Introduction to the Ad-papers Project
The Ad-papers project is a comprehensive and dynamic repository focused on the field of computational advertising. This project serves as an evolving library of academic papers, learning materials, and industry insights, offering valuable resources for professionals and enthusiasts in the computational advertising domain. The project not only acts as a personal summary of work done by Wang Zhe but also aims to facilitate accessibility to essential information in this field.
Purpose and Contributions
The primary goal of the Ad-papers project is to continuously curate and update a collection of significant papers and materials related to computational advertising. By doing this, the project not only consolidates the work and studies encountered by its creator, Wang Zhe, but also shares this knowledge with a broader audience, potentially speeding up learning and innovation within the advertising industry.
Wang Zhe invites collaboration and discussion with others passionate about computational advertising through various contact methods, including email, LinkedIn, and the Chinese platform Zhihu.
Notable Inclusions
Among the array of resources included in the Ad-papers project are notable papers and practices from industry leaders and significant contributions in technology and methodology. Examples include:
- Embedding Techniques: Innovative embedding strategies such as the real-time personalization system for search ranking developed by Airbnb in 2018, recognized as the best paper at the 2018 KDD conference.
- CTR Prediction: Alibaba's work on Deep Interest Evolution Network (DIEN) for Click-Through Rate (CTR) prediction showcases the advancement in understanding user interests.
Related Resources
The project also highlights additional resources and collaborations such as:
- RTB Papers: Contributions from Zhang Weinan, providing a comprehensive list of Real-Time Bidding papers.
- Spark-based CTR Models: Implementation of different CTR prediction models using the Spark MLlib framework, accommodating diverse algorithms such as LR, FM, RF, and more.
Sections Overview
The project is organized into several key thematic sections, each focusing on a specific aspect of computational advertising or related technological advances:
Optimization Method
This section covers various optimization techniques essential for enhancing advertising algorithms, including online optimization, parallel SGD, and FTRL among others. It provides a basis for understanding complex models through practical papers such as "Google Vizier: A Service for Black-Box Optimization."
Topic Model
Topic modeling is frequently used for extracting contextual features of advertisements. This section offers resources ranging from probabilistic language models to LDA, facilitating improved ad creativity and optimization.
Google Three Papers
A fundamental part of the big data infrastructure, these three seminal papers—covering HDFS, MapReduce, and BigTable—are pivotal for anyone working with large-scale data processing.
Factorization Machines
The Factorization Machines section provides insights into models that are crucial for computational advertising, featuring a selection of relevant papers and resources.
Embedding Techniques
Discussed here are various embedding methods used to enhance personalization and recommendation systems, including Word2Vec and Item2Vec methodologies.
Budget Control
This section delves into strategies for pacing and budget control in advertising systems, pivotal for effective management of ad campaigns and resource allocation.
Tree Model
Tree models and their boosting derivations are explored here. They show impressive performance across numerous prediction tasks, including CTR and CVR estimations.
Guaranteed Contracts Ads
This addresses online allocation and pricing strategies in contract-based advertising systems, presenting dynamic pricing models and yield optimization challenges.
Classic CTR Prediction
Finally, the project includes classic methods for predicting click-through rates, such practice is essential for improving the effectiveness of online advertisements.
In summary, the Ad-papers project is an invaluable asset for anyone interested in computational advertising, offering a wealth of information and a platform for engaging in discussions and collaborations within this rapidly evolving field.