Document-based Question Answering (DBQA) in Natural Language Processing (NLP) is important but difficult because of the long document and the complex question. Most of previous deep learning methods mainly focus on the similarity computation between two sentences. However, DBQA stems from the reading comprehension in some degree, which is originally used to train and test people's ability of reading and logical thinking. Inspired by the strategy of doing reading comprehension tests, we propose a unified model based on the human-like reading strategy. The unified model contains three major encoding layers that are consistent to different steps of the reading strategy, including the basic encoder, combined encoder and hierarchical encoder. We conduct extensive experiments on both the English WikiQA dataset and the Chinese dataset, and the experimental results show that our unified model is effective and yields state-of-the-art results on WikiQA dataset.

To give a more humanized response in Voice Dialogue Applications (VDAs), inferring emotion states from users’ queries may play an important role. However, in VDAs, we have tremendous amount of VDA users and massive scale of unlabeled data with high dimension features from multimodal information, which challenge the traditional speech emotion recognition methods. In this paper, to better infer emotion from conversational voice data, we proposed a semi-supervised multi-path generative neural network. Specifically, first, we build a novel supervised multi-path deep neural network framework. To avoid high dimensional input, raw features are trained by groups in local classifiers. Then high-level features of each local classifiers are concatenated as input of a global classifier. These two kinds classifiers are trained simultaneously through a single objective function to achieve a more effective and discriminative emotion inferring. To further solve the labeled-data-scarcity problem, we extend the multi-path deep neural network to a generative model based on semi-supervised variational autoencoder (semi-VAE), which is able to train the labeled and unlabeled data simultaneously. Experiment based on a 24,000 real-world dataset collected from Sogou Voice Assistant (SVAD13) and a benchmark dataset IEMOCAP show that our method significantly outperforms the existing state-of-the-art results.

Since many languages originated from a common ancestral language and influence each other, there would inevitably exist similarities between these languages such as lexical similarity and named entity similarity. In this paper, we leverage these similarities to improve the translation performance in neural machine translation. Specifically, we introduce an attention-via-attention mechanism that allows the information of source-side characters flowing to the target side directly. With this mechanism, the target-side characters will be generated based on the representation of source-side characters when the words are similar. For instance, our proposed neural machine translation system learns to transfer the character-level information of the English word "system" through the attention-via-attention mechanism to generate the Czech word "systém." Consequently, our approach is able to not only achieve a competitive translation performance, but also reduce the model size significantly.

Monolingual data have been demonstrated to be helpful in improving translation quality of both statistical machine translation (SMT) systems and neural machine translation (NMT) systems, especially in resource-poor or domain adaptation tasks where parallel data are not rich enough. In this paper, we propose a novel approach to better leveraging monolingual data for neural machine translation by jointly learning source-to-target and target-to-source NMT models for a language pair with a joint EM optimization method. The training process starts with two initial NMT models pre-trained on parallel data for each direction, and these two models are iteratively updated by incrementally decreasing translation losses on training data.In each iteration step, both NMT models are first used to translate monolingual data from one language to the other, forming pseudo-training data of the other NMT model. Then two new NMT models are learnt from parallel data together with the pseudo training data. Both NMT models are expected to be improved and better pseudo-training data can be generated in next step. Experiment results on Chinese-English and English-German translation tasks show that our approach can simultaneously improve translation quality of source-to-target and target-to-source models, significantly outperforming strong baseline systems which are enhanced with monolingual data for model training including back-translation.

User feedback can be an effective indicator to the success of the human-robot conversation. However, to avoid to interrupt the online real-time conversation process, explicit feedback is usually gained at the end of a conversation. Alternatively, users' responses usually contain their implicit feedback, such as stance, sentiment, emotion, etc., towards the conversation content or the interlocutors. Therefore, exploring the implicit feedback is a natural way to optimize the conversation generation process. In this paper, we propose a novel reward function which explores the implicit feedback to optimize the future reward of a reinforcement learning based neural conversation model. A simulation strategy is applied to explore the state-action space in training and test. Experimental results show that the proposed approach outperforms the Seq2Seq model and the state-of-the-art reinforcement learning model for conversation generation on automatic and human evaluations on the OpenSubtitles and Twitter datasets.

