Research Of Modeling And Methodologies For Web Financial Knowledge Service

With the rapid development of the internet technology and big data technologies,simultaneously, the needs of users for the financial knowledge services has become pro-gressively automated, diversified, and personalized. However, the traditional mode andmethods of financial knowledge service have been unable to meet the changing needsof users. On the other hand, heterogeneous forms of financial big data are mainly webinformation, PDF announcements, pictures, tables, real-time trading data, etc.. Thus, het-erogeneous information processing technology, data mining and other methods are urgentneed to support for the more e?cient and accurate services. In addition, there are betterapplication prospects and research value if we select the more concerned issues and activeapplications to research.For the above analysis, this thesis studied modeling and methodologies for web fi-nancial knowledge service, provided knowledge service by obtaining and retrieving ofweb heterogeneous information, using financial ontology and methods of time series anal-ysis to integrate and process information. For the three main branches stock, funds andbonds which users concern in financial sector, we focus on IPO(Initial Public Offerings),closed-end funds and enterprise bonds which have the high degree of activity, studiesthe modeling and methodology for knowledge service on the basis of constructing a webfinancial knowledge services framework.The main content is the follows.Firstly, overall design of financial knowledge service based on internet heteroge-neous information processing. Currently, structured and systematic exposition of hetero-geneous information processing methods for a particular field are rare, especially in thefinancial services sector. We have not found the presented summary of the financial infor-mation service processes, framework that from the perspective of Internet-based hetero-geneous information processing. Therefore, in order to improve the quality and e?cien-cy of financial knowledge services, we expound the financial heterogeneous informationprocessing methods and established the overall framework of financial heterogeneous in-formation processing. We implement the modularization of heterogeneous informationautomatically acquired, text and data relationship discovery and information verification,build financial knowledge services platform to provide real-time knowledge services forIPO, Closed-end funds, corporate bonds and other financial areas.Secondly, the analysis of the IPO gain based on Distribution and Unification GainModel is presented. We propose N-gram Distribution and Unification Gain problem,establish the model. we use dynamic programming and solve the issue by an optimizedoptimum path algorithm, gain calculation method and evaluation methods. Finally, themodel is applied to IPO benefit analysis, to calculate the optimal investment paths andmaximum investment yield over a period of time, to provide data for investors, researchinstitutions and regulatory agencies to transverse and longitudinal investment contrast,and to provide a reference for IPO investment in the future.Thirdly, the optimal policy model on new issue stock based on maximum entropyis proposed in this thesis. IPO investment has always been a hot spot in domestic in-vestment and financial research. Investment institutions and researchers analyzed the IPOprice and trends through a variety of methods. Combining stock subscription rules withthe feature of new stocks, this paper proposes the optimal policy model on new issuestock based on maximum entropy. This model differs from the traditional statistical soft-ware because of avoiding the dependence of the fixed algorithm and the limitation of datafeatures. Combining with LMT(Logistic Model Tree) classification, multiple linear re-gression and maximum entropy method overcomes the drawbacks of single algorithm.Recommending new stocks according to the results of prediction can avoid the uncertain-ty and large deviations which happens when we predict single stock price or yield directly.The contrast experiment in new stock investment income shows that the prediction resultof proposed method is close to the maximum income.Fourthly, closed-end fund timing series hybrid model based on neural network isdeveloped in this thesis. we propose a multi-model fusion method based on BP neuralnetwork. The input of this method is the maximum entropy model, trend fitting modeland support vector regression model. Each input model can handle different kinds offeatures, which avoid the limitation of features in single method. In the experiment ofclosed-end fund net valuation, the proposed model has taken a more accurate result bothin trend simulation and in valuation precision than the single methods.Fifthly, this thesis studied the methods of analysis for enterprise bond. In order toprovide automated, personalized and e?cient financial knowledge services to build the fi-nancial domain ontology, and to do research on the adaptation of ontology-based financialknowledge platform. We propose the Internet-based retrieval evaluation model on finan-cial field, the feature extraction model based on ontology rules adaptation, and the im-balance data classification based on optimized feature weight algorithm. These methodssolved the time-sensitive problem in financial information retrieval, improve the accuracyof automatic feature extraction, and enhance the granularity of minority features duringimbalance classification. To ensure accuracy of information, we verify the information onfinancial knowledge platform by direct and indirect automatic verification method. Thecomparative experiments of the key technologies on financial knowledge platform showthat the proposed models and methods can solve and improve the problems in financialinformation retrieval, feature automatic extraction and imbalance classification in the fi-nancial field.