Recent years have witnessed a growing interest in knowledge base construction (KBC) from both academia and industry. Knowledge base construction (KBC) refers to the process of populating a knowledge base (KB) with facts (or assertions) extracted from information sources, including documents, books, sensors, human, etc. While considerable efforts have been devoted, the state-of-the-art automatic KBC techniques, which rely on the information extraction (IE), natural language processing (NLP) and machine learning approaches, still have its limitations and can yield noisy or semantically meaningless knowledge facts. Human computation and crowdsourcing are becoming ever more popular paradigms in computing which employ the power of human knowledge and expertise to handle tasks that a...[ Read more ]

Recent years have witnessed a growing interest in knowledge base construction (KBC) from both academia and industry. Knowledge base construction (KBC) refers to the process of populating a knowledge base (KB) with facts (or assertions) extracted from information sources, including documents, books, sensors, human, etc. While considerable efforts have been devoted, the state-of-the-art automatic KBC techniques, which rely on the information extraction (IE), natural language processing (NLP) and machine learning approaches, still have its limitations and can yield noisy or semantically meaningless knowledge facts. Human computation and crowdsourcing are becoming ever more popular paradigms in computing which employ the power of human knowledge and expertise to handle tasks that are difficult for machines to handle alone. Crowdsourcing offers an alternative approach for KBC in which the crowd power can be incorporated to refine the knowledge extraction and acquisition process; moreover, as a natural source of knowledge, the crowd can be mined to obtain knowledge that resides in the human mind. However, the crowd alone cannot carry the whole burden of KBC due to the conflict between limited crowdsourcing resource and the large scales of real KBs.

In this thesis, to address the shortcomings of both automatic and human computation approaches, we propose hybrid human-machine computation frameworks for KBC to complement automatic knowledge base construction with the power of the crowd. To summarize, our study address the following problems:

● We propose a hybrid framework to combine the crowd and machine intelligence for taxonomy construction towards both high accuracy and high coverage.

● We study the problem of KBC by integrating existing large scale KBs in the new crowdsourcing perspective.

● We identify a subjective KBC problem which targets at subjective knowledge acquisition. We present two hybrid frameworks for subjective KBC powered by crowdsourcing and existing KBs.

We verify the effectiveness of the proposed frameworks with extensive experiments on real data sets and crowdsourcing platforms. In the end, we discuss future research direction of KBC with hybrid human-machine computation.