Unsupervised Query Categorization using Automatically-Built Concept Graphs

Unsupervised Query Categorization using Automatically-Built Concept Graphs is a paper from Yahoo! for automatic query classification without any training phase. The key idea is to build a network of cross-referencing terms extracted from a search engine's snippets. Categories are then hooked in this graph, by using few seed terms. The categorization algorithm is a variation of a truncated random walk. The results are compared with a basic SVM classifier and with KDD2005. Yahoo! UK is using this system in production.