CFSD中文金融情感词典_20190913_171605.zip_中文金融情感词典资源-CSDN文库

共12个文件

csv：4个

png：2个

txt：2个

NLP

需积分: 44 123 浏览量 2020-04-20 10:22:58 上传评论 11 收藏 3.64MB ZIP 举报

资源推荐

资源详情

资源评论

收起资源包目录

CFSD中文金融情感词典_20190913_171605.zip （12个子文件）

CFSD中文金融情感词典

.ipynb_checkpoints

中文会计&金融情感词典-checkpoint.ipynb 72B

img

正面s.gif 1.23MB

正面.png 215KB

负面.png 204KB

负面词典.txt 34KB

CFSD中文金融情感词典

负面词典.csv 10KB

正面词典.csv 8KB

A New Chinese Financial Sentiment Dictionary for Textual Analysis in Accounting and Finance 个副本.pdf 2.56MB

正面词典.txt 24KB

负面词典.csv 10KB

正面词典.csv 8KB

中文会计&金融情感词典.ipynb 10KB

A New Chinese Financial Sentiment Dictionary for Textual

Analysis in Accounting and Finance

Shibo Bian

, Dekui Jia

, Feng Li

, and Zhipeng Yan

First version: February 2018

This version: July 2019

Abstract: Using a multi-stage filtering procedure based on both algorithms and human

judgement, we develop a Chinese Financial Sentiment Dictionary (CFSD) with the

hope to help advance textual analysis of Chinese documents in accounting and finance.

There are 1,489 negative words and 1,108 positive words in the CFSD. We translate all

2,597 words into English. We also briefly discuss two unique settings in which

researchers can investigate some important and interesting research questions that are

rarely studied due to lack of data.

Key words: textual analysis; sentiment analysis; Chinese financial sentiment dictionary

This project is a part of the research, “Do small investors matter?” (Bian, Jia, Li and

Yan, 2019)

School of Statistics and Management, Shanghai University of Finance and Economics, Shanghai 200433, China,

Tel: 8613524558839, Email: bian.shibo@mail.sufe.edu.cn.

College of Business, Changzhou University, 1 Gehu Road, Wujin District, Changzhou Jiangsu, China 213164,

Tel: 8613817798472, Email: jdk@cczu.edu.cn. Jia acknowledges financial support from the Chinese National

Social Science Fund (Grant No. 17BJL036).

Shanghai Advanced Institute of Finance, Shanghai Jiao Tong University, Shanghai China 200030, Tel: 86-21-

6293-4533, Email: fli@saif.sjtu.edu.cn.

Martin Tuchman School of Management, New Jersey Institute of Technology, University Heights, Newark, NJ

07102-1982. Tel: (973) 596-3260, Email: zyan@njit.edu.

Electronic copy available at: https://ssrn.com/abstract=3446388

I. Introduction

Textual analysis is a surging field in accounting and finance. Most studies in this

area rely on textual information in the English language. Few studies examine large-

sample textual data in Chinese, the most spoken language in the world. The linguistic

characteristics of Chinese are drastically different from those of English. The thinking

patterns between Chinese and English native speakers also differ significantly

Sentiment analysis is an important aspect of textual analysis. It attempts to measure

the sentiment or tone of a financial document, by counting the number of words

associated with a particular sentiment word list scaled by the total number of words in

the document.

For researchers, the first step in the process of sentiment analysis is to decide which

dictionary should be used. Four different English word lists/dictionaries have been

extensively used by researchers: Henry (2008), Harvard’s GI, Diction, and Loughran

and McDonald (2011, hereinafter LM)

. It seems that the LM lists have become

predominant in more recent studies (Kearney and Liu, 2014). Due to the unique

linguistic characteristics of Chinese, we cannot translate the LM word list directly into

Chinese and use it for analyzing Chinese documents.

There are three commonly used Chinese sentiment classification dictionaries:

HOWNET, DLUTSD, and NTUSD. Similar to Harvard’s GI and Diction, all three

dictionaries are all for general use rather than for business usage. Loughran and

McDonald (2011) warn that researchers attempting to measure sentiment in business

communications should not use classification schemes derived outside the domain of

business usage. Instead, word lists designed specifically for business communication

should be used to measure the sentiment of business text (Loughran & McDonald 2016).

