Semanlink - Biterm Topic Model

Parents:

Topic Modeling over Short Texts

Biterm Topic Model

Word co-occurrence based topic model that learns topics by modeling word-word co-occurrences patterns. (In constrast, LDA and PLSA are word-document co-occurrence topic models)

A biterm consists of two words co-occurring in the same context (short text). BTM models the biterm occurrences in a corpus.

Conventional topic models exploit word co-occurrence patterns to reveal the latent semantic structure of a corpus in an implicit way by modeling the generation of words in each document. These approaches are sensitive to the shortness of documents since the word co- occurrence patterns in a single short document are sparse and not reliable. Instead, if we aggregate all the word co-occurrence patterns in the corpus, their frequencies are more stable and more clearly reveals the correlation between the words. With this idea, we developed the biterm topic model, which takes a novel way to reveal the latent topic components in a corpus by directly modeling the generation of word co-occurrence patterns.

Related Tags: