The computing cost of transformer self-attention often necessitates breaking long documents to fit in pretrained models in document ranking tasks. In this paper, we design Query-Directed Sparse attention that induces IR-axiomatic structures in transformer self-attention. Our model, QDS-Transformer, enforces the principle properties desired in ranking: local contextualization, hierarchical representation, and query-oriented proximity matching, while it also enjoys efficiency from sparsity. Experiments on four fully supervised and few-shot TREC document ranking benchmarks demonstrate the consistent and robust advantage of QDS-Transformer over previous approaches, as they either retrofit long documents into BERT or use sparse attention without emphasizing IR principles. We further quantify the computing complexity and demonstrates that our sparse attention with TVM implementation is twice more efficient that the fully-connected self-attention. All source codes, trained model, and predictions of this work are available at https://github.com/hallogameboy/QDS-Transformer.
Generative neural networks have been shown effective on query suggestion. Commonly posed as a conditional generation problem, the task aims to leverage earlier inputs from users in a search session to predict queries that they will likely issue at a later time. User inputs come in various forms such as querying and clicking, each of which can imply different semantic signals channeled through the corresponding behavioral patterns. This paper induces these behavioral biases as hypotheses for query generation, where a generic encoder-decoder Transformer framework is presented to aggregate arbitrary hypotheses of choice. Our experimental results show that the proposed approach leads to significant improvements on top-k word error rate and Bert F1 Score compared to a recent BART model.
Conventional word embeddings are trained with specific criteria (e.g., based on language modeling or co-occurrence) inside a single information source, disregarding the opportunity for further calibration using external knowledge. This paper presents a unified framework that leverages pre-learned or external priors, in the form of a regularizer, for enhancing conventional language model-based embedding learning. We consider two types of regularizers. The first type is derived from topic distribution by running LDA on unlabeled data. The second type is based on dictionaries that are created with human annotation efforts. To effectively learn with the regularizers, we propose a novel data structure, trajectory softmax, in this paper. The resulting embeddings are evaluated by word similarity and sentiment classification. Experimental results show that our learning framework with regularization from prior knowledge improves embedding quality across multiple datasets, compared to a diverse collection of baseline methods.
Many important email-related tasks, such as email classification or search, highly rely on building quality document representations (e.g., bag-of-words or key phrases) to assist matching and understanding. Despite prior success on representing textual messages, creating quality user representations from emails was overlooked. In this paper, we propose to represent users using embeddings that are trained to reflect the email communication network. Our experiments on Enron dataset suggest that the resulting embeddings capture the semantic distance between users. To assess the quality of embeddings in a real-world application, we carry out auto-foldering task where the lexical representation of an email is enriched with user embedding features. Our results show that folder prediction accuracy is improved when embedding features are present across multiple settings.