coefficient = − 0.124) and the business account had a weak negative effect (beta coefficient = − 0.042).
Table 4.
Regression model for the number of likes
Model
Non-standardised coefficients
Std. Coeff
t
Sig
B
Std. Error
Beta
Model 1 (adj. r2) = .040
Model 2 (adj. r2) = .542
1
(Constant)
542.964
83.771
6.482
BusinessAccount
−121.586
79.958
−.020
−1.521
postsCount
.821
.057
.195
14.476
2
(Constant)
956.560
61.416
15.575
.000
BusinessAccount
−248.850
56.429
−.042
−4.410
.000
postsCount
−.521
.045
−.124
−11.542
.000
followersCount
.014
.000
.705
66.462
.000
VerifiedAccount
2165.770
90.294
.228
23.986
.000
A second stepwise regression was then performed with the same independent variables, using the number of comments as the dependent variable (see Table 5). Both models were significant (p = 0.000); however, only 1.3% of the variance was explained in the first model (adjusted R2 = 0.013) and only 3.8% of the variance in the second model (adjusted R2 = 0.038). The second model shows that the number of followers had a significant, weak positive effect (beta coefficient = 0.184; p = 0.000), while the number of posts (beta coefficient = − 0.130; p = 0.000) and the business account (beta coefficient = − 0.1374; p = 0.000) had a significant but weak negative effect on the number of comments. The variable Verified Account had no significant effect on the number of comments (p = 0.175).
Table 5.
Regression model for the number of comments
Model
Non-standardised coefficients
Std. Coeff.
t
Sig
B
Std. Error
Beta
Model 1 (adj. r2) = .013
Model 2 (adj. r2) = .038
1
(Constant)
124.134
9.562
12.982
BusinessAccount
− 79.224
9.127
− .118
− 8.680
postsCount
− .019
.006
− .040
− 2.928
2
(Constant)
149.828
10.017
14.958
2
BusinessAccount
–91.424
9.203
− .137
− 9.934
.000
postsCount
− .061
.007
− .130
− 8.357
.000
followersCount
.000
.000
.184
11.991
.000
VerifiedAccount
− 19.959
14.727
− .019
− 1.355
.175
Next, the standardised mean values of the likes and comments of each cluster were compared to see if a significant difference could be found between the clusters1.
The t-SNE projection of the high dimensional image vectors indicates an acceptable overall cluster segmentation (Fig. 3).
Fig. 3.
Scatter plot of the nine clusters
×
The mean values for likes and comments per cluster are presented in Table 6.
Table 6.
Clusters by mean standardised likes and comments
Cluster name
Mean std. Likes
Cluster name
Mean std. Comments
C1
Summer Alps
0,068
C8
Products
0,0048
C2
Winter Alps
0,066
C1
Summer Alps
0,0026
C6
Lakes
0,052
C7
People
0,0024
C3
Urban
0,047
C6
Lakes
0,0020
C9
Diverse
0,035
C3
Urban
0,0016
C7
People
0,034
C9
Diverse
0,0016
C5
Parks
0,033
C2
Winter Alps
0,0016
C4
Cycling
0,025
C5
Parks
0,0011
C8
Products
0,022
C4
Cycling
0,0009
Looking at the mean standardised likes, it can be assumed that the content of the Summer Alps and Winter Alps clusters appealed to users the most. On the other hand, the content in the Bicycle and Products clusters seemed to be the least appealing, with the lowest mean standardised likes. Looking at the mean standardised comments, the content in Products generated the most comments, while the content in the Parks and Cycling clusters generated the least comments.
5 Discussion of the Results
The aim of this study was to examine the cognitive image of Destination Austria based on the content posted by influencers and to assess how different content is received by Instagram users. Furthermore, while this study provides insights based on engagement metrics, it is understood that selecting the most suitable influencer from a destination marketing perspective involves a comprehensive evaluation, including factors like image congruency, target audience, and past collaboration successes.
The content analysis revealed that the most prominent cluster was the mountain landscape, making the Alps the most well-known cognitive image of the Austrian destination as conveyed by influencers. The Alps are more strongly positioned not only as a ski destination in winter, but also as a cycling and hiking destination [29]. The distinction between alpine summer and winter landscapes is clearly visible in the clustering.
The Cycling cluster can be interpreted as the sport most represented by influencers in creating a destination image of Austria. Another cluster that contributes to the destination image of Austria is the Urban cluster, which promotes the sights and attractions of the capital Vienna, such as Schönbrunn Palace, the Vienna State Opera etc.
