22th edition: 115 teams from 86 universities in 28 countries

Create a recommendation model for a bookstore

In times of the corona pandemic, online trading has continued to gain in importance. Product recommendations are helpful for optimally advising customers in front of the screens. On the basis of the click and transaction data, a recommendation system in the online shop calculates relevant product recommendations that the customer will in all probability like.

This year the DATA MINING CUP is dedicated to this scenario. Students worldwide are asked to help a bookseller and therefore develop a model that calculates suitable book recommendations for customers. The recommendation model, which achieves the highest customer acceptance with its suggestions in the online shop, wins the DMC 2021.


Before the pandemic, Johannes Gutenberg managed a flourishing little book shop in the historic city center of Mainz. He took great joy in building personal relationships with each of his clients, recommending books catered to their personal taste and assisting in widening their literary palette. In his city and beyond he developed a formidable reputation with a respectable base of loyal customers, who considered him more of a connoisseur than a traditional salesman.

Unfortunately, this loyal base of customers is not enough to make his business profitable. And so, like many traditional retailers, Johannes also relies on walk-in customers.

At the beginning of the pandemic, this source of revenue vanished. To keep his employees and cover ongoing costs, Johannes had to find an alternative form of revenue.

With great initial reservation, he decided to expand his business by launching an online shop, which he believed would save his beloved business from imminent bankruptcy.

At first, Johannes and his employees tried their best to provide suitable recommendations for every product manually. But as the number of products increased and associates worked to keep at least some personal contact to clients via phone and email, this manual process was just not feasible.

Today, Johannes is looking for a reliable recommendation system to provide a targeted recommendation to every product page. This solution should meet his high personalization standards and only require a small amount of manual support to implement.

Since Johannes is only interested in the best process possible, he decides to organize a contest to identify the best recommendation solution.


The goal for each participating team is to create a recommendation model based on historical transactions and item features. For any given product, the model should return its five best recommendations. In order to create a recommender model, the participants are provided with historical transaction and descriptive item data in the form of structured text files (.csv).

The data is provided in three individual files. One file containing the transactions (“transactions.csv”) one the descriptive item data (“items.csv”) and the final one (“evaluation.csv”) containing the template for the result submission.

The solution file must match the specifications described in the task. Incorrect or incomplete submissions cannot be assessed.

The solution file must be uploaded as a structured text file (csv) to the DATA MINING CUP website:

Please make sure that the mandatory boxes on the form are correctly and fully completed before uploading the data.

The name of the text file consists of the team’s name and the file type:

.csv” (e.g. TU_Gutenberg_1.csv)

The team’s name was communicated to the team leaders when their registration was confirmed.




First Place:

Team Uni_Geneva_2
University of Geneva,

Prize: 2,000.00 EUR

Second Place:

Team Uni_Frankfurt_AS_1
Frankfurt University of Applied Sciences, Germany

Prize: 1,000.00 EUR

Third Place:

Team Uni_Aalto_1
Aalto University School of Science and Technology, Finland

Prize: 500.00 EUR

The best teams present their solutions