Abstract: The Web of Data is growing permanently as well as the industrial data sets. Induced by this movement the challenge for retrieving knowledge from such data sets has gained much importance in research and industry. Question Answering (QA) is tackling this challenge by providing an easy-to-use natural language interface for retrieving knowledge from large data sets comparable to well-known industrial solutions like Siri, Cortana, Google Now, or Wolfram Alpha, but applicable to common data sets or enterprise data.
Currently there are two main lines of implementing such systems: first, establish a QA system using a monolithic approach, secondly using a modular approach. The advantage of the later is the enabling of researchers to focus on and improve their particular research topics (e.g., entity linking) which will lead to improved quality. The disadvantage of the approach is the investment into infrastructure required to later provide the benefits. Luckily the research community was already provided by such a framework for QA systems (by the WDAqua project) called Qanary. It uses semantic technology (i.e., SPARQL and a vocabulary-driven approach) to represent the research challenges as RDF knowledge base. Additionally, it provides easy-to-use interfaces and a set of tools supporting the implementation and quality measuring (including frontends).
In the tutorial participants will be provided shortly by the foundations of creating Question Answering systems. Thereafter, the Qanary methodology and the equally named technical framework is used to implement an interactive Question Answering system capable answering natural-language questions on DBpedia.
Short Bio: Dr. Andreas Both is a computer scientist dedicated to applied research in an industrial context. He received a PhD in Computer Science for his research on component-based systems and service-oriented architectures at the University of Halle, Germany. At Web-driven companies he has worked for many years in leading research and development positions on different aspects of modern Web technologies. In particular, data-driven processes, data integration, information retrieval applications and web engineering topics are important topics. Currently, he is Head of Architecture, Web Technology and IT Research at DATEV eG a German company dedicated to business software ranking in the top 5 of German IT companies w.r.t. company size. He commits himself to advance in using the World Wide Web as knowledge base and developing the next generation of Web applications to open the capabilities of the WWW for both industry and users.
Abstract: Knowledge graph (KG) was first proposed by Google in 2012 to provide semantic search capabilities on the Web scale. On the meanwhile, chatbots have caught wide attentions from both academia and industry. Most well-known IT players like Google, Facebook, IBM, Microsoft, Amazon and Baidu are releasing bot frameworks or platforms for developers to build and deploy 3rd party bots. Knowledge graphs can play a big role to help understand user request, dialog management and even guide response generation. In this tutorial, we will first give a brief introduction of the chatbot history including describing Baidu DuerOS, Google Assistant, Facebook Messenger M, IBM Watson, Microsoft LUIS and Amazon Alexa in details. We then list the possible scenarios KG can be used as well as the challenges we will face when applying KG in the context of chatbots. Thirdly, we explain several relevant work focusing on representation learning for KG and question answering over KG. Finally, we will provide a hands-on session to build KG-based chatbot (a simplified version of Xiaobai, an emotional companion robot developed by Gowild Robotics Co. Ltd) step by step.
Short Bio: Dr. Haofen Wang was graduated from Apex Data & Knowledge Management Lab, Shanghai Jiao Tong University. He is now the CTO of a well-known AI startup company focusing on chatbots called Shenzhen Gowild Robotics Co. Ltd. His research interests mainly include knowledge graph and semantic technologies. He has published more than 75 high-quality international conference and journal papers. He has been serving as PC members of ISWC, WWW, AAAI and JIST for years. He won the 2nd place in Billion Triple Challenge, ISWC 2008, the 1st place of instance matching in OAEI 2012, and many other champions in knowledge graph related data contest. He has also published the first Chinese linked open data called Zhishi.me, which accelerates the development of Chinese knowledge graph.