To generate large scale high quality queries and query-document relevance labels, we sample query-document clicks from one year of Bing search engine’s logs. The initial query set gets filtered to remove queries that are rarely triggered, contain personally identifiable information, offensive content, adult content and those having no click connection to the ClueWeb22 document set. The resulting set includes queries triggered by many users, which reflects the real query distribution of a commercial web search engine.
The queries are split into train and test sets based on time, which is similar to real-world web scenarios training an embedding model using past data and serving future incoming web pages and queries. We sample around 10 million query-document pairs from the train set and 10 thousand query-document pairs from the test set. The documents in the query-document train and test sets are then merged into the 100 million train document set and test document set respectively (shown in right part of figure 1). To enable quality verification of the model during training, we split a dev query-document set from the train query-document set. Since the train and dev sets share the same document set, the dev set can be used to quickly verify the training correctness and model quality during training. For the 10B dataset, we use the same train, dev, and test queries but sample more query-document pairs.
Authors:
(1) Qi Chen, Microsoft Beijing, China;
(2) Xiubo Geng, Microsoft Beijing, China;
(3) Corby Rosset, Microsoft, Redmond, United States;
(4) Carolyn Buractaon, Microsoft, Redmond, United States;
(5) Jingwen Lu, Microsoft, Redmond, United States;
(6) Tao Shen, University of Technology Sydney, Sydney, Australia and the work was done at Microsoft;
(7) Kun Zhou, Microsoft, Beijing, China;
(8) Chenyan Xiong, Carnegie Mellon University, Pittsburgh, United States and the work was done at Microsoft;
(9) Yeyun Gong, Microsoft, Beijing, China;
(10) Paul Bennett, Spotify, New York, United States and the work was done at Microsoft;
(11) Nick Craswell, Microsoft, Redmond, United States;
(12) Xing Xie, Microsoft, Beijing, China;
(13) Fan Yang, Microsoft, Beijing, China;
(14) Bryan Tower, Microsoft, Redmond, United States;
(15) Nikhil Rao, Microsoft, Mountain View, United States;
(16) Anlei Dong, Microsoft, Mountain View, United States;
(17) Wenqi Jiang, ETH Zürich, Zürich, Switzerland;
(18) Zheng Liu, Microsoft, Beijing, China;
(19) Mingqin Li, Microsoft, Redmond, United States;
(20) Chuanjie Liu, Microsoft, Beijing, China;
(21) Zengzhong Li, Microsoft, Redmond, United States;
(22) Rangan Majumder, Microsoft, Redmond, United States;
(23) Jennifer Neville, Microsoft, Redmond, United States;
(24) Andy Oakley, Microsoft, Redmond, United States;
(25) Knut Magne Risvik, Microsoft, Oslo, Norway;
(26) Harsha Vardhan Simhadri, Microsoft, Bengaluru, India;
(27) Manik Varma, Microsoft, Bengaluru, India;
(28) Yujing Wang, Microsoft, Beijing, China;
(29) Linjun Yang, Microsoft, Redmond, United States;
(30) Mao Yang, Microsoft, Beijing, China;
(31) Ce Zhang, ETH Zürich, Zürich, Switzerland and the work was done at Microsoft.
