【作者】 吴茜；

【导师】 张德富；

【作者基本信息】 厦门大学 ， 计算机技术， 2019， 硕士

【摘要】 随着基于开放领域知识图谱的问答系统的发展,面向专业领域的问答系统因其能够利用大量的专业领域相关的知识来提高问答系统的实用性而受到许多领域专家的青睐。本文在开放领域问答系统实现的相关技术基础上,对基于知识图谱的农业问答系统进行研究,论文的主要工作包括:(1)农业知识图谱构建。目前最大的中文开放知识图谱上还没有农业相关的知识图谱,因此,本文利用知识图谱构建技术和基于本体的知识库构建技术,使用Scrapy和Protege工具,从互联网中收集农业数据,构建了涵盖农作物品种、农作物病虫害和农药肥料数据的农业领域知识图谱。(2)设计并实现了基于知识图谱的问答算法。将问答算法划分为农业领域命名实体识别和属性链接两个子任务。对命名实体识别这一任务,提出了多特征的CRF农业领域命名实体识别算法用以获取问句实体。对一个特征、多个特征组合及引入上下文信息的CRF模型训练结果表明,选择字本身与偏旁特征组合并引入一个窗口上下文信息的CRF实体识别算法对农业领域的命名实体识别有较好的效果,F1值为91.57%,其中对农作物实体的识别准确率达97.05%;对属性链接这一任务,将双向的LSTM模型与传统单向的LSTM模型进行对比,结果表明双向的LSTM模型能达到更好的效果。(3)设计并实现了交互式农业知识问答系统。能够支持农作物基本信息查询,对用户提出的农业领域问题给出精准的答案。更多还原

【Abstract】 With the development of question answering system based on open domain knowledge graph,the Q&A system for professional domain is favored by many experts because it can improve the practicability of Q&A system by using a large amount of knowledge related to professional domain.Based on the relevant technologies of the open field Q&A system,this paper studies the agricultural Q&A system based on knowledge graph.The main work of this paper includes:First,construction of agricultural knowledge graph.Agricultural knowledge graph is not found in the biggest Chinese knowledge graph.Therefore,in this paper uses the construction technology of knowledge graph and ontology knowledge base,Scrapy and Protege tools,and collecting the agricultural data from the Internet to build a knowledge graph in agricultural field which covers crop varieties,crop pests and diseases,fertilizer and pesticide data.Secondly,the Q&A algorithm based on knowledge graph is designed and implemented.The Q&A algorithm is divided into two sub-tasks,named entity recognition and attribute linking.For the task of entity recognition,a multi-feature CRF in agricultural field is proposed to obtain the question entity.The CRF model training results for a feature,multiple feature combinations and the introduction of contextual information show that the CRF model which combines the selection word itself with the radical feature,and introduces a window context information has a good effect on the named entity recognition in the agricultural field.The value of F1 is 91.57%,and the accuracy rate is 97.05%for crop entities recognition.For the task of attribute linking,the bidirectional LSTM model is compared with the traditional one-way LSTM model,and the results show that the Bi-LSTM model can achieve better effect.Thirdly,the interactive Q&A system of agricultural knowledge is designed and implemented.It can support basic information query of crops and provide accurate answers to the question raised by users in the field of agriculture.更多还原