| Summary: | 碩士 === 國立高雄應用科技大學 === 資訊管理研究所碩士班 === 104 === Most of the acadamic researchs about prediction of stock price in the past use fundamental and technical analysis rather than use keywords from news of the coporation, or further use subjects from a group of keywords to predict stocks price in the furture.
This paper investigates news of corporation in the cnYES, news categories include foods, semiconductors, and computer peripherals equipment, and those dated from September, 2014 to February, 2015 as the training data. We obtain keywords and topics through techniques such as text mining, tfidf, Latent Dirichlet Allocation(LDA) and so on. Using support vector machine(SVM) to train and establish models for prediction, we obtain a predictive value every other day after transferring the format of the keywords and topics. We also use the feature selection for topic models and expect to find the best topics to improve the accuracy and performance of model of prediction.
The experimental results show that the stock of food category perform the best after using feature selection. The accuracy of other category also improves after feature selection. Compared with the models only using keywords, there is more accuracy and improving 5% correctness on topic models.
Keywords: LDA, Text Mining, SVM, Feature Selection, Stocks Price Prediction
|
|---|
