近年來生物技術逐漸進步，大型實驗產生相當大量的資料與文件，如何在這些使用自然語言(如英文)的文件中萃取出有用的資訊，使得這些萃取出來的資料可以進一步分析變的越來越重要。 無論我們感興趣的是想從文件中了解生物體內每個環節的交互作用亦或是生物物質的註解，這項研究的第一步就是要先能讓電腦辨識出文件中，我們感興趣的物質名稱。這個研究即是在生物文件中，辨識出所有蛋白質的名稱。我們提出了一個系統來辨識出蛋白質或基因的名稱。這個系統主要依據人造的規則，外加機器學習機制讓系統表現的更好。這個系統在這個研究領域有名的文件集Yapex上，達到了F-score 73.8%的水準。 New high-throughput technologies have increased the accumulation of data about genes and proteins. However, such data is stored in natural language text. Further processing and integrating data into more complete and useful knowledge become harder for researchers because of tremendous amount of literature. Therefore, automatic literature mining is more and more important in recent years. The first step to extract knowledge from natural language text is to extract the named entities out of text, and then the relation between named entities can be constructed. Here we propose a new system to extract the named entities (especially named entities refer to proteins or genes) from the literature in biological domain such as MEDLINE abstracts. The system is mainly rule-based and combined with an SVM machine learning module for improving the system performance. It achieves an F-score 73.8% on the Yapex corpus.
