Machine Learning Contests: A Guidebook

Firstly, it takes common competition scenarios as a guide by giving the main processes of using Machine Learning to solve real-world problems, namely problem modelling, data exploration, feature engineering, model training. And then lists the main points of difficulties, general ideas with solutions in the whole process. Moreover, this book comprehensively covers several common problems in the field of Machine Learning competitions such as recommendation, temporal prediction, advertising, text computing, etc. This book is a systematic introduction to contests in the field of algorithms, not only explaining the theory behind the practice, but also elaborating in detail the guide to scoring and necessary skills needed from various angles, using different cases. Many Deep Learning algorithms have been applying to Natural Language Processing (NLP). The early word embedding model and the subsequent development of convolutional neural networks and recurrent neural networks have played a very important role for this time period, greatly improving the accuracy baseline of the original statistical method and achieving more generalized effects in different fields (such as translation, voice recognition, and other tasks). In the current latest environment, the self-attention mechanism models like transformer structures are applicable to a sea of data, which could then generate training models. This ability further helps natural language processing to develop, which has even achieved baseline scores that exceed what human can get on some tasks. The goal of Natural Language Processing technology is to recognize human language through various electronic machines such as computers, so as to understand human intentions. Natural Language Processing can better free labor from complicated tasks in some specific fields. Take Taobao as an example; dialog systems can analyze and identify customer questions, locate customer needs, and provide answers to corresponding questions or specific operations of the purchase process, thus saving merchants a lot of repetitive labor and time cost. In the vehicle-mounted voice system, the voice recognition system is combined with the natural language processing system to free up the hands of drivers while driving and provide corresponding services (such as wayfinding, playing music, etc.). Target Readers: The potential readers of this book will be divided into three categories: - Those who are interested in algorithmic contests. Interest is the biggest driving force. In order to make algorithm competitions more interesting and diverse, this book adds a lot of expanded and exploratory content, introducing and carrying out practice from multiple directions and in many fields. - Those who want to study Machine Learning and explore practice on algorithms in depth. Taking part in an algorithm competition is the best way for practice, which could enhance understanding of theory. This is also what the book emphasizes. - Those who major in Computer Science. Machine Learning or the Deep Learning algorithm, as a hot career in the computer industry, is worth further study. This book provides a very good explanation to real situations to help readers know how and why. This book systematically introduces the competitions in the field of algorithm and machine learning. The first author of the book has won 5 championships and 5 runner-ups in domestic and international algorithm competitions. Firstly, it takes common competition scenarios as a guide by giving the main processes of using machine learning to solve real-world problems, namely problem modelling, data exploration, feature engineering, model training. And then lists the main points of difficulties, general ideas with solutions in the whole process. Moreover, this book comprehensively covers several common problems in the field of machine learning competitions such as recommendation, temporal prediction, advertising, text computing, etc. The authors, also knew as "competition professionals", will explain the actual cases in detail and teach you various processes, routines, techniques and strategies, which is a rare treasure book for all competition enthusiasts. It is very suitable for readers who are interested in algorithm competitions and deep learning algorithms in practice, or computer-related majors. Cover Front Matter Part I. Half the Work, Twice the Effect 1. Guide to the Competitions 2. Problem Modeling 3. Data Exploration 4. Feature Engineering 5. Model Training 6. Model Integration Part II. Birds of a Feather Flock Together 7. User Profiles 8. Case Study: Elo Merchant Category Recommendation Part III. Learn from History to Create a Bright Future 9. Time Series 10. Case Study: Global Urban Computing AI Challenge 11. Case Study: Corporación Favorita Grocery Sales Forecasting Part IV. Precise Delivery, Optimized Experience 12. Computational Advertising 13. Case Study: 2018 Tencent Advertising Algorithm Competition—Audience Lookalike Expansion 14. Case Study: TalkingData AdTracking Fraud Detection Challenge Part V. Listen to What You Say and Understand What You Write 15. Natural Language Processing 16. Case Study: Quora Question Pairs