Open-domain Question Answering (OpenQA) is an important task in Natural Language Processing (NLP), which aims to answer a question in the form of natural language based on large-scale unstructured documents. Recently, there has been a surge in the amount of research literature on OpenQA, particularly on techniques that integrate with neural Machine Reading Comprehension (MRC). While these research works have advanced performance to new heights on benchmark datasets, they have been rarely covered in existing surveys on QA systems. In this work, we review the latest research trends in OpenQA, with particular attention to systems that incorporate neural MRC techniques. Specifically, we begin with revisiting the origin and development of OpenQA systems. We then introduce modern OpenQA architecture named ``Retriever-Reader'' and analyze the various systems that follow this architecture as well as the specific techniques adopted in each of the components. We then discuss key challenges to developing OpenQA systems and offer an analysis of benchmarks that are commonly used. We hope our work would enable researchers to be informed of the recent advancement and also the open challenges in OpenQA research, so as to stimulate further progress in this field.

Recently, Nogueira et al. [2019] proposed a new approach to document expansion based on a neural Seq2Seq model, showing significant improvement on short text retrieval task. However, this approach needs a large amount of in-domain training data. In this paper, we show that this neural document expansion approach can be effectively adapted to standard IR tasks, where labels are scarce and many long documents are present.

Search engines can quickly response a hyperlink list according to query keywords. However, when a query is complex, developers need to repeatedly refine the search keywords and open a large number of web pages to find and summarize answers. Many research works of question and answering (Q and A) system attempt to assist search engines by providing simple, accurate and understandable answers. However, without original semantic contexts, these answers lack explainability, making them difficult for users to trust and adopt. In this paper, a brain-inspired search engine assistant named DeveloperBot based on knowledge graph is proposed, which aligns to the cognitive process of human and has the capacity to answer complex queries with explainability. Specifically, DeveloperBot firstly constructs a multi-layer query graph by splitting a complex multi-constraint query into several ordered constraints. Then it models the constraint reasoning process as subgraph search process inspired by the spreading activation model of cognitive science. In the end, novel features of the subgraph will be extracted for decision-making. The corresponding reasoning subgraph and answer confidence will be derived as explanations. The results of the decision-making demonstrate that DeveloperBot can estimate the answers and answer confidences with high accuracy. We implement a prototype and conduct a user study to evaluate whether and how the direct answers and the explanations provided by DeveloperBot can assist developers' information needs.

Neural networks with deep architectures have demonstrated significant performance improvements in computer vision, speech recognition, and natural language processing. The challenges in information retrieval (IR), however, are different from these other application areas. A common form of IR involves ranking of documents -- or short passages -- in response to keyword-based queries. Effective IR systems must deal with query-document vocabulary mismatch problem, by modeling relationships between different query and document terms and how they indicate relevance. Models should also consider lexical matches when the query contains rare terms -- such as a person's name or a product model number -- not seen during training, and to avoid retrieving semantically related but irrelevant results. In many real-life IR tasks, the retrieval involves extremely large collections -- such as the document index of a commercial Web search engine -- containing billions of documents. Efficient IR methods should take advantage of specialized IR data structures, such as inverted index, to efficiently retrieve from large collections. Given an information need, the IR system also mediates how much exposure an information artifact receives by deciding whether it should be displayed, and where it should be positioned, among other results. Exposure-aware IR systems may optimize for additional objectives, besides relevance, such as parity of exposure for retrieved items and content publishers. In this thesis, we present novel neural architectures and methods motivated by the specific needs and challenges of IR tasks.

Small business classification is a difficult and important task within many applications, including customer segmentation. Training on small business names introduces gender and geographic origin biases. A model for predicting one of 66 business types based only upon the business name was developed in this work (top-1 f1-score = 60.2%). Two approaches to removing the bias from this model are explored: replacing given names with a placeholder token, and augmenting the training data with gender-swapped examples. The results for these approaches is reported, and the bias in the model was reduced by hiding given names from the model. However, bias reduction was accomplished at the expense of classification performance (top-1 f1-score = 56.6%). Augmentation of the training data with gender-swapping samples proved less effective at bias reduction than the name hiding approach on the evaluated dataset.

Such a query can be answered currently at a high human effort cost, by inspecting e.g., a JSON list of Assemblée elected officials (available from NosDeputes.fr) and manually connecting the names with those found in a national registry of companies. This considerable effort may still miss connections that could be found if one added information about politicians' and business people's spouses, information sometimes available in public knowledge bases such as DBPedia, or journalists' notes. No single query language can be used on such heterogeneous data; instead, we study methods to query the corpus by specifying some keywords and asking for all the connections that exist, in one or across several data sources, between these keywords. This problem has been studied under the name of keyword search over structured data, in particular for relational databases [49, 27], XML documents [24, 33], RDF graphs [30, 16]. However, most of these works assumed one single source of data, in which connections among nodes are clearly identified. When authors considered several data sources [31], they still assumed that one query answer comes from a single data source. In contrast, the ConnectionLens system [10] answers keyword search queries over arbitrary combinations of datasets and heterogeneous data models, independently produced by actors unaware of each other's existence.

