关键词: FrameNet, 框架消歧, 支持向量机, 条件随机场, 双层模型

Abstract:

Frame disambiguation refers to the frame that automatically identifies the ambiguous target word according to the context of the target word in a given sentence. Aiming at the problem that the traditional FrameNet frame disambiguation method does not consider the relationship between target words when it adopts the single classification model, which makes it difficult to extract the hidden features, and the classification results are relatively dependent on the performance of the classification model and the setting of parameters, a frame disambiguation method based on the two-stage model is proposed. This method uses the idea of divide and conquer to transform the frame disambiguation problem into the classification and sequence annotation of target words. The SVM model as the first stage roughly classifies the input corpus and obtains the classification labelsequence. The CRF model as the second stage takes the context sequence and classification label sequence of the SVM model as the input and add classification label to the feature template for further sequence annotation. Eighteen lexical elements and 2 614 sentences in the FrameNet are selected as the corpus of the experiment. Experimental results show that, compared with traditional single model, the two-stage model based on SVM and CRF has a higher accuracy in frame disambiguation, which proves that this method is a more suitable FrameNet frame disambiguation method.

Key words: FrameNet, frame disambiguation, support vector machine, conditional random field, two-stage model