Photo: Bruce Turner from AustinTX / Wikimedia Commons / CC BY 2.0
Text mining and network analysis power digital humanities because they help scholars study large groups of texts and records. Text mining finds patterns in words. Network analysis maps links among people, texts, ideas, or places. Together, these methods support broad reading while keeping the source material in view.
I think the key point is scale. A scholar can read one novel closely. A computer can count words across thousands of novels. It can also find repeated phrases, names, dates, and word groups. This does not replace close reading. It helps show where close reading may be most useful.
Text mining means using a computer to study written language. The simplest form counts words in a set of texts, called a corpus. More advanced methods find phrases, compare texts, group related words, or identify topics. A topic model is a computer method that groups words that often appear near each other. These groups can suggest themes for further study.
The results depend on the text file. A scan may contain errors. Old spelling may confuse a search. A name may appear in several forms. A word count may also ignore meaning. The word “bank,” for example, can refer to money or land beside a river. A count alone cannot decide which meaning applies.
This is why text mining works best as a way to ask better questions. A pattern may show that a term grows more common over time. It may show that two authors use similar language. It may reveal words that appear together across many texts. Each result needs a return to the source and its historical setting.
Network analysis adds a second view. It represents items as points, often called nodes. It represents links between them as edges. A node might be an author, book, character, place, or concept. An edge might mean that two authors worked together, two books share a citation, or two characters appear in the same scene.
The choice of nodes and edges shapes the result. A network of authors can show collaboration. A network of books can show shared references. A network of words can show repeated co-occurrence. These are different questions, even when they use the same corpus.
One common measure is centrality. It estimates which nodes have an important position in a network. Another measure can show clusters, or groups with many links among themselves. Such measures can help locate major authors, key texts, or terms that connect separate groups.
These methods often work together. Text mining can extract names, places, or phrases from documents. Network analysis can then map the links among those items. A study of letters, for example, might count terms about travel and map the people who exchanged letters. The count shows language use. The network shows social ties.
The method also supports what scholars call distant reading. This means studying patterns across many texts rather than reading each text from start to finish. Distant reading can show broad change. Close reading can then test what those broad patterns mean in selected passages.
The order matters. First comes the corpus. A corpus is the group of texts chosen for study. Its size, date range, language, and selection rules affect every result. A network built from a narrow set of texts may be useful for that set. It cannot speak for all writing from the same period.
Then comes cleaning. Computers need consistent data. Scholars may need to remove scan errors, mark speakers, join name forms, or separate editions. These choices are part of the research. They are not minor technical tasks. A different cleaning rule can produce a different pattern.
Metadata also matters. Metadata is information about a record, such as its author, date, title, or place of publication. Good metadata can support strong networks. Missing or uneven metadata can hide links. A database may list one author in full and another by initials. A computer may treat them as different people.
The main limit is interpretation. Text mining detects patterns in recorded data. It does not understand a text in the full human sense. Network analysis displays relations that researchers define. It does not prove that those relations were meaningful to people at the time.
A graph can look exact because it uses numbers and lines. That appearance can be misleading. A large node may reflect many records, not great cultural importance. A missing link may reflect poor cataloging, lost evidence, or a rule in the dataset. The image cannot explain these limits on its own.
There is also a risk of treating digital collections as neutral samples. Digitized texts often reflect past collecting choices. Some languages, regions, genres, and communities are better represented than others. Copyright rules and scanning projects also shape what scholars can access. A large corpus may still offer a narrow view.
For that reason, I treat the method as evidence for inquiry, not as an automatic answer. The strongest work states what the data contains, how it was prepared, and what it leaves out. It also separates a computer result from a human claim about history or meaning.
The practical search tip is simple: begin with a small, clear unit. Search for one term, one name form, or one known relation before building a large network. This makes errors easier to see. It also shows whether the source has enough detail for the question.
Text mining and network analysis power digital humanities by joining scale with structure. One method finds patterns in language. The other makes relationships visible. Their value depends on careful source choice, clear data rules, and modest claims.
That is the promise of The Source List: one digital source worth knowing, one search tip, and one honest limitation. For this method, the limitation belongs beside the result, not after it.