Named entity recognition (NER) is a well-studied task in natural language processing. However, the widely-used sequence labeling framework is difficult to detect entities with nested structures. In this work, we view nested NER as constituency parsing with partially-observed trees and model it with partially-observed TreeCRFs. Specifically, we view all labeled entity spans as observed nodes in a constituency tree, and other spans as latent nodes. With the TreeCRF we achieve a uniform way to jointly model the observed and the latent nodes. To compute the probability of partial trees with partial marginalization, we propose a variant of the Inside algorithm, the \textsc{Masked Inside} algorithm, that supports different inference operations for different nodes (evaluation for the observed, marginalization for the latent, and rejection for nodes incompatible with the observed) with efficient parallelized implementation, thus significantly speeding up training and inference. Experiments show that our approach achieves the state-of-the-art (SOTA) F1 scores on the ACE2004, ACE2005 dataset, and shows comparable performance to SOTA models on the GENIA dataset. Our approach is implemented at: \url{https://github.com/FranxYao/Partially-Observed-TreeCRFs}.

Slot-filling refers to the task of annotating individual terms in a query with the corresponding intended product characteristics (product type, brand, gender, size, color, etc.). These characteristics can then be used by a search engine to return results that better match the query's product intent. Traditional methods for slot-filling require the availability of training data with ground truth slot-annotation information. However, generating such labeled data, especially in e-commerce is expensive and time-consuming because the number of slots increases as new products are added. In this paper, we present distant-supervised probabilistic generative models, that require no manual annotation. The proposed approaches leverage the readily available historical query logs and the purchases that these queries led to, and also exploit co-occurrence information among the slots in order to identify intended product characteristics. We evaluate our approaches by considering how they affect retrieval performance, as well as how well they classify the slots. In terms of retrieval, our approaches achieve better ranking performance (up to 156%) over Okapi BM25. Moreover, our approach that leverages co-occurrence information leads to better performance than the one that does not on both the retrieval and slot classification tasks.

The ubiquity of smart voice assistants has made conversational shopping commonplace. This is especially true for low consideration segments like grocery. A central problem in conversational grocery is the automatic generation of short product titles that can be read out fast during a conversation. Several supervised models have been proposed in the literature that leverage manually labeled datasets and additional product features to generate short titles automatically. However, obtaining large amounts of labeled data is expensive and most grocery item pages are not as feature-rich as other categories. To address this problem we propose a pre-training based solution that makes use of unlabeled data to learn contextual product representations which can then be fine-tuned to obtain better title compression even in a low resource setting. We use a self-attentive BiLSTM encoder network with a time distributed softmax layer for the title compression task. We overcome the vocabulary mismatch problem by using a hybrid embedding layer that combines pre-trained word embeddings with trainable character level convolutions. We pre-train this network as a discriminator on a replaced-token detection task over a large number of unlabeled grocery product titles. Finally, we fine tune this network, without any modifications, with a small labeled dataset for the title compression task. Experiments on Walmart's online grocery catalog show our model achieves performance comparable to state-of-the-art models like BERT and XLNet. When fine tuned on all of the available training data our model attains an F1 score of 0.8558 which lags the best performing model, BERT-Base, by 2.78% and XLNet by 0.28% only, while using 55 times lesser parameters than both. Further, when allowed to fine tune on 5% of the training data only, our model outperforms BERT-Base by 24.3% in F1 score.

Named entity recognition (NER) is a critical step in modern search query understanding. In the domain of eCommerce, identifying the key entities, such as brand and product type, can help a search engine retrieve relevant products and therefore offer an engaging shopping experience. Recent research shows promising results on shared benchmark NER tasks using deep learning methods, but there are still unique challenges in the industry regarding domain knowledge, training data, and model production. This paper demonstrates an end-to-end solution to address these challenges. The core of our solution is a novel model training framework "TripleLearn" which iteratively learns from three separate training datasets, instead of one training set as is traditionally done. Using this approach, the best model lifts the F1 score from 69.5 to 93.3 on the holdout test data. In our offline experiments, TripleLearn improved the model performance compared to traditional training approaches which use a single set of training data. Moreover, in the online A/B test, we see significant improvements in user engagement and revenue conversion. The model has been live on homedepot.com for more than 9 months, boosting search conversions and revenue. Beyond our application, this TripleLearn framework, as well as the end-to-end process, is model-independent and problem-independent, so it can be generalized to more industrial applications, especially to the eCommerce industry which has similar data foundations and problems.