Layer 1: Places (city, district?), no inter-connection but edges to Layer 1 and Layer 2 Layer 2: 1 node for every user with inter-connections (user-user edge with annotation for Similarity Measure 1) and connections to layer 2 (visited places) as well as to layer 3 (hotels stayed at, with annotation for reviews containing aspect ratings) Layer 3: Hotels, no inter-connections but edges to Layer 2 and Layer 1

Collaborative filtering for every hotel: Matrix containing all users with at least 5 reviews x aspects of the reviews => Cosine Similarity

For all hotels a user has visited calculate:

1. For every aspect: deviation of a user's rating to the average rating of all users
2. Average hotel class (+variance) + price range

• Crawl city names / location data
• Crawl hotel class