The use of AI has increased rapidly in the recent times and much of it is due to enhanced hardware at low cost. Expedia has also realized the potential of AI and thus has decided to use it to help hotel put the right photos in front of right people. According to a research conducted by Expedia, travellers spend less than a second on deciding the perfect place for them. So, the company wants to make sure that each image is of high quality to impress customers.
According to Nuno Castro, director of data science at Expedia,” The human brain is designed to process images first. People will right away know whether they like that image or not. Expedia will be using image recognition and deep learning to develop a system that will automatically choose the appropriate hotel images and decide which one will be visible to which customers and when. He also said that” The order of image also plays an important role especially when the viewer is in rush. If all the pictures describe the same thing then the viewer is most likely to shift to another hotel.”
According to Expedia, they have more than 10 million photos of over 300,000 affiliated networks. Previously, Expedia used to select the first photo for each hotel manually. The other photos were randomly sorted (or grouped by what they picture) which resulted in less customer satisfaction. But, this method is neither practical nor beneficial in the longer run. For example, you may want to see the image of the bathroom to make sure it is according to your need but all you can see are the bedroom images in first few photos. This will likely to make you shift to other hotels.
The use of AI will mean that customers will see more relevant photos first. The company is also working on deep learning models which will tailor images to customer types. It means that business travellers will see the gym photos and the family traveller will see the pool images first. The company has used NVIDIA Tesla GPU accelerators in the Amazon Cloud and GeForce GTX GPUs locally.
The company first built the dataset of more than 100,000 Expedia images via crowd sourcing. Then they taught a convolutional neural network to classify each image into one of more than 800 image categories.
Expedia is surely looking to expand its business using the latest technology which will enhance the user experience.
Nuno Castro - Ranking hotel images using deep learning
Video Source: PyData