两种基于向量化策略SVM分类器的对比分析

以股票涨跌趋势预测精度为评价指标,针对传统股票数据特征训练过程中预测精度不高的情况,考虑引入两种不同的向量化策略对股民评论、新闻关键词等文本信息进行非结构化数据特征的捕捉,利用词意的积极、消极程度对客观因素进行处理,进而将向量化后的特征作为新的非线性特征项扩充原有的结构化特征集合。文中分别以词向量化和句向量化为出发点设计两种启发式的SVM分类器,其目标是在拟合每支股票的情况下尽可能预测出其未来的走势,挖掘出更具有增长潜力的股票样本。经过2018年6月至12月半年沪市股票数据集的实验结果表明,相比于词向量化策略,采用句向量化策略设计的SVM分类器不仅能够更好地预测股票涨跌,并且能够更有效地挑选出潜在增长的股票样本。

Comparison Analysis between Two Vectorization Strategy Based SVM Classifiers

With the accuracy of stock trend prediction as the evaluation index,two different vectorization strategies are introduced to capture the unstructured data characteristics of shareholders’comments,news keywords and other text information in the light of the low accuracy in the traditional stock data training process.Based on the positive and negative degree of lexical meaning,the objective factors are processed,and the vectorized features are used as new nonlinear features to expand the original structural feature set.We design two kinds of heuristic SVM classifiers from the perspective of word vectorization and sentence vectorization respectively so as to predict the future trend of each stock as far as possible under the condition of fitting each stock and dig out the stock samples with more growth potential.The experimental results of the Shanghai Stock Market data set from June to December 2018 show that compared with the word vectorization strategy,the SVM classifier designed by the sentence vectorization strategy can not only better predict the stock trend,but also pick out the stock samples with potential growth more effectively.