Unigrams

Unigrams are the simplest unit in n-gram language models. A unigram is a single element of a sequence, most commonly a word. In word-based unigram models, the probability of a text is approximated by the product of the probabilities of its individual words. Unigrams can also refer to single characters in character-level modeling, where the alphabet letters are treated as tokens.

Use cases: In text classification and information retrieval, unigrams form the basis of bag-of-words representations, where documents are converted into vectors of word counts or frequencies. They are used in term frequency (TF) and TF-IDF weighting, as well as in probabilistic classifiers such as Naive Bayes and in simple language models that estimate P(w) from observed frequencies.

Advantages and limitations: Unigrams are simple and robust to small corpora, fast to compute, and provide a broad view of vocabulary usage. They, however, ignore word order and multiword expressions, and the resulting models can miss important context. Also, rare words can cause high-dimensional sparse representations, requiring smoothing or dimensionality reduction.

Variants and related concepts: In character-level modeling, unigrams are single characters; higher-order n-grams (bigrams, trigrams) capture local word sequences. A unigram model assumes independence among tokens, which is a strong simplification but serves as a useful baseline and feature set.