We identify the core topics of research applying machine learning to finance. We use a probabilistic topic modeling approach to make sense of this diverse body of research spanning across multiple disciplines. Through a latent Dirichlet allocation topic modeling technique, we extract 15 coherent research topics that are the focus of 5942 academic studies from 1990 to 2020. We find that these topics can be grouped into four categories: Price-forecasting techniques, financial markets analysis, risk forecasting and financial perspectives. We first describe and structure these topics and then further show how the topic focus has evolved over the last three decades. A notable trend we find is the emergence of text-based machine learning, for example, for sentiment analysis, in recent years. Our study thus provides a structured topography for finance researchers seeking to integrate machine learning research approaches in their exploration of finance phenomena. We also showcase the benefits to finance researchers of the method of probabilistic modeling of topics for deep comprehension of a body of literature.

中文翻译：

---

金融中的机器学习：一种主题建模方法

我们确定了将机器学习应用于金融的研究的核心主题。我们使用概率主题建模方法来理解这种跨越多个学科的多样化研究。通过潜在的狄利克雷分配主题建模技术，我们提取了 15 个连贯的研究主题，这些主题是 1990 年至 2020 年 5942 项学术研究的重点。我们发现这些主题可以分为四类：价格预测技术、金融市场分析、风险预测和财务观点。我们首先描述和构建这些主题，然后进一步展示主题焦点在过去三年中是如何演变的。我们发现的一个显着趋势是近年来出现了基于文本的机器学习，例如情绪分析。因此，我们的研究为寻求将机器学习研究方法整合到金融现象探索中的金融研究人员提供了一个结构化的拓扑结构。我们还向金融研究人员展示了主题概率建模方法对深入理解大量文献的好处。