依存信息在蛋白质关系抽取中作用.pdfVIP

25 2 V ol . 25, No . 2 20 11 3 J OU RNA L OF CH INESE IN FO RM AT ION PR OCESSIN G M ar., 2011 : 2011) , , , , 2 15006; , 2 15006) : 基于核函数的蛋白质关系 PP I) 抽取可以捕获结构化信息, 取得了较高的性能, 但其计 复杂度过高该 文结合词汇句法等信息, 重点探讨了依存信息对基于特征向量的蛋白质关系 PP I) 抽取的影响在多 个P PI 语料 库上的实验表明, 依存信息和基本短语块信息可以有效提高基于特征向量的 PP I 抽取性能特别要指出, 在 AIM ed 语料上的 PPI 抽取取得了54. 7 的 F 测度, 是目前基于特征向量的P PI 抽取系统的最好水平 : 蛋白质关系抽取; 支持向量机; 依存信息 : T P39 1 : A The Role of Dependency Information for ProteinProtein Interaction Extraction L IU Bing, Q IA N L on ghua, XU Hua, ZHOU Gu odong Scho ol of Comput er Science T echnolog y, Soochow U niver sity, Suzhou Jiang su 2 15006; Natura l L ang uage P rocessing L ab, Soo chow U niver sity , Suzhou J iangsu 2 15006) Abstract: K ernelbased P PI P ro teinPro tein Inter action) ex traction sy stem s can achieve better perfo rmance becau se of their capabilit y to capture str uctura l info rmatio n, but at the expense o f hig h co mputatio na l complex it y. T his paper inv est igat es the combination of div erse lex ical , syntactic and par ticular ly dependency info rmatio n for f eatur e based pr oteinprot ein int er action extr action using SV M s . Ex per iment al evaluation o n multiple P PI cor po ra rev eals that dependency info rmation as w ell as base phr ase chunking infor matio n is very eff ective fo r f eat ur ebased PP I ex tr act ion . P articularly, our method achieves a pr omising perfo rmance of 54 . 7 in Fmeasur e on t he A IM ed co rpus, sur passing o ther stateoftheart f eatur ebased o nes . Key words: PPI ex tr act ion ; SV M ; dependency inf ormat ion