Short Bio: Dr. Dou Shen, Senior Vice President at Baidu, responsible for Baidu App, Search, News Feed, Haokan (Short Videoplatform), etc. His research interests include information retrieval, data mining and artificial intelligence. Before joining Baidu, Dr. Shen cofounded Buzzlabs, which was acquired by CityGrid. He published dozens of papers over TOIS, TKDE, IJCAI, AAAI, ICML, SIGIR, and so on. He was the vice general chair for SIGKDD 2012 and Industry PC chair for CIKM 2012 and SIGKDD 2016.
Title: Latest Practice in Newsfeed Recommendation Engine
Abstract: Recommendation systems have become common in recent years, which help users discover information and settle on choices. Baidu newsfeed, a personalized news recommendation engine, is one of the core components in Baidu App and is serving hundreds of millions of users every day. This talk will share the latest practice in Baidu newsfeed, In which I will first show the overview of Baidu newsfeed recommendation engine, and then introduce the main techiques in user modeling, content modeling and recommendation algorithms including collaborative filtering , pointwise/listwise ranking, user interest exploration etc.
Short Bio: Thorsten Joachims is a Professor in the Department of Computer Science and in the Department of Information Science at Cornell University, and he is an Amazon Scholar. His research interests center on a synthesis of theory and system building in machine learning, with applications in information access, language technology, and recommendation. His past research focused on counterfactual and causal inference, support vector machines, text classification, structured output prediction, convex optimization, learning to rank, learning with preferences, and learning from implicit feedback. He is an ACM Fellow, AAAI Fellow, and Humboldt Fellow.
Title: Learning Recommendation Policies from Logged Interventions
Abstract: Placing an item into a user's information feed is an active intervention for which we can measure the effect (e.g. click, dwell time). This intervention now only consists of the choice of item, but also how the item is presented (e.g. composition of heading and summary). Training from logs of such intervention-effect pairs is fundamentally different from conventional supervised learnings, since we only observe``bandit feedback'' limited to the particular intervention chosen by the system. We don't get to see how the user would have responded, if we had chosen a different intervention. Such logged bandit feedback is plentiful not only in information-feed applications, but lies at the heart of virtually all intelligent information systems (e.g. recommender systems, ad placement). In this talk, I will explore learning methods for batch learning from logged bandit feedback (BLBF). Following the inductive principle of Counterfactual Risk Minimization for BLBF, this talk presents an approach to training linear models and deep networks from propensity-scored bandit feedback. Moreover, the talk will comment on the particular challenges in designing machine-learning approaches for information-feed applications.
Short bio: Jie Tang is a Full Professor and the Vice Chair of the Department of Computer Science and Technology at Tsinghua University. His interests include data mining, social networks, knowledge graph, machine learning, and artificial intelligence. He has been visiting scholar at Cornell University, Hong Kong University of Science and Technology, and Southampton University. He has published more than 200 journal/conference papers and holds 20 patents. His papers have been cited by more than 12,000 times. He served as PC Co-Chair of CIKM’16, WSDM’15, Associate General Chair of KDD’18, and Acting Editor-in-Chief of ACM TKDD, Editors of IEEE TKDE/TBD and ACM TIST. He leads the project AMiner.org for academic social network analysis and mining, which has attracted more than 10 million independent IP accesses from 220 countries/regions in the world. He was honored with the UK Royal Society-Newton Advanced Fellowship Award, CCF Young Scientist Award, NSFC for Distinguished Young Scholar, and KDD’18 Service Award.
Title: Intelligent Feed Recommendation
Abstract: Recommendation in feed systems has faced several challenges. First, cold-start problems are extremely challenging, in particular with very limited training data in the feed systems. Second, the displayed content might be heterogeneous including text, image, audio, video, or their combination. To address the heterogeneity, we propose a unified framework to address this problem. The framework supports both transductive and inductive learning. We also give the theoretical analysis of the proposed framework, showing its connection with previous works and proving its better expressiveness. To address the limited number of training data, we combine the scenario-specific learning with a model-agnostic sequential meta-learning and unify them into an integrated end-to-end framework, namely Scenario-specific Sequential Meta learner (or s2Meta ). Extensive experiments on various real-world datasets demonstrate that our proposed model can achieve significant gains over the state-of-the-arts in online recommendation. Deployment is at the Guess You Like session, the front page of the Mobile Taobao.
Copyright © 2018 Intelligent Information Feed - All Rights Reserved.