Extracting and Visualizing Semantic Relationships from Chinese Biomedical Text

Copyright 2012 by Qingliang Miao, Shu Zhang, Bo Zhang, and Hao Yu 26th Pacific Asia Conference on Language,Information and Computation pages 99–107 Extracting and Visualizing Semantic Relationships from Chinese Biomedical Text Qingliang Miao, Shu Zhang, Bo Zhang, Yao Meng, Hao Yu Fujitsu R&D Center Co., Ltd. {qingliang.miao, zhangshu, zhangbo, mengyao, yu}@cn.fujitsu.com Abstract In this paper, we study how to automatically extract and visualize food (or nutrition) and disease relationships from Chinese publications of Nutritional Genomics. Different from previous approaches that mostly apply handcrafted rules or co-occurrence patterns, we propose an approach using probabilistic models and domain knowledge. In particular, we first utilize encyclopedia to construct a domain knowledge base, and then develop a sentence simplification model to simplify complicated sentences we meet. Afterwards, we treat relation extraction issue as a sequence labeling task and adopt Conditional Random Fields (CRFs) models to extract food and disease relationships. Finally, these relationships are visualized. Experimental results on real-world datasets show that the proposed approach is effective. 1 Introduction Advancements in biomedical science has led to large volume of published research articles, especially in Nutritional Genomics, an emerging interdisciplinary that studies the relationship between human genome, food and diseases (Hakenberg et al., 2010; Sharma et al., 2010; Tsuruoka et al., 2011). For example, many researches in Nutritional Genomics study the relationships between “green tea”, “soy”, “fish oil” and “tumor diseases”. Mining and drawing a full picture of these relationships can be adopted in many practical fields, such as public health services, drug discovery, etc. However, due to the considerable number of unstructured data, it is unrealistic to go through and obtain the panoramas of relationships manually. Consequently, automatically relation extraction and visualization techniques become ever more important and necessary. Some prior work has studied how to extract food and disease relationships from English biomedical text (Yang et al., 2011). On Chinese biomedical text, however, there is relatively little investigation conducted on food and disease relation mining. In this paper, we focus on extracting and visualizing food and disease relationship from Chinese biomedical text. S1 "金 雀 异 黄 素 能 够 影 响 恶 性 黑 色 素 瘤 的 体 外 生 长 ， 并 抑 制 紫 外 线 诱 导 的 D N A氧 化 损 伤 。 " "G enistein c ould affec t the grow th of m alignant m elanom a in vitro and inhibit ultraviolet light -induc ed oxidative DN A dam age." S2 "研 究 表 明 绿 茶 能 够 预 防 人 肝 癌 细 胞 HepG 2。 " "It suggests that green tea c ould prevent Hum an hepatom a c ell HepG 2." Figure 1: Example of relation-bearing sentences in Chinese and their English translation. Figure 1 shows two examples of Chinese biomedical sentences and their English translation. The objective of semantic relationship mining is to extract all the binary semantic relationships between food and diseases, such as <金雀异黄素, 影响, 黑色素瘤> (<genistein, affect malignant melanoma>), <绿茶, 预防, 人肝癌细胞 HepG2> (<green tea, prevent, …

Content truncated to 500 words