Research Of Stock Clustering For Financial Knowledge Service

For massive data of the financial industry emerging constantly,it is necessary to seek an effective way for mining and processing such information.With the rapid development of computer science and graph theory,interdisciplinary study that uses the theory of complex network in combination with other disciplines has achieved substantial results.In the financial field,it is very important to classify stocks correctly to build a portfolio and reduce risk in modern portfolio theory.Being different from traditional method in which financial markets classifying stocks by industry,region,concept and some other basic aspects,we apply complex network theory in the Chinese stock markets and use community detection algorithm to perform cluster analysis of the stock correlation networks in this paper.Based on the threshold method of edge correlation,we build stock networks according to time series data similarity and unstructured data similarity respectively.Finally,we implement an online analysis system of stock data and deploy it on Haitian Yuan finance service platform.The main research of this context contains the followings:Information collection and pretreatment.In this paper,information used in our research includes structured market data,web form and unstructured data.Different method is used to clean and deal with different data.Finally,the unstructured financial data is transformed into structured data.These accurate data provide guarantee for the construction of correlation network and community discovery.Construction of stock correlation network.The stock correlation network is constructed by price fluctuation correlation coefficient,and anther stock relation network based on text information is constructed by the stock relevance excavated from stock news and announcement data.In this paper,we take the threshold method of edge correlation coefficient to build the final network.Community discovery of stock correlation networks.In this paper,varieties of community discovery methods are used to find cluster of the stocks.Except Newman fast algorithm and Fast unfolding method,a community discovery method based on LDA topic model is proposed to analyze the stock correlation network community.Then we compare different result of different method and analyze the result with the stock market.Implementation of Haitian Yuan finance service platform and c onstruction of stock analysis system.The purpose of this paper is to apply the data-mining algorithm and machine learning method to the real financial market and build an online system to serve the user.The system display the cluster results in graphical way,which can provide a reference for the construction of portfolio and risk aversion.