Can Machine Learning Improve Your Next Holiday?

The Myplanet Concepts team focuses on applying emerging technologies to real-world situations. For our latest project, we applied visual recognition tools to the travel sector. Our aim was to explore ways to ease travel planning stress using AI — in this instance, a recommendation engine for vacation activities.

Challenge

Most avid travellers will tell you the joy of travel is not in the planning of a trip, but rather in the trip itself. Searching for the best flight prices, gathering activity information from different sources, and booking itineraries to match specific interests is a time-consuming, often tedious, task. And as people hopscotch through various sites, travel companies lose out on customers. Users scramble to find relevant information from any source available, instead of finding the one applicable to them.

We wanted to figure out how travel marketers might better match travel itineraries to customers. Could we solve this potential lost revenue stream? And was there a way for machine learning to help us get a clearer grasp of their interests?

Concept Overview

Indulging in local delicacies? Communing with nature? Rocking out to a favourite band? The things we share regularly on social media say a lot about our preferences. As such, the daily life of an individual can be a great predictor of activities they might enjoy on a vacation. Based on this, we prototyped a system that seeks to understand the travel interests of customers, analyzing their social profile and making relevant suggestions for vacation activities and tours.

Key Workflow 



To make the platform work, we began by creating archetypal users using data from Instagram. These archetypes became the foundation of the travel profiler’s training data. Classifications were then established via IBM’s visual recognition technology. Our aim was to help the system understand what a travel persona consisted of (such as a passionate foodie or an enthusiastic nature lover).

Once the archetypes were in place, we were able to experiment with real world users.


For a new user, the first step is to set-up basic parameters about their travel preferences, such as budget and pace.



The user’s Instagram account is then analyzed via the IBM visual recognition service and compared to the initial training set. Finally, based on the priority level assigned to the categories, the travel profiler predicts the user’s interests and makes recommendations.

Flexibility in travel planning is key, so we built in the option for users to adjust the settings as they go. This allows a user to further refine their preferences and influence the recommendations.

Beyond Convenience

For the consumer, the time saving benefits of an AI-powered travel planner are clear. But there is also a clear business opportunity for travel marketers and suppliers. A preference-driven system like the travel profiler is an ideal learning ground for marketers. It can allow them to optimize their offerings by more accurately matching trip itineraries to traveller preferences.

Increased Personalization

With enough aggregated data, the system can get smarter over time and make more nuanced recommendations. For instance, users with concerts documented on Instagram may start out with a basic “music lover” profile. But as it learns, the system may move from general festival or concert recommendations to more genre or artist-specific events.

More Accurate Predictions

Having a massive data set to work with also means we can gather insights, such as optimal activity combinations for specific personas. The system could become more sophisticated at activity pairing.

This could be especially useful for non-obvious connections. For example, the system may connect culinary enthusiasts with adrenaline-filled outdoor activities, due to their naturally adventurous spirit. This type of insight could inform future system predictions on what may explicitly or implicitly appeal to a specific travel persona.

Highlighting Unexpected Correlations

An algorithmically-driven system could also reveal insights on market demands that might be missed by humans. A surge in visitors to Rome for sport-centred activities may signal the surprise emergence of a new mecca for sport lovers, for example. This type of information is especially relevant for travel companies, who could use it to adjust or build upon their activity offerings.

Travel planning continues to be a pain point for users, so it’s easy to see the benefits of an AI-powered travel profiler. And for an online travel agency or direct marketing site, more accurate recommendations and a reduction in workload can create major opportunities. AI solutions like the travel profiler have the potential to generate higher customer conversions and a lower rate of cross shopping.

Interested in the other innovative work we have been doing with Watson? Fascinated by what the shifting landscape of big data can do? You can reach us here to find out more about how we can apply the latest in smart tech to improve your business.

Written by

Fiona Chung

Fiona Chung

Sign up for our newsletter