Font Size: a A A

Quantifying The Effect Of Social Media On Stock Market From The Perspective Of Big Data

Posted on:2019-12-18Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z L XieFull Text:PDF
GTID:1488306125469094Subject:Information technology and economic management
Abstract/Summary:PDF Full Text Request
By the end of 2017,the number of list companies in China's stock market has reached 3485 with a total market value of RMB 56.7086077 trillion Yuan,which enables the stock market to become the instrumental element of China's national economy.Stock market is also rated by many as the barometer of national economy;therefore,its stable functioning not only lays the solid foundation for the healthy development of the economy,but also plays a significant role in maintaining social stability and safeguarding financial risks.The 19th National Congress of the CPC and the 2018 sessions of NPC and CPPCC point out that the future risk of China's economy is primarily in financial field and financial risks must be prevented.The securities supervision department has given top priority to market monitoring.The reason that bear and bull markets repeatedly appear in China's stock market in the past 28 years is because that such immature behaviours as over-optimism,excessive horror,buying the winners lead to drastic fluctuation in stock market.Market sentiment is the most significant representation of this kind of immature behaviours for most of the investors in China's stock market are individual investors.Their sentiment will largely influence the market and their emotional trading is more popular compared with institutional investors.According to modern behavioral finance,share price is determined not only by its inner value but also by the sentiment and behaviours of the information-influenced investors.Based on the analysis the current documents,most of the researchers measure the investors through indirect sentiment indicator and direct sentiment indicator to focus on the impact on the fluctuation of the stock market by the investors'sentiment.Based on the construction of historical data,indirect sentiment indicator,by indirect reflection of the investors through objective measurement,usually lags behind the times,whereas the direct sentiment indicator often takes the sentiment of investors when surveyed directly as the samples.As a result,this kind of measurement means big error and high cost for the samples do not necessarily mean the sentiment when making decisions.With the advent of the internet,people find it easier and easier to get the access to information and the internet has become the major source for seeking the information.Investors can easily acquire,release and spread the relevant information of stock market through social network,some of which involve the sentiment of the investors.The explosive information has attracted the researchers to study the investors'sentiment and its impact on the fluctuation of stock market through natural language processing and traditional machine learning model.However,due to the limitation of information technology and cross-disciplinary fields,the current research of information on social network,investors'sentiment and even the research method are far from satisfaction.This thesis uses distributed crawler to obtain complete text information from social network and summarizes text sentiment through Chinese Sentence Convolutional Neural Network(CSCNN)core algorithm and the analysis of the grammar and semantic structure of the Chinese language.This thesis also integrates the structural features of social network text information to build Social Media Investor Sentiment Index(SMISI)which can delegate the investor sentiment of social media.Based on the investor sentiment of Sentiment-driven Long Short-Term Memory(S-LSTM)depth neural network algorithm,this thesis constructs social media quantitative intelligent platform(SMQIP)of stock market,which can explore the depth and breadth of the influence of social media investor sentiment on the stock market.This these consists of eight chapters:Chapter one is the introduction.It mainly introduces the background,significance,method,structure and major innovative points of the research.Chapter two systematically reviews the existing research from social media quantification,market sentiment and social media's impact on the fluctuation of stock market.This chapter provides the theoretical basis for the analysis of social media quantification,social media information,stock market fluctuation and the ways to prevent systemic financial risksChapter three summarizes the logical module and process from system overall design,illustrating the data processing of the system.This chapter also tries to make sure the relationship between the modules of social media intelligence solutions in the stock market and the whole process from data capture,sentiment extraction to in-depth study of neural network so as to push the system module and research smoothly.Chapter four studies the social media quantification and investors'sentiment extraction.This chapter first describes the extraction,filtration,pre-processing and vocabulary quantification,then decides the emotional polarity of the text from grammar,semantic structure and CSCNN core algorithm,finally compares the affective decision models.Chapter five uses the traditional exponential construction principle and the feature of information text structure of social network to put forward a social media statement weight algorithm,which is based on content similarity matrix,reference relation matrix and reply relation matrix,to calculate statement weight.This chapter also combines user impact factor,reading quantity factor and thumb-up quantity factor to create investment emotion index(SMISI),which provides characteristic variable for further study.Chapter six uses empirical study method.It combines SMISI and Fama to verify the systematic impact of SMISI on the rate of return in stock market.It further uses VAR model to study how deeply SMISI will impact the fluctuation in stock market.Chapter six also puts forward the core algorithm,which is based on social media emotion-driven S-LSTM to accurately capture the impact of social media investors'emotion on the stock market and verify the feasibility of SMISI applied in quantitative investment through simulation.Chapter seven analyzes the inner principle how social media triggers market emotion,leads to price fluctuation and affects financial stability from three angles, regulators,list companies and investors.This chapter also verifies the availability of SMQIP from these three angles.Chapter eight concludes the whole thesis,summarizes the existing loopholes of this thesis and expects possible future research plan from the perspective of financial intelligence.Based on the existing research,this study quantifies the media effects of big market-based securities market along the main line of“social media–investor sentiment–securities market volatility”.The innovation points of this thesis may include:The first is about the study of the collection and transaction of public sentiments on the basis of the content and release structure of the information,based on the deep learning method.This thesis applies convolutional neural network of the Chinese language to judge the sentiment under the environment of stock market,and make full use of the unique structure of web forum:posting,replying,forwarding and quoting,innovatively puts forward a set of extraction method of public sentiments.By doing so,this thesis may successfully summarize dominant speech and extract the sentiment tendency toward the list companies,board or even the whole market from the massive,chaotic information in the web forum.The second innovation point is about the research and design of sentiment exponent of new type social media in stock market.This thesis innovatively applies exponential construction principle in statistics to give weight to positive sentiment and negative sentiment so as to construct investment emotion index(SMISI)and influence mechanism and transmission mechanism in the historical transaction data in stock market.In the meantime,this thesis also subdivides sentiment exponents into main board list companies'social media sentiment,medium and small-sized list companies'social media sentiment and GEM's social media sentiment by combining the sentiment exponent of constituent stock.All these may provide important referential value for market regulators.The third innovation point is about the research and design of intelligence analysis model of stock market and key algorithm on the basis of in-depth study neural network.This thesis innovatively puts forward a solution for the integration of continuous time series data and discrete time series data by transforming LSTM,adding mood enhancement door,changing the data structure of forgetting gate,input gate and output gate,and constructing sequential neural network driven by social media sentiment.The SMQIP on the basis of the previous study may provide theoretical reference and decision basis for market regulators,list companies,investors and relevant researchers.The sequential neural network driven by social media sentiment also develops a new way out for issues of continuous time series data and discrete time series data in other fields.
Keywords/Search Tags:Social Media, Investor Sentiment, Deep Neural Network, Securities Market, Quantitative Investment
PDF Full Text Request
Related items