Time: 2014Posted in: EMNLPOriginal file: http://pan.baidu.com/s/1i3phG49
Main content: Using news events to predict: 1. 2. Market trend of U.S. stocks; The trend of the selected 15 stocks.
Detailed content: Main work step: 1. Extract Financial News 2. The news title is parser and the event is extracted. The event extraction is open information extraction, which is an open-ended template, but an open extraction. The result of extraction is: (subject, verb, object, time). Among them, each element will do stemming, verb will do clustering, become verb category, such as: get_class this category. 3. Generalization of the extracted results, such as: (Subject, verb, object, time) is generalized into several inputs, (subject, verb), (verb, object) and so on, so as to avoid the original input data sparse problem. 4. The above input to the classification model, such as: SVM, or deep learning, binary classification, so as to predict the market and the single stock of the ups and downs
Experiment: On the U.S. stock market, we used Reuters and Bloomberg's news on the financial channel to predict the next day, the next week, and the next one months, respectively. Contrast baseline is a model that uses only the word-bag input feature (SVM or deep learning)
Experimental results: 1. Predicting the accuracy of a day is higher than the predicted time, indicating that the event is 2 more important for short-term stock forecasting. The title data is the most useful, adding content data, but the prediction accuracy rate decreased by 3. Deep learning is more useful than SVM, but hidden Layer 2 layer effect is best, 3 layer is not good, not try more layers
Related work: 1. Previous people used the word bag feature + classifier to predict 2. On the basis of the feature of the word bag, an extension method is to add the phrase features, such as: noun phrase 3. On the basis of the feature of the word bag, another way to expand is to add semantic features, such as: Through the general user on Twitter sentiment index to predict the market trend
Copyright NOTICE: This article for Bo Master original article, without Bo Master permission not reproduced.
' Deep learning ' Using structured Events to Predict the Stock price Movement:an empirical investigation
