Topic Model

Information Retrieval

Evaluation and Interpretation

Guided LDA to seed topics with specific words or more recently CatE.

Dynamic Topic Model: each topic has a distinct distribution over words for each time period

cDTM: time is considered continuously

Author-Topic model: an author has a collection of topics that they write about and each document is a combination of the topics that its set of authors care about.

Inheritance topic model: adapt LDA to a citation network

Dynamic influence model: Find most influential documents

LinkLDA: relationship between documents

Short documents

• it can confuse topic modeling algorithms see these studies (see for a specific application in Twitter; also, interesting on how to apply and analyze topic models)
• to capture trends over time or across users, algorithms must know these connections between users or messages
• Topic model for sparse data or short texts

Supervised Topic model

Results vary based on the number of topics. The nested Dirichlet process provide a way of selecting the appropriate number of topics.

• Labelled LDA: it can do multi-class (see application on Twitter)