It can be concluded that influencers contribute to the cognitive image of Austria as a destination mainly through content about the Alps, the capital Vienna and cycling, which seems to be in line with the current marketing objectives. The remaining five clusters contribute rather little to the image of the destination. The most promising among them is the Lakes cluster. Although this cluster contains quite heterogeneous content, the promotion of lake destinations such as Hallstadt and Salzkammergut, Neusiedler See, Wörthersee, Zell am See, etc. could potentially benefit from further influencer marketing campaigns. As mentioned above, travel influencers are very successful in convincing tourists about their choice of destination [8].
5.1 Effects on the Engagement Rate
With regard to the selection of suitable influencers for destination marketing campaigns, the following recommendation can be derived. If DMOs opt for micro-influencers with fewer followers, they should choose active influencers who regularly post relevant content. Regarding meso-influencers, a business account will not receive more likes. DMOs should therefore not perceive a business account label as an advantage when opting for a meso-influencer.
For comments, the situation is more complex, as a higher number of followers does not lead to a higher number of comments. It can be assumed that it takes more effort to achieve a higher level of engagement, e.g. through comments, than just having a wider follower base. This also means that the difference between micro and meso influencers in achieving higher levels of engagement in the form of comments is insignificant. These findings are consistent with some previous studies showing that both meso and micro-influencers seem to be as successful as their prominent online counterparts when it comes to brand promotion [30].
When comparing the clusters based on the standardised number of likes, it can be seen that users like the pictures of the Alps best. In contrast, the lowest number of likes was achieved in the “bicycle” and “products” clusters. Some previous studies have shown that posts with a face or physical representation of an influencer increase user engagement [16, 31]. However, in this case, the cluster “people” that only showed pictures of people did not result in a higher number of likes or comments compared to other clusters. According to a study by Silva et al. [16] influencers try to engage users by seeking direct interaction with their posts by asking questions, inviting comments, etc. It can be concluded that the inclusion of product images in a destination marketing campaign, along with direct invitations to comment, can be beneficial even if the product images appear less visually appealing to users.
According to the standardized number of likes analysis combined with the regression analysis, it can be assumed that when promoting a destination with a more appealing theme and a higher number of likes, such as the Alps and Lake destinations, the use of a micro-influencer would be sufficient. On the other hand, destinations in this case study with themes belonging to the less popular clusters would rather benefit from a meso-influencer or a verified Instagram account as they seem to receive more likes.
6 Conclusion and Implications for Future Research
This research contributes to the current literature by expanding the knowledge base of influencer marketing in the context of the tourism industry. The study provides a deeper understanding of how users are influenced by an influencer's number of followers, account type and posted content. However, it's important to emphasize that while engagement metrics like likes and comments provide valuable data, they are just one facet of influencer suitability. The alignment between marketing objectives, selected content, and influencers who authentically represent this content is crucial. Other factors, such as image congruency between destination and influencer and the influencer's broader brand and audience, also play a pivotal role in determining the success of an influencer marketing campaign. In addition, the study offers a methodological contribution by demonstrating the benefits of machine learning and data mining models in the increasing analysis of social media data.
The results of this study have practical implications for the tourism industry as they provide guidance on what type of influencer would be suitable for a destination marketing campaign. This study clearly shows that the content itself influences the number of likes and comments and should therefore be taken into account when deciding on the optimal influencer type. The alignment between marketing objectives, selected content and influencers who authentically represent this content is of great importance. The business implications of this study show how DMOs can reduce the cost of an influencer campaign by actively using both meso and micro influencers, taking into account the impact of the content on engagement.
One of the limitations of this study is that it was not possible to distinguish between generic posts and paid sponsorships. Therefore, it is unclear whether the results of the clustering are biased. Furthermore, the cognitive image of the destination was analysed solely on the basis of influencer posts, without reference to the image of the destination among tourists. A comparative content analysis of UCG posts and influencer posts could provide additional insights into whether a certain aspect of the destination image is missing or could be improved through an influencer marketing campaign. Furthermore, due to the large scale of the destination under study and the large number of contributing influencer accounts, only a limited number of influencer accounts were included in the research.
Open Access This chapter is licensed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license and indicate if changes were made.
The images or other third party material in this chapter are included in the chapter's Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the chapter's Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.
The comparison of each cluster's mean values of the standardized number of likes and comments (the Kruskal – Wallis and the Post-hoc tests) can be seen here: https://tinyurl.com/ANOVA-Appendix.
