This paper provides a structured and extensive survey on context learning methods. It groups context learning methods into four categories, in order of complexity: explicit analysis, implicit analysis, neural-network-based analysis, and composite analysis.

Among explicit analysis methods, the paper details m-gram models, link-based analysis, explicit semantic analysis, and ensemble-based analysis. It also benchmarks explicit analysis methods, and points out that ensemble-based analysis methods outperform other explicit analysis methods.

Among implicit analysis methods, the paper details latent semantic indexing (LSI), mixture of unigrams, probabilistic LSI (PLSI)-based analysis, and latent Dirichlet allocation (LDA)-based analysis. It also benchmarks implicit analysis methods, and points out that LDA-based analysis outperforms other implicit analysis methods.

Among neural-network-based analysis methods, the paper details the neural-network-based bi-gram model, the neural probabilistic language model (NPLM) and its variants, and restricted Boltzmann machine-based analysis. The paper points out the advantages of neural-network-based analysis over explicit and implicit analysis methods.

Among composite models, the paper details explicit analysis plus neural networks, explicit analysis plus implicit analysis, and implicit analysis plus neural networks.

Overall, this paper presents an extensive survey on context learning, grouped into four different categories, which is a new categorization in this field. The paper can be used as an informative guide for anyone interested in context learning methods.