The Open Knowledge System for TAC KBP 2017 Zixuan Li, Yunqi Qiu, Fan Yang, Xiaolong Jin, Yuanzhuo Wang, Yantao Jia, Haoran Yan, Kailin Zhao, and Jialin Su CAS Key Laboratory of Network Data Science and Technology, Institute of Computing Technology, Chinese Academy of Science School of Computer and Control Engineering, University of Chinese Academy of Sciences lizixuan@software.ict.ac.cn Abstract This paper presents the Open Knowledge System (OKS) developed for the Cold Start KB task in TAC KBP 2017. In order to complete this task, we developed seven modules, namely, the English entity discovery module, the relation extraction module, the event nugget detection module, the entity linking and clustering module, the standalone BeSt module, inference module and post processing module. Particularly, the relation extraction module combines three existing methods: RNN-based relation extraction, OpenIE and Implicit Relation Extraction. 1 Introduction The goal of TAC KBP 2017 is to develop and evaluate technologies for building and populating knowledge bases (KBs) from unstructured text. It contains several tracks, and we participated in the Cold Start Track (KB variant, English) this year, which contains five components: Entity Discovery and Linking (EDL), Slot Filling (SL), Event Nugget Detection and Coreference (EN), Event Argument Extraction and Linking (EAL), and Sentiment. The Cold Start KBP track builds a knowledge base from scratch using a given document collection and a predefined schema for the entities and relations that will compose the KB. The KB schema for Cold Start 2017 consists of: • Entities: entities and entity mentions as defined in the main task of the EDL track; • SF Relations: entity attributes ("slots") as defined in the SF track; • Events: events (hoppers) and event nuggets as defined in the EN track; • Event Arguments: event arguments as defined in the EAL track; • Sentiment: Sentiment from a source entity toward a target entity as defined in the Belief and Sentiment (Best) track. For this purpose, we proposed a system consisting of seven modules to finish this task. The paper is organized as follows. Section 2 describes the architecture of the developed system. Document-processing and entity discovery are explained in Section 3, including Pre-processing, entity discovery and intra-document coreference resolution. Section 4 describes the details of event nugget extractor. Section 5 presents the three methods and the strategy for combining them together. The sentiment module is introduced in Section 6, the entity linking and clustering module are explained in Section 7 as well as the inference step are explained in Section 8 and the post- processing module in Section 9. Finally, we conclude the paper in Section 10 and present related references. 2 The System Architecture Our proposed system starts with the entity discovery module which extracts all named