学术文本词汇功能识别--基于标题生成策略和注意力机制的问题方法抽取

学术文本词汇功能识别的目的是实现学术文本中表征问题、方法和对象等词汇的抽取。针对传统识别方法中训练难以获取所导致的识别准确率低、召回率有限和泛化能力差等问题,本研究提出了一种基于深度学习和标题生成策略的学术文本词汇功能识别方法,将任务形式由信息抽取转化为特定形式的标题生成问题。本研究采用构建seq2seq模型和引入注意力机制的方式捕获词汇多层语义信息,最终实现学术文本中问题和方法指代词的生成和获取。实验结果表明,通过应用深度学习方法和标题生成策略,本研究提出的模型能够从摘要中有效识别学术文献的主要研究问题和主要研究方法,并较已有方法在识别效果上有明显提升。

Recognition of Lexical Functions in Academic Texts:Problem Method Extraction Based on Title Generation Strategy and Attention Mechanism

The purpose of academic text problem and method identification is to extract research questions and methods from academic text.Aimed at solving the problems of low recognition accuracy,limited recall rate,and poor generalization ability caused by the difficulty of obtaining the training set in traditional recognition methods,this study proposes an academic text problem recognition method based on a deep learning and title generation strategy.The method converts the extraction and recognition of the problem method into the form of title generation in a specific form.By constructing a seq2seq model and introducing an attention mechanism,multi-layer semantic word information was captured to generate and obtain the problem and method pronouns in academic texts.The experimental results showed that through the application of deep learning methods and title generation strategies,this study effectively identified core research problems and core research methods in academic literature.