Muthusamy Chelliah (Flipkart.com)
Saptarshi Ghosh (Indian Institute of Technology, Kharagpur)
Date: 8th December (9:30 - 13:00)
Query interpretation helps infer user intent better and augmentation improves search experience. Techniques like classification, tagging and segmentation as well as suggestion and reformulation respectively are abound in IR literature. Compared with general Web search, product retrieval affords unique opportunities: structured data backend (catalog) and rich semantics due to readily available attribute-value pairs. Lack of domain knowledge, limited/dynamic inventory, long/sparse tail of query distribution and monetary impact of search on purchase transactions are challenging to tackle for helping focused/exploratory buyers shopping for the right deal in online marketplaces.
In this tutorial, we first review relevant methods from industrial research such as semantic query relations, mapping query modifiers to product specs based on user behaviour and leveraging query category for term deletion. Beyond such practical implementations, we next highlight recent results as future research directions, e.g., unsupervised methods for query clustering/annotation, graphical model for query facet extraction and transition graph for term/topic-level suggestions. Our intent is thus 2-fold for the audience: extend existing shopping vertical search engines with new advances in query understanding/enhancement and based on this experience develop further such state-of-the-art results in query mining/rewrite.