Due to their high retrieval efficiency and low storage cost for cross-modal search task, cross-modal hashing methods have attracted considerable attention. For supervised cross-modal hashing methods, how to make the learned hash codes preserve semantic structure information sufficiently is a key point to further enhance the retrieval performance. As far as we know, almost all supervised cross-modal hashing methods preserve semantic structure information depending on at-least-one similarity definition fully or partly, i.e., it defines two datapoints as similar ones if they share at least one common category otherwise they are dissimilar. Obviously, the at-least-one similarity misses abundant semantic structure information. To tackle this problem, in this paper, we propose a novel Deep Cross-modal Proxy Hashing, called DCPH. Specifically, DCPH first learns a proxy hashing network to generate a discriminative proxy hash code for each category. Then, by utilizing the learned proxy hash code as supervised information, a novel $Margin$-$SoftMax$-$like\ loss$ is proposed without defining the at-least-one similarity between datapoints. By minimizing the novel $Margin$-$SoftMax$-$like\ loss$, the learned hash codes will simultaneously preserve the cross-modal similarity and abundant semantic structure information well. Extensive experiments on two benchmark datasets show that the proposed method outperforms the state-of-the-art baselines in cross-modal retrieval task.
Node classification in structural networks has been proven to be useful in many real world applications. With the development of network embedding, the performance of node classification has been greatly improved. However, nearly all the existing network embedding based methods are hard to capture the actual category features of a node because of the linearly inseparable problem in low-dimensional space; meanwhile they cannot incorporate simultaneously network structure information and node label information into network embedding. To address the above problems, in this paper, we propose a novel Deep Kernel Supervised Hashing (DKSH) method to learn the hashing representations of nodes for node classification. Specifically, a deep multiple kernel learning is first proposed to map nodes into suitable Hilbert space to deal with linearly inseparable problem. Then, instead of only considering structural similarity between two nodes, a novel similarity matrix is designed to merge both network structure information and node label information. Supervised by the similarity matrix, the learned hashing representations of nodes simultaneously preserve the two kinds of information well from the learned Hilbert space. Extensive experiments show that the proposed method significantly outperforms the state-of-the-art baselines over three real world benchmark datasets.
Currently, open-domain generative dialog systems have attracted considerable attention in academia and industry. Despite the success of single-turn dialog generation, multi-turn dialog generation is still a big challenge. So far, there are two kinds of models for open-domain multi-turn dialog generation: hierarchical and non-hierarchical models. Recently, some works have shown that the hierarchical models are better than non-hierarchical models under their experimental settings; meanwhile, some works also demonstrate the opposite conclusion. Due to the lack of adequate comparisons, it's not clear which kind of models are better in open-domain multi-turn dialog generation. Thus, in this paper, we will measure systematically nearly all representative hierarchical and non-hierarchical models over the same experimental settings to check which kind is better. Through extensive experiments, we have the following three important conclusions: (1) Nearly all hierarchical models are worse than non-hierarchical models in open-domain multi-turn dialog generation, except for the HRAN model. Through further analysis, the excellent performance of HRAN mainly depends on its word-level attention mechanism; (2) The performance of other hierarchical models will also obtain a great improvement if integrating the word-level attention mechanism into these models. The modified hierarchical models even significantly outperform the non-hierarchical models; (3) The reason why the word-level attention mechanism is so powerful for hierarchical models is because it can leverage context information more effectively, especially the fine-grained information. Besides, we have implemented all of the models and already released the codes.
In this work, we formulate cross-lingual language model pre-training as maximizing mutual information between multilingual-multi-granularity texts. The unified view helps us to better understand the existing methods for learning cross-lingual representations. More importantly, the information-theoretic framework inspires us to propose a pre-training task based on contrastive learning. Given a bilingual sentence pair, we regard them as two views of the same meaning, and encourage their encoded representations to be more similar than the negative examples. By leveraging both monolingual and parallel corpora, we jointly train the pretext tasks to improve the cross-lingual transferability of pre-trained models. Experimental results on several benchmarks show that our approach achieves considerably better performance. The code and pre-trained models are available at http://aka.ms/infoxlm.
