AI can be an extremely useful tool for improving DMO marketing efforts. This opinion piece focuses on recent research into developing a visual classifier model for analysing visual social media content.
The DTTT launched a new podcast series entitled Beyond Tourism's Horizon, aimed at building bridges between academia and the tourism industry. This series raises awareness of the capabilities of new technologies that can improve operational efficiency and tourism experiences, enabling the constant development of the sector and its offerings.
This first episode is with Lyndon Nixon - an Assistant Professor in Applied Data Science at MODUL University in Vienna. He is also the Chief Technology Officer at the spin-off research centre Modul Technology, which focuses on AI and data science. With his background in Semantic Web and media annotation, Lyndon has developed a visual classifier for destination brands using images. This can enable tourism stakeholders to better understand how their destination is being communicated visually, for example on social media.
Text is a media that computers can interpret in a relatively straightforward manner. There has already been a significant volume of research conducted using text analysis from TripAdvisor to identify a destination's image and the characteristics associated with them. However, it is much more complex to automatically analyse the meaning behind images and videos. Consequently, visual elements from consumer reviews are less understood by the tourism sector. Lyndon's research aims to identify how computers can better understand non-textual elements, and how AI can be employed to describe photos and videos (which can be considered as a time-based moving image). With the emergence of AI and neural networks over the last 10 years this enables better accuracy in such a sentiment analysis. User-Generated-Content (UGC) is now a significant contributor to destination images, and social media has become increasingly important as an information source; which has overtaken more traditional forms of travel media, such as brochures. Lyndon has looked into how to extract knowledge based on visuals from Instagram content as one of the most popular social networks.
Lyndon highlights how the research conducted into visual classifiers can benefit destination managers and marketers. He explains that various AI models are available online that can be connected directly through APIs or downloaded. He explains that these AI models are used to label photos and cluster them into various categories.
However, there is a gap in moving from off-the-shelf solutions to organisations knowing and being able to adapt or fine-tune these AI models to the actual situation of their destination. For example, the dataset from ImageNet has approximately 1,000 labels, but doesn't have a label for deserts, which means that the AI analysis instead highlights specific objects contained within the picture, such as camels or 4X4s. He highlights that the key barrier for destinations to implement these AI models is the need for additional training. He outlines the process that destinations need to follow in order to train models for application to destination-specific requirements:
In order to evaluate destination image and branding, it is important to capture all of the attributes related to each destination. However, there is no definitive list of what attributes to focus on. To analyse photos, labels need to be exclusive and clearly defined. Therefore, training the data set can be a time-consuming process.
Lyndon developed 18 key labels in his classifier model based on an aggregation of user and expert surveys. He explains that his visual classifier model looks at the cognitive and functional elements that are visible through photos. Therefore, aspects such as emotions require an extra level of analysis. Lyndon also explains that individual differences aren't taken into account by AI tools. Consequently, a room in a Turkish cave hotel may not get labelled under the category of accommodation as it does not have the same appearance as a traditional hotel room.
The projected images are those characteristics that the destination intends to promote, while the perceived images are how the destination is viewed by tourists and can be observed in UGC. Tourists only post certain aspects of a destination and these are the associations considered as being the most important to them. Comparing projected and perceived destination images helps to understand if marketing activities are working and brand messaging is aligned with tourist perceptions of the destination. However, the differences between UGC and DMOs projected images can be more important as they can highlight aspects of a destination that may be missed in traditional marketing materials.
Lyndon explains that his research analysed a destination image based on a vector to compare and visualise the representation of destinations on Instagram. For instance, Bora Bora and the Maldives were grouped together as coastal destinations, while New York and New Orleans were urban destinations. New Orleans was highlighted strongly in terms of being a shopping and gastronomy destination, which should be incorporated into marketing activities.
With the rise of Instagrammable experiences, the social media site has created new instances of overtourism. Some unknown locations have therefore risen to prominence as a result of UGC. Destinations have multiple attributes and it's important for DMOs to identify whether there is too much reliance on promoting specific attributes. For instance, Bali has a higher proportion of urban infrastructure than in the Maldives and Bora Bora. This in turn may help highlight the diversity of the destination as more than just a beach resort. Therefore, DMOs must consider the balance of the destination's promotion across multiple attributes and not just on bucket list attractions to attract more diverse audiences and help disperse visitors throughout the destination.
