Abstract:

To overcome the feature selection limitation in the research field of mega-event trend prediction based on massive public news data, based on former researchesand artificial intelligence technologies, it proposes a trend prediction method for mega-event by fusing semantics and event characteristics. First, it uses web crawler technology to assist in capturing relevant news data. Second, it uses topic model and technology of event extraction to assist in feature set construction and vector representation, because the high frequency noise word contained by news data may weak the feature capturing ability of traditional LDA topic model, it proposes an improved topic model IDFLDA. Finally, it uses a machine learning classification model to output prediction result. Taking North Korea’s nuclear behavior trend prediction as the research object to verify this method, the experimental results show that the prediction performance of this method is better than the traditional method based on expert knowledge for feature set construction. It can effectively predict the trend of mega-events and provide support for strategic decision-making.

Key words: mega-event, trend prediction, topic model, event extraction, feature fusion

摘要：

针对当前基于海量公开新闻数据的重大事件趋势预测研究在特征选择上的局限性问题，结合人工智能相关技术对现有方法进行优化改进，提出一种融合语义与事件特征的重大事件趋势预测方法。利用网络爬虫技术辅助数据采集；利用主题模型与事件抽取技术辅助海量新闻数据的特征集构建与向量表示，并针对LDA主题模型在特征词提取上存在偏向性的问题，提出一种改进模型IDFLDA；利用机器学习分类模型进行预测结果输出。以朝鲜核行为预测为例对提出方法进行验证，预测结果表明，该方法的预测性能优于依赖专家知识进行特征集构建的传统方法，能有效进行重大事件的趋势预测，为战略决策提供辅助支持。

关键词: 重大事件, 趋势预测, 主题模型, 事件抽取, 特征融合