My personal highlights from the accepted paper presentations include:
- Claudio Carpineto et al. – Semantic Search Log k-Anonymization with Generalized k-Cores of Query Concept Graph. (shared best paper) For k-anonymization of search engine log files, unique or infrequent queries can be removed in order to prevent individual users from being identifiable. Massive pruning can, however, significantly reduce the coverage of log files. The authors propose a clustering method to anonymize log files based on query similarity rather than identity.
- Yashar Moshfeghi et al. – Understanding Relevance: An fMRI Study. (shared best paper) Are there neural differences in the processing of relevant and non-relevant documents? How does the human brain react to relevance? The authors show a first investigation into this domain.
- Aleksandr Chuklin et al. – Using Intent Information to Model User Behavior in Diversified Search. (best student paper) The authors propose an intent-aware click model to estimate relevance based on the user’s underlying search intent.
- Marc Bron et al. – Example Based Entity Search in the Web of Data. Using positive examples for entity search, the authors show performance gains when enriching entity queries with knowledge gained from the context of provided examples.
- Van Dang et al. – Two-Stage Learning
to Rank for Information Retrieval. The authors introduce a multi-stage bootstrapped learning to rank process.
- Dongyi Guan et al. – Increasing Stability of Result Organization for Session Search. For faceted search, the authors employ external resources such as Wikipedia to improve the performance of the underlying result clustering.
- Xiaofei Zhu et al. – Recommending High Utility Query via Session-Flow Graph. Based on random walks in the clickthrough graph, the authors motivate a query recommendations scheme that focuses on high-utility queries. Such queries have been estimated to return more useful results for the user.
- Maksim Zhukovskii et al. – URL Redirection Accounting for Improving Link-Based Ranking Methods. The authors show how redirections can significantly obscure web graphs used to compute PageRank and other structural quality indicators.
- Nima Asadi et al. – Training Efficient Tree-Based Models for Document Ranking. The authors investigate the balanced creation of CART trees for LambdaMART learning to rank schemes. By biasing algorithms towards balanced, shallow tree ensembles, they show significant efficiency gains at only minuscule losses in ranking performance.