Abstract: The success of machine (deep) learning amazes the audience from various domains. Recent efforts on Natural Language Processing (NLP), for instance, transcend humanity with cognitive awareness for a number of predefined tasks. In an active cyber-ecosystem, text content may often accompany rich heterogeneous data context, e.g., user behaviors, knowledge graphs, social networks, and expert judgments, which can be potentially important to further enhance the NLP model performance. From the data science viewpoint, data heterogeneity is being used for innovative discovering, modeling, integrating and sharing knowledge. In this talk, I will share my most recent research work in NLP by leveraging very large scale user behavior and heterogeneous graph data. The proposed joint representation learning algorithms and multitask learning models can be used to address a number of novel NLP applications, e.g., dynamic content generation, pornography product/ebook detection, and cross-domain sentiment analysis.

Short Bio: Dr. Xiaozhong Liu is an Associate Professor at School of Informatics, Computing, and Engineering, Indiana University Bloomington. He also serves as a senior consultant of Alibaba DAMO Academy for NLP Research and Legal AI Research. Over the past 15 years, Dr. Liu's research has primarily focused on human-centered data science in the context of AI, big data, and human computing, which affiliate his work with several closely interrelated research communities: Information Retrieval/Recommendation, Cyberlearning, and Computational Social Computing. Currently, his algorithms are used in a number of eCommerce platforms and search engines more than 150 million times per day and outperform existing methods by 10% on the average. He has filed 8 patents and published more than 80 peer-reviewed articles addressing various audiences and crossing disciplinary boundaries in a wide range of venues, e.g., AAAI, SIGIR, PNAS, ACL, EMNLP, JASIST, and WWW. As the project lead, his recent projects are funded by federal agencies such as NSF and IMLS, and also industry partners.