Abstract Comparing and contrasting qualitative and quantitative methods for social media data exploration, this article describes and demonstrates the topic modeling approach for the descriptive analysis of large unstructured text data. Using a sample of tweets with the #WhyIStayed and #WhyILeft hashtags (n = 3,068), a Twitter conversation describing the reasons individuals left or stayed in abusive relationships, a traditional thematic analysis was used to qualitatively code the tweets. The same tweet sample was subject to a series of quantitative topic models. Results suggest topic modeling as a comparable approach to first-round qualitative analysis, with key differences: topic modeling and traditional thematic analysis are both inductive and phenomenon-oriented, but topic modeling results in a lexical semantic analysis, in contrast to the compositional semantic analysis offered by the qualitative approach. An evaluation of topics and codes using the Linguistic Inquiry and Word Count (LIWC) software further supports these findings. We argue topic modeling is a useful method for the descriptive analysis of unstructured social media data sets, and is best used as part of a mixed-method strategy, with topic model results guiding deeper qualitative analysis. Implications for human service intervention development and evaluation are discussed.

中文翻译：

---

定性数据探索的计算社会科学视角：使用主题模型对社交媒体数据进行描述性分析*

摘要比较和对比定性和定量的社交媒体数据探索方法，本文描述并演示了用于大型非结构化文本数据描述性分析的主题建模方法。使用带有#WhyIStayed和#WhyILeft主题标签（n = 3,068）的推文样本，在Twitter对话中描述个人离开或停留在虐待关系中的原因，使用传统的主题分析定性地对推文进行编码。相同的推文样本受一系列定量主题模型的约束。结果表明，主题建模是第一轮定性分析的可比方法，但有以下主要区别：主题建模和传统主题分析都是归纳的和面向现象的，但是主题建模会产生词法语义分析，与定性方法提供的成分语义分析相反。使用语言查询和单词计数（LIWC）软件对主题和代码进行的评估进一步支持了这些发现。我们认为主题建模是一种用于非结构化社交媒体数据集描述性分析的有用方法，并且最好用作混合方法策略的一部分，主题模型结果可指导更深入的定性分析。讨论了对人类服务干预发展和评估的影响。最好用作混合方法策略的一部分，主题模型结果指导进行更深入的定性分析。讨论了对人类服务干预发展和评估的影响。最好用作混合方法策略的一部分，主题模型结果指导进行更深入的定性分析。讨论了对人类服务干预发展和评估的影响。