Organisations should consider how new technologies can benefit them, which requires investment in time and resources to evaluate the different solutions and their usefulness. There are lots of exciting opportunities provided by AI for destination management. For example, AI can be used to solve overcrowding in destinations by automatically monitoring video livestreams to calculate how many people are at specific attractions. Therefore, DMOs would have sufficient data for estimating crowd levels at specific times and therefore be able to develop and implement strategies to reduce pressure on specific attractions and better disperse tourists throughout the destination to improve experiences.
Research knowledge tends to stay within the same communities despite the desire for it to be commercialised and benefit the industry. The increasing usage of open-source data within AI research means that research outcomes can be shared freely and built upon by others. Lyndon highlights Huggingface.co where his destination image classifier model is available to download.
Lyndon explains that the real test in terms of AI's value is only in its real-life usage and to see the practical usefulness for decision-making and updating destination strategies. Companies regularly provide new functionalities, and DMO dashboards may be used for social media analytics that already includes elements of visual classifiers. However, companies only add new functionalities when there is a proven high accuracy, meaning the latest research takes a few years to be applied to actual business scenarios.
DMOs are small and have tight budgets, but numerous opportunities abound for them to reach out to universities. Lyndon highlights the innovation grant in Austria for collaboration between industry and researchers on feasibility studies for new technological developments and prototyping. Organisations need to consider these innovation programmes and benefit from external knowledge and APIs to pilot and test whether there will be a sufficient return on their investments.
The DTTT launched a new podcast series entitled Beyond Tourism's Horizon, aimed at building bridges between academia and the tourism industry. This series raises awareness of the capabilities of new technologies that can improve operational efficiency and tourism experiences, enabling the constant development of the sector and its offerings.
This first episode is with Lyndon Nixon - an Assistant Professor in Applied Data Science at MODUL University in Vienna. He is also the Chief Technology Officer at the spin-off research centre Modul Technology, which focuses on AI and data science. With his background in Semantic Web and media annotation, Lyndon has developed a visual classifier for destination brands using images. This can enable tourism stakeholders to better understand how their destination is being communicated visually, for example on social media.
Text is a media that computers can interpret in a relatively straightforward manner. There has already been a significant volume of research conducted using text analysis from TripAdvisor to identify a destination's image and the characteristics associated with them. However, it is much more complex to automatically analyse the meaning behind images and videos. Consequently, visual elements from consumer reviews are less understood by the tourism sector. Lyndon's research aims to identify how computers can better understand non-textual elements, and how AI can be employed to describe photos and videos (which can be considered as a time-based moving image). With the emergence of AI and neural networks over the last 10 years this enables better accuracy in such a sentiment analysis. User-Generated-Content (UGC) is now a significant contributor to destination images, and social media has become increasingly important as an information source; which has overtaken more traditional forms of travel media, such as brochures. Lyndon has looked into how to extract knowledge based on visuals from Instagram content as one of the most popular social networks.
Lyndon highlights how the research conducted into visual classifiers can benefit destination managers and marketers. He explains that various AI models are available online that can be connected directly through APIs or downloaded. He explains that these AI models are used to label photos and cluster them into various categories.
However, there is a gap in moving from off-the-shelf solutions to organisations knowing and being able to adapt or fine-tune these AI models to the actual situation of their destination. For example, the dataset from ImageNet has approximately 1,000 labels, but doesn't have a label for deserts, which means that the AI analysis instead highlights specific objects contained within the picture, such as camels or 4X4s. He highlights that the key barrier for destinations to implement these AI models is the need for additional training. He outlines the process that destinations need to follow in order to train models for application to destination-specific requirements:
In order to evaluate destination image and branding, it is important to capture all of the attributes related to each destination. However, there is no definitive list of what attributes to focus on. To analyse photos, labels need to be exclusive and clearly defined. Therefore, training the data set can be a time-consuming process.
Lyndon developed 18 key labels in his classifier model based on an aggregation of user and expert surveys. He explains that his visual classifier model looks at the cognitive and functional elements that are visible through photos. Therefore, aspects such as emotions require an extra level of analysis. Lyndon also explains that individual differences aren't taken into account by AI tools. Consequently, a room in a Turkish cave hotel may not get labelled under the category of accommodation as it does not have the same appearance as a traditional hotel room.
