【作者】 王鹏；

【导师】 张春祥；

【作者基本信息】 哈尔滨理工大学 ， 计算机技术（专业学位）， 2023， 硕士

【摘要】 词义消歧是自然语言处理领域的一个重要课题。句子或文章中存在一词多义现象,即歧义词在不同的语境中语义不同。词义消歧是通过一定的算法或策略自动确定歧义词的真实语义的过程。词义消歧的发展可以提高计算机理解自然语言的能力,因此在机器翻译、情感分析、语音识别和文本分类等领域有广泛的应用。目前,词义消歧的主要难题是缺乏有效的算法模型提高消歧准确率。为了解决消歧准确率较低的问题,本文使用LSTM模型、Bi LSTM模型、孪生网络模型确定歧义词的真实语义,探究不同网络模型的消歧效果。为了提高消歧准确率,引入Attention机制和Ada Boost算法进行增强。本文主要进行了以下四个部分的研究:1.介绍了词义消歧的研究背景、应用场景和意义,以及国内外学者在词义消歧领域的研究现状。介绍了本文使用的MSH WSD数据集和不同的词义消歧方法,选择有监督的方法构建消歧模型进行实验。2.介绍了消歧特征的选择,为了引入更多的语言学知识,为句子中的词汇添加词性信息和语义类信息,抽取歧义词汇左右各两个单元组成消歧特征。分别使用LSTM模型和Bi LSTM模型进行词义消歧,比较模型的消歧性能。3.介绍了Attention机制和Ada Boost算法,在此基础上提出增强LSTM融合模型和增强Bi LSTM融合模型,进行实验分析融合模型较单一模型在消歧性能上的提升。4.介绍了孪生神经网络词义消歧模型,以相似性作为分类的基础。在训练阶段不断优化模型,使模型能准确的进行相似性度量;在测试阶段根据模型输出的相似性概率分布,确定歧义词的真实语义。由实验可知消歧效果具有一定提升。更多还原

【Abstract】 Word sense disambiguation(WSD)is a significant topic in the field of natural language processing.There is polysemy in sentences or articles,that is,ambiguous words have different meanings in different contexts.Word sense disambiguation refers to the process of automatically determining the true semantics of ambiguous words using certain algorithms or strategies.The development of word sense disambiguation can improve the ability of computer to understand natural language,so it is widely used in machine translation,emotion analysis,speech recognition,text classification and other fields.At present,the main problem of word sense disambiguation is the lack of an effective algorithm model to improve the accuracy of disambiguation.In order to solve the problem of low disambiguation accuracy,this paper uses LSTM model,Bi LSTM model and siamese network model to determine the true semantics of ambiguous words and explore the disambiguation effects of different network models.In order to improve the accuracy of disambiguation,Attention mechanism and Ada Boost algorithm are introduced for enhancement.This paper mainly studies the following four parts:1.This paper introduces research background,application scenarios and significance of word sense disambiguation,as well as the research status of domestic and foreign scholars in the field of word sense disambiguation.MSH WSD corpus and different word sense disambiguation methods used in this paper are introduced,and selects the supervised method to construct the disambiguation model for experiment.2.The selection of disambiguation features is introduced.In order to introduce more linguistic knowledge,part of speech information and semantic information are added to the sentence,and the disambiguation features are composed of two units on the left and right sides of the ambiguous words.LSTM model and Bi LSTM model are respectively used for word sense disambiguation to compare the disambiguation performance of the model.3.Attention mechanism and Ada Boost algorithm are introduced.On this basis,the enhanced LSTM integration model and enhanced Bi LSTM integration model are proposed.Experiments are conducted to analyze the improvement of disambiguation performance of the integration model compared with the single model.4.This paper introduces the siamese neural network word sense disambiguation model,which uses similarity as the basis of classification.In the training phase,the model is continuously optimized so that the model can accurately measure the similarity;In the test phase,the real semantics of the ambiguous words are determined according to the similarity probability distribution of the model output.It can be seen from the experiment that the disambiguation effect has been improved to a certain extent.更多还原