To the best of our knowledge, there is not a sentiment dictionary in Chinese

available for business research. To fill this gap, we develop a Chinese Financial

Sentiment Dictionary (CFSD) in this paper with the hope to advance textual analysis of

Prior research provides evidence that Chinese native speakers rely primarily on visual representations whereas

English native speakers rely primarily on phonological representations (Schmitt et al. 1994).

See Loughran and McDonald (2016) for detailed discussion.

Electronic copy available at: https://ssrn.com/abstract=3446388

Chinese documents in accounting and finance.

II. Construction of CFSD

We first create a basic dictionary using HOWNET, DLUTSD and NTUSD and then

use online roadshow transcripts, earnings conference call transcripts, IPO prospectus

reports, and annual reports to create a basic corpus. Using a multi-stage filtering

procedure based on both algorithms and human judgement, we finally create CFSD.

The complete process is described in detail as follows.

Step 1: Create a basic sentiment dictionary by merging the positive and negative

terms in HOWNET, DLUTSD and NTUSD, and removing the duplicates.

Step 2: Collect 1,411 online roadshow transcripts, 7,138 earnings conference call

transcripts, 2,043 IPO prospectus reports, and 29,737 annual reports. Jieba (a Python

package) is used to segment these documents to create a basic corpus.

Step 3: Calculate the word frequency of every positive and negative terms in the

basic corpus according to the basic sentiment dictionary created in Step 1. The terms

with zero frequency are excluded. Terms that are irrelevant to the financial documents

are manually excluded before the basic Chinese financial sentiment dictionary (CFSD

0.0) is created.

Step 4: All the terms in CFSD 0.0 are from HOWNET, DLUTSD and NTUSD.

Since these three dictionaries do not include some frequently used negative and positive

terms in finance, 100 frequently used financial negative words and 100 positive words

are manually added to produce the dictionary CFSD 0.1.

Step 5: Gensim (a Python package) is used to transform the basic corpus into a

word vector model. A word vector model is used to find out the first 50 similarities (or

synonyms) for each term in the CFSD 0.1. CFSD 0.2 is created after we manually

remove all the similar terms (or synonyms) that are irrelevant to the financial documents.

Step 6: the CFSD 0.0, CFSD 0.1, and CFSD 0.2 are merged to remove the

duplicates to get the CFSD.

There are 1,488 negative words and 1,107 positive words in the CFSD. Table 1

Electronic copy available at: https://ssrn.com/abstract=3446388

reports the constituion of HOWNET, DLUTSD, NTSUSD, and CFSD. It shows that

CFSD contains much fewer positive and negative words than the other three general

dictionaries. We translate all 2,595 words into English. Due to the vast differences

between these two languages, some words cannot be meaningfully translated into

English unless they are interpreted within a concrete context. In addition, some

linguistic nuances of Chinese words can only be understood by a native speaker and

cannot be precisely translated. Nonetheless, we list all words alphabetically according

to their English translations (Part I and Part II, Appendix).

III. Applications of CFSD

For textual analysis in accounting and finance, researchers can apply the CFSD in

examining news articles, the China Securities Regulatory Commission (CSRC,

equivalent to the SEC in the U.S.) filings, stock exchange filings and comment letters,

text from social media, such as weibo, China’s Twitter, and so on.

Moreover, China’s security markets and financial services platforms offer quite a

few unique settings in which researchers can investigate some important and interesting

research questions that are rarely studied due to lack of data. In this section, we briefly

discuss two such settings that can help further our understanding of the role played by

small investors: online eanrings communication conferences (ECC) and online

roadshows.

1. Online Earnings Communication Conferences (ECC)

In late 2005, the Shenzhen Stock Exchange of China mandated that each company

listed on the Small and Medium Enterprise (SME) Board must hold an online ECC,

within ten trading days of its annual report release. An ECC must have an open question

and answer (Q&A) session, during which small investors can ask management any

business-related questions and the management is obligated to answer them. In 2009,

the China’s hi-tech board - ChiNext imposed the same set of rules on its listed

companies.

