Chemical Compounds Knowledge Visualization with Natural Language Processing and Linked Data

Chemical Compounds Knowledge Visualization with Natural Language Processing and Linked Data Kazunari Tanaka1,2 Tomoya Iwakura1,2 Yusuke Koyanagi1,2 Noriko Ikeda1 Hiroyuki Shindo2,3 Yuji Matsumoto2,3 FUJITSU LABORATORIES LTD.1, RIKEN Center for Advanced Intelligence Project2, NARA INSTITUTE of SCIENCE and TECHNOLOGY3 {tanaka.kazunari, iwakura.tomoya, koyanagi.yusuke, nona}@jp.fujitsu.com, {shindo, matsu}@is.naist.jp Abstract This paper proposes a visualization system for chemical compounds. New chemical compounds are being produced by every moment and registration of chemical compounds to databases strongly depends on human labor. Our system uses Natural Language Processing technologies for extracting information of chemical compounds from text and for storing the extracted results as Linked Data (LD). By combining the extracted results with LD-based existing chemical compound knowledge, our system provides visualization of chemical compound information such as integrated view of several databases and chemical compounds that have similar structures. Keywords: Information extraction, chemical compounds 1. INTRODUCTION Knowledge of chemical compounds has great value for developing new materials, new drugs, and so on. Therefore, databases of chemical compounds are being created. For example, CAS [1], one of the largest databases, includes information on over 100 million chemical compounds. However, the creation of such databases strongly depends on manual labor since chemical compounds are being produced at every moment. In addition, the database creation mainly focuses on English text. Therefore, in other words, chemical compound information other than English is not good enough to be available. For example, although Japan has one of the largest chemical industries and has large chemical compound information written in Japanese text documents, such information is not exploited well so far. We propose a visualization system based on chemical compound extraction results with Japanese Natural Language Processing and structured databases represented as Linked Data (LD). Figure 1 shows an overview of our system. First, chemical compound names in text are recognized. Then, aliases of chemical compound names are identified. The extraction results and existing chemical compound databases are represented as LD. By combining these LD-based chemical compound knowledge, our system provides different views of chemical compounds. 2. Chemical Compound Information Extraction Unknown chemical compounds usually first appear in unstructured text such as scientific papers and patents. In order to extract chemical compound names from text, we use a Japanese Named Entity (NE) recognition method [2]. An NE recognizer is trained with a distant supervision [3] using a dictionary of chemical compound names as a lexical resource. The dictionary is compiled from Nikkaji [4]. The dictionary is used to recognize chemical compound names in text. The chemical compound names in text recognized with the dictionary are used as positive samples for training. Negative samples are documents that would be irreverent with chemical domains such as politics and sports. Then, the automatically created training data was used for training an NE recognizer. By using a chemical compound name recognizer based on an NE recognition method, we expect to obtain new chemical compound names from text with contextual information like “boiling point” and “fusing points” or, prominent words consisting of chemical compounds like “acid” and “methylphenol”. …

Content truncated to 500 words