Track1 task: Personalized recommendation system in social networks

According to the user attributes (User Profile) in Tencent Weibo, SNS social relationships, and interaction records in social networks ( retweet, comment, at), etc., as well as the historical item recommendation records in the past 30 days, to predict the list of recommended items that are most likely to be accepted by users

KDD Cup 2012 winner:

Champion: ACMClass@SJTU (ACM Class of Shanghai Jiao Tong University)

Runner-up: Shanda Innovations (Shanda Innovation Institute)

Third-place: SYSU_Wargreymon

Track2 task : pTCR click-through rate estimation of the search advertising system

Provides users’ query terms (query) searched in Tencent, displayed advertising information (including advertising title, description, URL, etc.), and the relative position of the advertising ( Ranking in multiple ads) and user clicks, as well as advertiser and user attribute information, to predict user clicks on ads in the subsequent time

KDD Cup 2012 Track2 winner:

Champion: Catch Up (National Taiwan University)

Runner-up: Opera Solutions (Opera Inc.)

Third-place: Steffen Rendle Track1 Task: Music Rating Prediction

< p>According to the user's historical rating record of the item on Yahoo Music, predict the difference between the user's rating of other items (including songs, albums, etc.) and the actual rating RMSE (minimum mean square error). At the same time, information such as the album, singer, and genre of the song is provided

KDD Cup 2011 Track1 winner:

Winner: National Taiwan University (National Taiwan University)

p>

Runner up: Commendo (Commendo Inc.)

Third runner up: InnerPeace (Shanghai Jiao Tong University)

Track2 task: Identify whether music has been rated by users

Each user provides 6 candidate songs, 3 of which have been rated by the user, and the other 3 are songs that have not been rated by the user but are from songs with higher overall ratings among users. Song attribute information (album, singer, genre, etc.) is also provided. Participants give binary classification results (0/1 classification), and the final ranking is calculated based on the overall accuracy

KDD Cup 2011 Track2 winner:

Winner: National Taiwan University ( National Taiwan University)

First Runner-up: The Art of Lemon (Chinese Academy of Sciences)

Second Runner-up: Commendo (Commendo Inc.) Based on the interaction logs between the intelligent teaching tutoring system and students, Predicting students' test scores on mathematics questions. This task is of both practical importance and scientific interest. The competition provides 3 development data sets and 2 challenge data sets, and each data set is divided into a training part and a test part. The test part of the Challenge data set is hidden, and contestants need to develop a learning model to accurately predict the performance of this hidden part

KDD Cup 2010 winner:

Champion: National Taiwan University

First Runner-up: Zhang and Su

Third Runner-up: BigChaos@KDD (Commendo Inc. and AT&T Labs) French telecom operator Orange’s large-scale data center , accumulated a large number of customer behavior records. Competitors need to design a good customer relationship management system (CRM) and use a fast and stable method to predict three-dimensional attributes of customers, including:

1. Loyalty: the possibility of users switching operators (Churn);

2. Appetency: the possibility of purchasing new services (Appetency);

3. Value-added: the possibility of customers upgrading or purchasing additional high-profit products (Up-selling).

Results are evaluated using AUC curves

KDD Cup 2009 winner:

Track1 (Fast Track):

Winner: IBM Research (IBM Research)

First Runner-up: ID Analytics Inc.

Third Runner-up: Old dogs with new tricks (Professor David Slate, Professor Peter W. Frey, Northwestern University, USA)

Track2 (Slow Track)

Winner: University of Melbourne

Runner-up: Financial Engineering Group, Inc. Japan

Third-place: National Taiwan University An important application of medical imaging is the detection of breast cancer. Every year, 465,000 female patients around the world are killed by breast cancer. However, after the introduction of X-ray detection in 1990, signs of the disease can be successfully detected and treated in the early stages, successfully reducing the disease mortality rate by 30%. The traditional method of using radiologists to detect X-rays is time-consuming and labor-intensive, and in order to improve the accuracy of detection, multiple doctors are often required to repeatedly interpret the images.

Based on this background, this year’s competition is divided into two tracks, one is to design a computer-aided detection system (Computer-Aided Detection, CAD) to judge whether images contain signs of breast cancer; and the other is to design a two-part system. Classifier (binary classification) to determine whether a sample requires repeated interpretation by a doctor.

KDD Cup 2008 Winner:

Track1:

Winner: IBM Research (IBM Research)

Runner-up: National Taiwan University (National Taiwan University)

Second runner-up: Wayne State University (Wayne State University, USA)

Track2:

Winner: IBM Research (IBM Research)

First runner-up: TZ Team

Second runner-up: National Taiwan University