This paper contributes to a critical methodological discussion that has direct ramifications for policy studies: how computational methods can be concretely incorporated into existing processes of textual analysis and interpretation without compromising scientific integrity. We focus on the computational method of topic modeling and investigate how it interacts with two larger families of qualitative methods: content and classification methods characterized by interest in words as communication units and discourse and representation methods characterized by interest in the meaning of communicative acts. Based on analysis of recent academic publications that have used topic modeling for textual analysis, our findings show that different mixed‐method research designs are appropriate when combining topic modeling with the two groups of methods. Our main concluding argument is that topic modeling enables scholars to apply policy theories and concepts to much larger sets of data. That said, the use of computational methods requires genuine understanding of these techniques to obtain substantially meaningful results. We encourage policy scholars to reflect carefully on methodological issues, and offer a simple heuristic to help identify and address critical points when designing a study using topic modeling. 本文为一项重要的方法讨论做出了贡献，并可以直接影响政策研究，即如何在不损害科学完整性的情况下，将计算方法具体地纳入现有的文本分析和解读过程中。我们将重点放在主题建模（TM）的计算方法，并研究它如何与以下两类定性方法相互作用：其一为内容和分类方法，其特点是对作为交流单位的词语感兴趣；其二为话语和表征方法，这种方法专注于交流行为的含义。通过分析最近使用主题建模进行文本分析的学术文献，我们发现，将TM与两组方法相结合时，我们应该采用不同的混合方法进行研究设计。我们的主要结论是，主题建模能够帮助研究学者将政策理论和概念应用到更大的数据集。尽管如此，使用计算方法时需要真正理解这些方法才能取得实质上有意义的成果。我们鼓励政策研究者仔细思考方法问题，并提供一种简单的启发式算法，以识别和解决在设计主题建模研究时遇到的关键问题。