A Review of Graph-Based Methods in Semi-Supervised Learning

Nowadays, due to the increasing growth of information bulk, it seems necessary to have a system to automatically classify the texts. In past 10 years, management based on text content has gained more account as a consequence of rapid growth and availability of textual documents in digital form. Text classification is used for the practice of subject-based labeling of natural language texts, in accordance with a pre-determined set. Currently, text classification is practical in wide range of contexts, from text indexing, based on a controlled glossary, to text filtering, automatic production of metadata, word clarification, production of hierarchical catalogues from the existing web sources, and in general in any case wherein the aim is to organize the documents or distribute them selectively and comparatively in a certain way. This paper deals with graph-construction methods, surveying five graph-based methods in semi-supervised learning, namely Min-cut method, Manifold Regulation, multiple label, harmonical compositions, and harmonical function by means of Laplasian Matrix.

کلیدواژه ها:

data-mining ، comparison ، graph construction ، semi-supervised learning