All online participants can see all the questions and answers in real-time during an

ECC. Manual inspection of 50 ECC transcripts reveals that investor questions are very

Electronic copy available at: https://ssrn.com/abstract=3446388

diverse. They cover all aspects of a company’s business and more. Many questions deal

with understanding the company’s operations and accounting practice, including

questions about local government subsidies, earnings forecast, related party

transactions, major competitors, dividend payout, and so on. Some are about corporate

culture, governance and social responsibility, such as minority shareholder protection,

executive compensation and environmental protection. A handful questions are

seemingly irrelevant to a company’s operations but informative. For instance, a

participant asked

, “I am a small retail investor. Can I visit your company? I called the

number (for investor relationship) several times. No one answered, and I am very

anxious.” Quite a few questions are blunt and challenging. For example, one question

was repeatedly asked at many an ECC, - “Who is the real boss in your company?”

During the 2015 ECC of H.Brothers

, one asked its chairman, “Mr. Chairman, we as

investors do not like your diplomatic rhetoric. How can you win back investors’

confidence?” And the answer was, “Thank you for your advice.” The funniest question

goes to “Uncle Zhu, I have been crying and fainting in the bathroom. Shenzhen

Government has imposed restrictions on real estate purchases. What is your take on

this?”

This set of questions are quite different from those typically asked by professional

analysts or large investors. The online IDs of these participants also tend to contain their

cell phone numbers and personal touches, further suggesting that they are likely

individual small investors.

Utilizing this unique setting, Bian, Jia, Li and Yan (2019) study the role of small

investors in corporate governance by examining whether they ask value-relevant

questions when interacting with managers. They find that market reaction and future

firm performance are positively associated with the tone of small investors’ questions.

When managers dodge the questions from small investors, as measured using the degree

of dissimilarity between investors’ questions and mangers’ answers, the market reacts

less favorably and future performance becomes worse. The effects are stronger for firms

The question was asked during the 2007 ECC of Zhejiang NHU Co., Ltd. (stock ticker: 002001). The question

was directed to Mr. Baiji Hu, Chairman of the company.

The company is Huayi Bros. Media Group, a leading entertainment group in China. Its stock ticker is 300027.

The question was asked during the 2015 ECC of Zhaoshang Shekou (stock ticker: 001979), a real estate

company in Shenzhen. The question was directed to Mr. Wenkai Zhu, Deputy general manager of the company.

Electronic copy available at: https://ssrn.com/abstract=3446388

评论收藏

内容反馈

�JW.Megalodon

粉丝: 2
资源: 1

CFSD中文金融情感词典_20190913_171605.zip

会计金融情绪词典.zip

中文情感词典

最全中文情感词库

中文情感分析语料库

基于深度学习的领域情感词典自动构建_——以金融领域为例.pdf

【深度学习语料库】常见金融领域词汇词典

金融词汇字典.txt

最全中文情感和语义词库

中文情感分析词典数据集（基于词典）.zip

中文会计&金融情感词典.ipynb

最常用的中文情感词典.zip

基于词典方法和机基于器学习方法的中文情感倾向分析系统.zip

基于词典方法和机器学习方法的中文情感倾向分析（毕业设计：附完整代码）.zip

金融专业术语词典

专用EFI包.rar

中文实体词典(NLP必备)

中英文会计&金融情绪词典.rar

用python实现基于情感词典的情感分析

中文情感词汇库-极值表

中文情感分析常用词典-chinese_sentiment_dictionary-master.zip

完整车牌号识别程序，可以识别车牌和颜色，可以集成到项目中 支持win7+

ChatGPT教程（终极版）最全整理

博客中Kmeans以及FCM算法数据（免积分）

神经网络回归预测--气温数据集

hugging face的models-openai-clip-vit-large-patch14文件夹

最新资源

完整车牌号识别程序，可以识别车牌和颜色，可以集成到项目中支持win7+