Affective image understanding has been extensively studied in the last decade since more and more users express emotion via visual contents. While current algorithms based on convolutional neural networks aim to distinguish emotional categories in a discrete label space, the task is inherently ambiguous. This is mainly because emotional labels with the same polarity (i.e., positive or negative) are highly related, which is different from concrete object concepts such as cat, dog and bird. To the best of our knowledge, few methods focus on leveraging such characteristic of emotions for affective image understanding. In this work, we address the problem of understanding affective images via deep metric learning and propose a multi-task deep framework to optimize both retrieval and classification goals. We propose the sentiment constraints adapted from the triplet constraints, which are able to explore the hierarchical relation of emotion labels. We further exploit the sentiment vector as an effective representation to distinguish affective images utilizing the texture representation derived from convolutional layers. Extensive evaluations on four widely-used affective datasets, i.e., Flickr and Instagram, IAPSa, Art Photo, and Abstract Paintings, demonstrate that the proposed algorithm performs favorably against the state-of-the-art methods on both affective image retrieval and classification tasks.

Designing an e-commerce recommender system that serves hundreds of millions of active users is a daunting challenge. To our best knowledge, the complex brain activity mechanism behind human shopping activities is never considered in existing recommender systems. From a human vision perspective, we found two key factors that affect users’ behaviors: items’ attractiveness and their matching degrees with users’ interests. This paper proposes Telepath, a vision-based bionic recommender system model, which simulates human brain activities in decision making of shopping, thus understanding users from such perspective. The core of Telepath is a complex deep neural network with multiple subnetworks. In practice, the Telepath model has been launched to JD’s recommender system and advertising system and outperformed the former state-of-the-art method. For one of the major item recommendation blocks on the JD app, click-through rate (CTR), gross merchandise value (GMV) and orders have been increased 1.59%, 8.16% and 8.71% respectively by Telepath. For several major ad publishers of JD demand-side platform, CTR, GMV and return on investment have been increased 6.58%, 61.72% and 65.57% respectively by the first launch of Telepath, and further increased 2.95%, 41.75% and 41.37% respectively by the second launch.

It is a challenging and practical research problem to obtain effective compression of lengthy product titles for E-commerce. This is particularly important as more and more users browse mobile E-commerce apps and more merchants make the original product titles redundant and lengthy for Search Engine Optimization. Traditional text summarization approaches often require a large amount of preprocessing costs and do not capture the important issue of conversion rate in E-commerce. This paper proposes a novel multi-task learning approach for improving product title compression with user search log data. In particular, a pointer network-based sequence-to-sequence approach is utilized for title compression with an attentive mechanism as an extractive method and an attentive encoder-decoder approach is utilized for generating user search queries. The encoding parameters (i.e., semantic embedding of original titles) are shared among the two tasks and the attention distributions are jointly optimized. An extensive set of experiments with both human annotated data and online deployment demonstrate the advantage of the proposed research for both compression qualities and online business values.

Recently, domain adaptation based on deep models has been a promising way to deal with the domains with scarce labeled data, which is a critical problem for deep learning models. Domain adaptation propagates the knowledge from a source domain with rich information to the target domain. In reality, the source and target domains are mostly unbalanced in that the source domain is more resource-rich and thus has more reliable knowledge than the target domain. However, existing deep domain adaptation approaches often pre-assume the source and target domains balanced and equally, leading to a medium solution between the source and target domains, which is not optimal for the unbalanced domain adaptation. In this paper, we propose a novel Deep Asymmetric Transfer Network (DATN) to address the problem of unbalanced domain adaptation. Specifically, our model will learn a transfer function from the target domain to the source domain and meanwhile adapting the source domain classifier with more discriminative power to the target domain. By doing this, the deep model is able to adaptively put more emphasis on the resource-rich source domain. To alleviate the scarcity problem of supervised data, we further propose an unsupervised transfer method to propagate the knowledge from a lot of unsupervised data by minimizing the distribution discrepancy over the unlabeled data of two domains. The experiments on two real-world datasets demonstrate that DATN attains a substantial gain over state-of-the-art methods.

Instance Search (INS) is a fundamental problem for many applications, while it is more challenging comparing to traditional image search since the relevancy is defined at the instance level. Existing works have demonstrated the success of many complex ensemble systems that are typically conducted by firstly generating object proposals, and then extracting handcrafted and/or CNN features of each proposal for matching. However, object bounding box proposals and feature extraction are often conducted in two separated steps, thus the effectiveness of these methods collapses. Also, due to the large amount of generated proposals, matching speed becomes the bottleneck that limits its application to large-scale datasets. To tackle these issues, in this paper we propose an effective and efficient Deep Region Hashing (DRH) approach for large-scale INS using an image patch as the query. Specifically, DRH is an end-to-end deep neural network which consists of object proposal, feature extraction, and hash code generation. DRH shares full-image convolutional feature map with the region proposal network, thus enabling nearly cost-free region proposals. Also, each high-dimensional, real-valued region features are mapped onto a low-dimensional, compact binary codes for the efficient object region level matching on large-scale dataset. Experimental results on four datasets show that our DRH can achieve even better performance than the state-of-the-arts in terms of mAP, while the efficiency is improved by nearly 100 times.