The project concentrates on developing new applications of machine learning for the digital cruise advisor Cruisewatch. A company which offers cruise trips from various companies on a webpage. Cruisewatch maintains a huge database of (meta-)information about upcoming and historical trips. Accordingly, it allows to apply machine learning in various directions. Current developed algorithms include trip price prediction, customer behavior analysis and customer review analysis which are helpful for travel recommendation.
The project consists of travel price forecasting, customer behavior analysis and customer rating analysis.
Price Forecast: A model based on more than seven years of price history has been developed to predict the sales price of cruises in the weeks leading up to departure. After analyzing the sales price provided by Cruisewatch, the model analyzed for which constellations and dimensions the price changes differently. Therefore the model is trained separately for these groups and the results are optimized accordingly.
Analysis of customer behaviour: Based on the existing customer information and behaviour (e.g. which trip / ship / shipping company the customers prefer) the algorithm provides recommendations for further trips. In order to be able to recommend suitable trips to new customers, an interactive tool was developed which finds the most relevant offers very efficiently, i.e. with as few questions as possible. In addition, it also examines which characteristics of cruise ships are particularly important for customers in order to adapt the consulting process accordingly.
Customer rating analysis: Based on customer ratings for their cruises, features are extracted to automatically determine whether the ratings are positive or negative. In addition, the feedback is presented by aggregating it in clusters in such a way that the customer gets an overview of the advantages and disadvantages of a cruise with little effort.