The projected images are those characteristics that the destination intends to promote, while the perceived images are how the destination is viewed by tourists and can be observed in UGC. Tourists only post certain aspects of a destination and these are the associations considered as being the most important to them. Comparing projected and perceived destination images helps to understand if marketing activities are working and brand messaging is aligned with tourist perceptions of the destination. However, the differences between UGC and DMOs projected images can be more important as they can highlight aspects of a destination that may be missed in traditional marketing materials.
Lyndon explains that his research analysed a destination image based on a vector to compare and visualise the representation of destinations on Instagram. For instance, Bora Bora and the Maldives were grouped together as coastal destinations, while New York and New Orleans were urban destinations. New Orleans was highlighted strongly in terms of being a shopping and gastronomy destination, which should be incorporated into marketing activities.
With the rise of Instagrammable experiences, the social media site has created new instances of overtourism. Some unknown locations have therefore risen to prominence as a result of UGC. Destinations have multiple attributes and it's important for DMOs to identify whether there is too much reliance on promoting specific attributes. For instance, Bali has a higher proportion of urban infrastructure than in the Maldives and Bora Bora. This in turn may help highlight the diversity of the destination as more than just a beach resort. Therefore, DMOs must consider the balance of the destination's promotion across multiple attributes and not just on bucket list attractions to attract more diverse audiences and help disperse visitors throughout the destination.
Organisations should consider how new technologies can benefit them, which requires investment in time and resources to evaluate the different solutions and their usefulness. There are lots of exciting opportunities provided by AI for destination management. For example, AI can be used to solve overcrowding in destinations by automatically monitoring video livestreams to calculate how many people are at specific attractions. Therefore, DMOs would have sufficient data for estimating crowd levels at specific times and therefore be able to develop and implement strategies to reduce pressure on specific attractions and better disperse tourists throughout the destination to improve experiences.
Research knowledge tends to stay within the same communities despite the desire for it to be commercialised and benefit the industry. The increasing usage of open-source data within AI research means that research outcomes can be shared freely and built upon by others. Lyndon highlights Huggingface.co where his destination image classifier model is available to download.
Lyndon explains that the real test in terms of AI's value is only in its real-life usage and to see the practical usefulness for decision-making and updating destination strategies. Companies regularly provide new functionalities, and DMO dashboards may be used for social media analytics that already includes elements of visual classifiers. However, companies only add new functionalities when there is a proven high accuracy, meaning the latest research takes a few years to be applied to actual business scenarios.
DMOs are small and have tight budgets, but numerous opportunities abound for them to reach out to universities. Lyndon highlights the innovation grant in Austria for collaboration between industry and researchers on feasibility studies for new technological developments and prototyping. Organisations need to consider these innovation programmes and benefit from external knowledge and APIs to pilot and test whether there will be a sufficient return on their investments.
As part of DTTT's joint Research Membership with IFITT, we have launched a new podcast series entitled Beyond Tourism's Horizon. This series is aimed at helping to build bridges between academia and the tourism industry and to raise awareness of the capabilities of new technologies that can improve operational efficiency and tourism experiences, enabling the constant development of the sector and its offerings.
This podcast focuses on recent research into developing a visual classifier model for analysing visual social media content.
As part of DTTT's joint Research Membership with IFITT, we have launched a new podcast series entitled Beyond Tourism's Horizon. This series is aimed at helping to build bridges between academia and the tourism industry and to raise awareness of the capabilities of new technologies that can improve operational efficiency and tourism experiences, enabling the constant development of the sector and its offerings.
This podcast focuses on recent research into developing a visual classifier model for analysing visual social media content.
As part of DTTT's joint Research Membership with IFITT, we have launched a new podcast series entitled Beyond Tourism's Horizon. This series is aimed at helping to build bridges between academia and the tourism industry and to raise awareness of the capabilities of new technologies that can improve operational efficiency and tourism experiences, enabling the constant development of the sector and its offerings.
This podcast focuses on recent research into developing a visual classifier model for analysing visual social media content.
.webp)