When the Netflix approach doesn’t work: Personalized Travel Recommendations

The Challenge

The customer needed to personalize hotel and travel offers to their customers across their entire portfolio. Crucially, very little information about the customers was available. In particular no Login was being captured, so typical recommender approaches didn’t quite work.

Results

Kineo.ai implemented a machine learning Algorithm using Factorization machines that was able to provide highly personalized travel recommendations, based on very few features such as place of origin, number of people and travel destination and duration. In addition we could make use of contextual features, such as location, week day and time of year.

The project leverages big data level amounts of data through the usage of Spark and a S3 Data Lake.