Recently, open-domain dialogue systems have attracted growing attention. Most of them use the sequence-to-sequence (Seq2Seq) architecture to generate responses. However, traditional Seq2Seq-based open-domain dialogue models tend to generate generic and safe responses, which are less informative, unlike human responses. In this paper, we propose a simple but effective keywords-guided Sequence-to-Sequence model (KW-Seq2Seq) which uses keywords information as guidance to generate open-domain dialogue responses. Specifically, KW-Seq2Seq first uses a keywords decoder to predict some topic keywords, and then generates the final response under the guidance of them. Extensive experiments demonstrate that the KW-Seq2Seq model produces more informative, coherent and fluent responses, yielding substantive gain in both automatic and human evaluation metrics.
As a key component in a dialogue system, dialogue state tracking plays an important role. It is very important for dialogue state tracking to deal with the problem of unknown slot values. As far as we known, almost all existing approaches depend on pointer network to solve the unknown slot value problem. These pointer network-based methods usually have a hidden assumption that there is at most one out-of-vocabulary word in an unknown slot value because of the character of a pointer network. However, often, there are multiple out-of-vocabulary words in an unknown slot value, and it makes the existing methods perform bad. To tackle the problem, in this paper, we propose a novel Context-Sensitive Generation network (CSG) which can facilitate the representation of out-of-vocabulary words when generating the unknown slot value. Extensive experiments show that our proposed method performs better than the state-of-the-art baselines.
Open-domain generative dialogue systems have attracted considerable attention over the past few years. Currently, how to automatically evaluate them, is still a big challenge problem. As far as we know, there are three kinds of automatic methods to evaluate the open-domain generative dialogue systems: (1) Word-overlap-based metrics; (2) Embedding-based metrics; (3) Learning-based metrics. Due to the lack of systematic comparison, it is not clear which kind of metrics are more effective. In this paper, we will first measure systematically all kinds of automatic evaluation metrics over the same experimental setting to check which kind is best. Through extensive experiments, the learning-based metrics are demonstrated that they are the most effective evaluation metrics for open-domain generative dialogue systems. Moreover, we observe that nearly all learning-based metrics depend on the negative sampling mechanism, which obtains an extremely imbalanced and low-quality dataset to train a score model. In order to address this issue, we propose a novel and feasible learning-based metric that can significantly improve the correlation with human judgments by using augmented POsitive samples and valuable NEgative samples, called PONE. Extensive experiments demonstrate that our proposed evaluation method significantly outperforms the state-of-the-art learning-based evaluation methods, with an average correlation improvement of 13.18%. In addition, we have publicly released the codes of our proposed method and state-of-the-art baselines.
In end-to-end dialogue modeling and agent learning, it is important to (1) effectively learn knowledge from data, and (2) fully utilize heterogeneous information, e.g., dialogue act flow and utterances. However, the majority of existing methods cannot simultaneously satisfy the two conditions. For example, rule definition and data labeling during system design take too much manual work, and sequence-to-sequence methods only model one-side utterance information. In this paper, we propose a novel joint end-to-end model by multi-task representation learning, which can capture the knowledge from heterogeneous information through automatically learning knowledgeable low-dimensional embeddings from data, named with DialogAct2Vec. The model requires little manual work for intervention in system design and we find that the multi-task learning can greatly improve the effectiveness of representation learning. Extensive experiments on a public dataset for restaurant reservation show that the proposed method leads to significant improvements against the state-of-the-art baselines on both the act prediction task and utterance prediction task.
Multilingual pretrained language models (such as multilingual BERT) have achieved impressive results for cross-lingual transfer. However, due to the constant model capacity, multilingual pre-training usually lags behind the monolingual competitors. In this work, we present two approaches to improve zero-shot cross-lingual classification, by transferring the knowledge from monolingual pretrained models to multilingual ones. Experimental results on two cross-lingual classification benchmarks show that our methods outperform vanilla multilingual fine-tuning.
We introduce a new scientific named entity recognizer called SEPT, which stands for Span Extractor with Pre-trained Transformers. In recent papers, span extractors have been demonstrated to be a powerful model compared with sequence labeling models. However, we discover that with the development of pre-trained language models, the performance of span extractors appears to become similar to sequence labeling models. To keep the advantages of span representation, we modified the model by under-sampling to balance the positive and negative samples and reduce the search space. Furthermore, we simplify the origin network architecture to combine the span extractor with BERT. Experiments demonstrate that even simplified architecture achieves the same performance and SEPT achieves a new state of the art result in scientific named entity recognition even without relation information involved.