Temporal and Contextual Evaluation of Background Knowledge Discovery for Short Text Classification
Background Knowledge (BK) plays an essential role in machine learning for short-text and non-topical classification. In this paper the authors present and evaluate two Information Retrieval techniques used to assemble four sets of BK in the past seven years. These sets were applied to classify a commercial corpus of search queries by the apparent age of the user. Temporal and contextual evaluations were used to examine results of various classification scenarios providing insight into choice, significance and range of tuning parameters. The evaluations also demonstrated the impact of the dynamic Web collection on classification results, and the advantages of Automatic Query Expansion (AQE) vs. basic search. The authors discuss other results of this research and its implications on the advancement of short text classification.