Separate and Attend in Personal Email Search

Abstract

In personal email search, user queries often impose different requirements on different aspects of the retrieved emails. For example, the query "my recent flight to the US" requires emails to be ranked based on both textual contents and recency of the email documents, while other queries such as "medical history" do not impose any constraints on the recency of the email. Recent deep learning-to-rank models for personal email search often directly concatenate dense numerical features 1 (e.g., document age) with embedded sparse features (e.g., n-gram embeddings). In this paper, we first show with a set of experiments on synthetic datasets that direct concatenation of dense and sparse features does not lead to the optimal search performance of deep neural ranking models. To effectively incorporate both sparse and dense email features into personal email search ranking, we propose a novel neural model, SepAttn. SepAttn first builds two separate neural models to learn from sparse and dense features respectively, and then applies an attention mechanism at the prediction level to derive the final prediction from these two models. We conduct a comprehensive set of experiments on a large-scale email search dataset, and demonstrate that our SepAttn model consistently improves the search quality over the baseline models.

Leveraging User Behavior History for Personalized Email Search

“…Shen et al [27] clustered queries based on the frequent n-grams of results retrieved with a baseline ranker and then used the query cluster information as an auxiliary objective in multi-task learning. There has also been research that studies how to combine sparse and dense features in a unified model effectively [22], combine textual information from queries and documents with other side information to conduct effective and efficient learning to rank [24], and transfer models learned from one domain to another in email search [30].…”

Overcoming Prior Misspecification in Online Learning to Rank

“…In offline LTR, it is assumed that the ground truth utility of the lists has been provided and the goal is to learn a scoring function which can be used to rank the items (Zamani et al, 2017;Mitra and Craswell, 2017;Shen et al, 2018;Meng et al, 2020). In this setting, it is implicitly assumed that user behavior is time-invariant, however, in many real-world problems user preferences can change dynamically.…”