建筑质量投诉文本分类与知识问答系统研究

汪旭

摘  要

建筑工程质量关系人民生命财产安全，一直受到建筑领域研究人员密切关注。随着我国建筑业持续发展和信息化时代的到来，中国政府积极推进电子政务的发展，顺应我国建筑业发展的时代需求。建筑质量投诉是建筑质量高低的真实表现，也是中国建筑质量缺陷的集中表现，应当受到研究人员的重视。在建筑质量投诉数量和建筑业质量领域知识不断增长的背景下，人工分类建筑质量投诉的方式无法满足当前需求，投诉处理人员和相关从业者也没有统一化电子平台查询和学习建筑质量知识。

本文使用机器学习方法卷积神经网络用于解决建筑质量投诉文本自动分类的问题，有助于加快建筑质量投诉处理过程。基于武汉市质量监督站建筑质量投诉文本，使用机器学习方法朴素贝叶斯和支持向量机用于对比试验，得出基于卷积神经网络的分类模型计算时间短和分类准确率最高的结论。为提高建筑质量投诉管理及处理效率，本文构建问答系统及其功能作为建筑质量知识的信息化管理平台，设计问答原型系统中不同组成部分的实现流程，并以实例对问答系统的实现过程进行了说明。

本文对建筑质量投诉文本自动分类的研究，有助于拓展建筑质量领域中文文本分类方法，对建筑质量知识问答原型系统的构建，有助于建筑质量知识共享。在今后的研究中，可以将文本分类模型应用于更多的短文本领域，并对建筑质量知识问答原型系统进行补充和开发，应用于实际建设工程环境中。

关键词：建筑工程质量；文本分类；卷积神经网络；问答系统；知识图谱

Abstract

The quality of construction projects relates to people's lives and property safety, and has been considered as the key topic by researchers in the field of AEC. With the advent of the information age and the sustainable development of China's construction industry, Chinese government has actively promoted the development of e-government to satisfy the need of China's construction industry. Complaints about construction quality are the reflections of the quality of construction, which refer to architectural quality defects of China. Currently, the number of complaints in construction quality and related domain knowledge are growing. However, the classification of construction quality complaints under manual method can’t meet above situation. What’s more, complaints handling personnel and related participants have no unified e-learning platform to query and learn construction quality knowledge.

This paper uses Convolutional Neural Network (CNN), one of the machine learning method, to solve the problem of automatic classification of construction quality complaint texts, which helps speed up the construction quality complaint handling process. Based on the construction quality complaints texts from Wuhan Quality Supervision Station, the machine learning methods, Naive Bayes and Support Vector Machines, are used for comparison experiments. The results show that the classification method based on CNN model has the highest classification accuracy and shorter calculation time. Besides, this paper has proposed the structure of a Question Answering System (QAS) with related functions as the management platform for quality complaints. The realization flow of different components of the QAS prototype system are designed, and examples are given to illustrate the implementation of the QAS.

The study on the automatic classification of construction quality complaint texts will help to improve the text classification method in the field of construction quality. The establishment of the QAS prototype for building quality contributes to knowledge sharing. In the future, the proposed classification model can be applied to more research areas, especially for short texts. Besides, the construction quality QAS prototype can be developed and applied to the practical construction projects.

Keywords: Quality of construction project; Text categorization; Convolutional neural networks; Question answering system; Mapping knowledge domain