Integrating structured knowledge into language model representations increases recall of domain-specific information useful for downstream tasks. Matching between knowledge graph entities and text entity mentions can be easily performed when entity names are unique or there exists entity linking data. When extending this setting to new domains, newly mined knowledge contains ambiguous and incorrect information, with no explicit linking information. In such settings, we design a framework to robustly link relevant knowledge to input texts as an intermediate modeling step while performing end-to-end domain fine-tuning tasks. This is done by first computing the similarity of the existing task labels with candidate knowledge triplets to generate relevance labels. We use these labels to train a relevance model, which predicts the relevance of the inserted triplets to the original text. This relevance model is integrated within a language model, leading to our Knowledge Relevance BERT (KR-BERT) framework. We test KR-BERT for entity linking tasks on a real-world e-commerce dataset as well as a public linking task, where we show performance improvements over strong baselines.

Hierarchies are common structures used to organize data, such as e-commerce hierarchies associated with product data. With these product hierarchies, we aim to learn hierarchy-aware product text embeddings to improve fine-tuning performance on a variety of downstream e-commerce tasks. Existing methods leverage hierarchies by either aligning the text embeddings to separate hierarchical embeddings or by aligning the hierarchical information implicitly within a unified text Transformer. Although these models optimize to predict hierarchy information, performing further fine-tuning on new tasks is non-trivial. To bridge this gap, we propose a pre-training architecture to implicitly encode the hierarchy within the product text and then directly leverage a sub-set of the pre-training model during fine-tuning. Pre-training is done through Semantic and Structural Transformers (SST) where the Semantic-Transformer first encodes the product text into a contextual embedding, which is then used by the Structural-Transformer to infer the product’s path in the hierarchy. Fine-tuning is done using only the initial Semantic-Transformer, now that hierarchy-aware text embeddings are learned. With this design, we eliminate the need of linking each fine-tuning dataset with corresponding hierarchies. This leads to fine-tuning performance improvements on critical e-commerce downstream tasks over the existing state-of-the-art hierarchy models, even when hierarchy data $is$ available during fine-tuning. Moreover, this improvement is consistent even after augmenting our baseline models to support fine-tuning. We conclude by discussing how such implicit structural encodings can be leveraged beyond the e-commerce domain.

Bag-of-words based lexical retrieval systems are still the most commonly used methods for real-world search applications. Recently deep learning methods have shown promising results to improve this retrieval performance but are expensive to run in an online fashion, non-trivial to integrate into existing production systems, and might not generalize well in out-of-domain retrieval scenarios. Instead, we build on top of lexical retrievers by proposing a Term Weighting BERT (TW-BERT) model. TW-BERT learns to predict the weight for individual n-gram (e.g., uni-grams and bi-grams) query input terms. These inferred weights and terms can be used directly by a retrieval system to perform a query search. To optimize these term weights, TW-BERT incorporates the scoring function used by the search engine, such as BM25, to score query-document pairs. Given sample query-document pairs we can compute a ranking loss over these matching scores, optimizing the learned query term weights in an end-to-end fashion. Aligning TW-BERT with search engine scorers minimizes the changes needed to integrate it into existing production applications, whereas existing deep learning based search methods would require further infrastructure optimization and hardware requirements. The learned weights can be easily utilized by standard lexical retrievers and by other retrieval techniques such as query expansion. We show that TW-BERT improves retrieval performance over strong term weighting baselines within MSMARCO and in out-of-domain retrieval on TREC datasets.

The great success of supervised learning has initiated a paradigm shift from building a deterministic software system to a probabilistic artificial intelligent system throughout the industry. The historical records in enterprise domains can potentially bootstrap the traditional business into the modern data-driven approach almost everywhere. The introduction of the Deep Neural Networks (DNNs) significantly reduces the efforts of feature engineering so that supervised learning becomes even more automated. The last bottleneck is to ensure the data quality, particularly the label quality, because the performance of supervised learning is bounded by the errors present in labels. In this paper, we present a new Active Deep Denoising (ADD) approach that first builds a DNN noise model, and then adopts an active learning algorithm to identify the optimal denoising function. We prove that under the low noise condition, we only need to query the oracle with log n examples where n is the total number in the data. We apply ADD on one enterprise application and show that it can effectively reduce 1/3 of the prediction error with only 0.1% of examples verified by the oracle.