Research On The Detection Method Of Company Financial Events In Weibo

With the continuous development of the Internet and social networks self-media in recent years,social platforms have transformed into important positions for information dissemination and acquisition.Weibo has become a representative social platform in China due to its fast response,concise language,and official certification.Financial markets are sensitive to financial events and change rapidly.Due to the low cost of information acquisition on social platforms such as Weibo,investors can use it to reduce the investment impact of information asymmetry.Enterprises can obtain financial information in real time,identify competitors' strategies,and assist companies in making decisions.Therefore,tracking company information and detecting financial events will not only assist investors in investment decision-making and investment behavior formulation,but also provide competitive intelligence and business insights for company strategy formulation,which is of great research significance.At present,domestic and foreign scholars mainly detect financial events from online news or company annual reports,which data characteristics are completely different from social media sources.It is not possible to directly transplant previous research methods to social platforms for event detection.And with the development of deep learning algorithm in recent years,text semantic acquisition and classification through deep learning algorithms have been greatly improved over traditional machine learning algorithms.In order to detect financial events more accurately,this thesis conducts research on Weibo short text representation and financial event detection methods,including the following three parts: 1)A short text representation method for the financial field is studied.In this method,Word2 vec is used to expand the financial events trigger dictionary,and the expanded financial events trigger dictionary is used to identify the trigger words in the short text,and then Word2 vec is used to express them semantically,so as to realize the weighted vectorization representation of the short text.2)A data stream classification algorithm based on ensemble SVM is proposed and applied to company financial events detection.This method is mainly based on the data stream detection framework,which is composed of multiple SVM base classifiers.It can detect concept drift through hypothesis test and replace the base classifier dynamically.This model can effectively detect new instances of concept drift and accurately detect events in dynamic environments.3)A CNN and LSTM classificationalgorithm based on attention mechanism is constructed.This algorithm is a deep learning classification method composed of the convolutional neural network,long short-term memory neural networks and the attention mechanism.Firstly,the local semantic features of the text are obtained through a convolutional neural network;secondly,the global features of the text are obtained through long short-term memory neural networks;finally,the useful information is focused on through the attention mechanism.This model not only solves the problem that the convolutional neural network obtains a single local feature,but also effectively improves the training efficiency,and can accurately detect financial events on large-scale Weibo corpus.The experimental results show that compared with the traditional text representation scheme,the method of short text representation for financial field can further obtain the semantic features of the text.The data stream classification algorithm based on ensemble SVM is suitable for dynamic detection environment,and it is significantly better than commonly ensemble classification methods in the detection of financial events.On the other hand,the CNN and LSTM classification algorithm based on attention mechanism is better than commonly used deep learning methods in terms of feature acquisition and classification accuracy.Compared with the data stream classification algorithm based on ensemble SVM,it is more suitable for large-scale data sets,and the accuracy is better.