Approaches, Applications and Verification for Large-scale Knowledge Linking and Textual Semantic Computing
Abstract：In response to the emergent needs of natural language processing in the big data era, this project aims to explore novel approximate semantic computing theories and approaches based on knowledge linking from the following five perspectives:1) propose “entity-event relation networks”, to provide a unified description for both static and dynamic knowledge in knowledge bases and natural language texts from multiple sources, model the combination of logic expressions and representation learning, in order to support multi-source heterogeneous knowledge fusion and uncertain textual inference and computation; 2) investigate deep knowledge fusion approaches with the focus on semantic linking to build fundamental knowledge foundations for open-domain text computation; 3）Exploit knowledge bases and information redundancy in texts to generate weakly annotated data, and investigate open-domain text computation approaches through knowledge-driven representation learning, to provide cutting-edge techniques for deep text understanding; 4） investigate the approaches of link prediction on large-scale knowledge bases and open-domain textual inference based on the combination of logic reasoning and representation learning，to shed light on new directions of semantic computation;5) build a Chinese deep Question Answering system and evaluation platform，to verify the proposed knowledge linking, text semantic computation and knowledge inference approaches. The research outcome of this project will provide key technical support for critical applications such as e-learning and intelligent health care.
Keywords: natural language processing; textual semantic computing; question answering; knowledge base; knowledge linking
Contact:
ZHAO Jun
E-mail: jzhao@nlpr.ia.ac.cn