Evans, James and Pedro Aceves. 2016. “Machine Translation: Mining Text for Social Theory”. Annual Review of Sociology 42:21-50. DOI: 10.1146/annurev-soc-081715-074206
In the paper, you mention how, with many unsupervised learning methods, small tweaks can produce very different results (e.g., clustering methods on pg. 32); how can a researcher maximize the flexibility and power of these methods without inadvertently 'fixing' their conclusions (or just capturing noise)? Moreover, how does one strike a balance between producing results that are interpretable and avoiding telling a 'just-so' story?
In the paper, you mention how, with many unsupervised learning methods, small tweaks can produce very different results (e.g., clustering methods on pg. 32); how can a researcher maximize the flexibility and power of these methods without inadvertently 'fixing' their conclusions (or just capturing noise)? Moreover, how does one strike a balance between producing results that are interpretable and avoiding telling a 'just-so' story?
Thanks!