Documentterm

Document-term is a concept used in information retrieval and text mining to describe the relationship between a document and the terms it contains. It is central to the construction of representations that map text data to mathematical structures, such as vectors or matrices. Depending on context, document-term may refer to a pair (document, term) or to the term's occurrence within a document.

The document-term matrix (DTM) is a common representation. It is a two-dimensional sparse matrix with one row per document and one column per term in the vocabulary. Each entry records a weight for the term in the document, most commonly a raw count or a weighted value such as TF, IDF, or TF-IDF.

Term weighting helps distinguish informative terms from common words. TF counts reflect how often a term appears; IDF downscales terms that appear across many documents; TF-IDF combines both effects. Other schemes include binary indicators, log scaling, and sublinear TF.

Construction typically proceeds by collecting a corpus, tokenizing text, normalizing case, removing stopwords, and optionally applying stemming or lemmatization. The resulting vocabulary defines the document-term space, and the chosen weighting scheme determines the numeric matrix used for analysis.

Applications include document classification, clustering, information retrieval, and topic modeling. Limitations include high dimensionality, sparsity, and potential loss of semantic information. Dimensionality reduction, regularization, and more advanced representations such as word embeddings can mitigate these issues.