This data contains a record of user interactions with the Entree Chicago restaurant recommendation system. This is an interactive system that recommends restaurants to the user based on factors such as cuisine, price, style, atmosphere, etc. or based on similarity to a restaurant in another city (e.g. find me a restaurant similar to the Patina in Los Angeles). The user can then provide feedback such as find a nicer or less expensive restaurant.
Robin Burke University of California, Irvine Department of Information and Computer Science Irvine, CA 92697Date Donated: March 9, 2000
This data records interactions with Entree Chicago restaurant recommendation system (originally http://infolab.cs.uchicago.edu/entree) from September, 1996 to April, 1999. The data is organized into files roughly spanning a quarter year -- with Q3 1996 and Q2 1999 each only containing one month.
Each line in a session file represents a session of user interaction with the system. The (tab-separated) fields are as follows:
Date, IP, Entry point, Rated restaurant1, ..., Rated restaurantN, End point
A = Atlanta B = Boston C = Chicago D = Los Angeles E = New Orleans F = New York G = San Francisco H = Washington DC
L = browse (move from one restaurant in a list of recommendations to another) M = cheaper (search for a restaurant like this one, but cheaper) N = nicer ( " " , but nicer) O = closer (unused in the production version of the system) P = more traditional (search for a restaurant like this, but serving more traditional cuisine) Q = more creative (search for a restaurant serving more creative cuisine) R = more lively (search for a restaurant with a livelier atmosphere) S = quieter (search for a restaurant with a quieter atmosphere) T = change cuisine (search for a restaurant like this, but serving a different kind of food) Note that with this tweak, we would ideally like to know what cuisine the user wanted to change to, but this information was not recorded.
Some potentially useful data is missing. In many cases, we don't know the starting point because the user input a set of selection criteria (such as "inexpensive traditional Mexican") using a form submission, rather than starting from a known restaurant. These queries were not recorded. This is denoted by a 0 in the entry point field. Some sessions do not have a known end point. This is marked by -1 in the end point field.
In addition to the user's interactions, there is also data linking the restaurant ID with its name and features such as "fabulous wine lists", "good for younger kids", and "Ethopian" cuisine. This data is stored by city (e.g. Atlanta, Boston, etc.) and is in the following format:
restaurant id [tab] restaurant name [tab] restaurant features (3 digits ids separated by spaces)
Burke, R. The Wasabi Personal Shopper: A Case-Based Recommender System. In Proceedings of the 11th National Conference on Innovative Applications of Artificial Intelligence, pages 844-849. AAAI, 1999.
Burke, R. Knowledge-based Recommender Systems. To appear in the Encyclopedia of Library and Information Science.