摘要
随着信息时代的快速发展，各行各业在网络上都会集聚大量的信息，这些信
息通过挖掘分析能够补充现阶段的片面研究，就目前来看，互联网的广泛使用使
得网络平台上存在着大量有关金融市场、经济发展、企业风险等相关的信息。从
文本信息源来看，股吧作为投资者全方位交流分享平台，其内中包含着大量有关
企业盈亏和经营状况的信息，这些信息从侧面反映出企业的信用风险，进一步影
响银行等金融机构展开客观评估。基于此，为了能直观评价股吧信息是否有助于
信用风险评估，本文在传统纯财务评估指标体系基础上将股吧信息情感值作为新
兴指标引入进来，分析股吧信息在信用风险评估中的作用。
本文的主要内容包括以下几个方面：第一，通过梳理前人研究成果以及分析
股吧信息的特点，从股吧传递有关企业价值的重要信息、股吧信息反映企业违约
倾向、股吧信息反映企业信用状况恶化、股吧信息对投资者心理和行为的影响四
个方面阐述股吧信息对信用风险评估的作用机理；第二，根据实证研究内容和条
件需求共选取 75家中小板上市公司作为研究对象，结合基础情感词典、领域情
感词典、网络情感词典和修饰词词典，构建股吧信息情感词典，并从词语搭配和
句型两个方面设计基于情感词典的情感分析算法，运用 Python3.6.2软件完成股
吧信息爬取和情感分析计算；第三，通过随机森林模型，运用 R语言软件分别
对纯财务指标模型和引入股吧信息情感值的混合财务指标模型进行中小板上市
公司信用风险评估实证研究。
实证结果表明：引入股吧信息情感值的混合财务指标体系模型对中小板上市
公司信用风险评估准确率比纯财务指标体系模型平均提高 4.44%，对“ST”上市
公司的信用风险识别效果较好；混合财务指标体系模型和纯财务指标体系模型的
信用风险评估准确率都随着时间的往前推移而逐年下降，且对“ST”上市公司
的信用风险评估准确率呈逐年下降的趋势，而对“非 ST”上市公司的信用风险
评估准确率几乎没有变化。
关键词：股吧信息；情感分析；随机森林；信用风险评估
I

ABSTRACT
With the rapid development of the information age, all walks of life will gather a
lot of information on the network, which can supplement the one-sided research at
this stage through mining and analysis. At present, the extensive use of the Internet
makes a lot of information about financial market, economic development, enterprise
risk and other related information on the network platform. From the text information
source, as an all-round exchange and sharing platform for investors, the Guba
contains a large number of information about the profits and losses and operating
conditions of enterprises, which reflects the credit risk of enterprises from the side,
and further affects the objective evaluation of banks and other financial institutions.
Based on this, in order to intuitively evaluate whether Guba messages is helpful to
credit risk assessment, this paper introduces the emotional value of Guba messages as
a new indicator based on the traditional pure financial evaluation index system, and
analyzes the role of Guba messages in credit risk assessment.
The main content of this paper includes the following aspects: Firstly, by
combing the previous research results and analyzing the characteristics of stock bar
information, this paper expounds the mechanism of Guba messages on credit risk
assessment from four aspects: the important information about corporate value, the
tendency of corporate default reflected by Guba messages, the deterioration of
corporate credit status reflected by Guba messages, and the impact of Guba messages
on investors' psychology and behavior. Secondly, according to the empirical research
content and condition demand, 75 small and medium-sized board listed companies are
selected as the research objects. Combined with the basic emotion dictionary, domain
emotion dictionary, network emotion dictionary and modifier dictionary, the Guba
messages emotion dictionary is constructed. The algorithm of sentiment analysis
based on emotion dictionary is designed from two aspects of word collocation and
sentence pattern. The information crawling and sentiment analysis calculation of
Guba are completed by using Python 3.6.2 software. Thirdly, through the random
forest model, we use R language software to conduct an empirical study on the credit
risk assessment of Listed Companies in small and medium-sized board by using R
language software.
The empirical results show that: the mixed financial index system model with
stock bar information emotional value has an average increase of 4.44% in the credit
II

risk assessment of small and medium-sized board listed companies than the pure
financial index system model, and has a good effect on the credit risk identification of
"ST" listed companies. The accuracy rate of credit risk assessment of mixed financial
index system model and pure financial index system model decreased year by year
with the passage of time, and the accuracy rate of credit risk assessment of "ST" listed
companies showed a downward trend year by year, while the accuracy rate of credit
risk assessment of "non ST" listed companies almost did not change.
Key words: Guba Messages; Sentiment Analysis; Random Forest; Credit Risk
Evaluation
III
。。。以下略