Title :
EXOS: Expansion on session for enhancing effectiveness of query auto-completion
Author :
Fang-Hsiang Su;Manas Somaiya;Shrish Mishra;Rajyashree Mukherjee
Author_Institution :
Columbia University, New York, NY USA
Abstract :
Query auto-completion (QAC) is an important feature of a search system that helps users reach their target queries faster. Personalization is a powerful technique that improves the relevance of QAC to the current query. This paper proposes a novel framework, EXOS, to augment the native query by important tokens in the user´s search contexts to provide personalized auto-completions dynamically. EXOS is composed of three major steps: expansion, selection and boosting. Each token from the user´s search contexts is evaluated and stored by EXOS. The tokens with the highest importance are chosen by the expansion step to augment the native query from the user. These augmented queries are used to retrieve QAC results. The selection and boosting models then extract qualified results, merge and re-rank these results based on their query importance to provide the user with a personalized QAC. We designed two large-scale experiments with over six million queries to evaluate the effectiveness of EXOS. Compared with an advanced MostPopularCompletion system, EXOS has 126% gain on Mean Reciprocal Rank (MRR) and 270% gain on Success Rate at Top-1 in a prefix-based experiment. EXOS shows 10.5% gain on first-token MRR and about 9% gain on Token Saved Rate in a token-based experiment. The results of our experiments show that EXOS significantly enhances the relevance of QAC by predicting the user´s target query with using fewer keystrokes.
Keywords :
"Context","Footwear","Search engines","Feature extraction","Predictive models","Context-aware services","Boosting"
Conference_Titel :
Big Data (Big Data), 2015 IEEE International Conference on
DOI :
10.1109/BigData.2015.7363869