In the scenario of unsupervised extractive summarization, learning high-quality sentence representations is essential to select salient sentences from the input document. Previous studies focus more on employing statistical approaches or pre-trained language models (PLMs) to extract sentence embeddings, while ignoring the rich information inherent in the heterogeneous types of interaction between words and sentences. In this paper, we are the first to propose an unsupervised extractive summarizaiton method with heterogeneous graph embeddings (HGEs) for Chinese document. A heterogeneous text graph is constructed to capture different granularities of interactions by incorporating graph structural information. Moreover, our proposed graph is general and flexible where additional nodes such as keywords can be easily integrated. Experimental results demonstrate that our method consistently outperforms the strong baseline in three summarization datasets.
The pre-trained conversational models still fail to capture the implicit commonsense (CS) knowledge hidden in the dialogue interaction, even though they were pre-trained with an enormous dataset. In order to build a dialogue agent with CS capability, we firstly inject external knowledge into a pre-trained conversational model to establish basic commonsense through efficient Adapter tuning (Section 4). Secondly, we propose the ``two-way learning'' method to enable the bidirectional relationship between CS knowledge and sentence pairs so that the model can generate a sentence given the CS triplets, also generate the underlying CS knowledge given a sentence (Section 5). Finally, we leverage this integrated CS capability to improve open-domain dialogue response generation so that the dialogue agent is capable of understanding the CS knowledge hidden in dialogue history on top of inferring related other knowledge to further guide response generation (Section 6). The experiment results demonstrate that CS\_Adapter fusion helps DialoGPT to be able to generate series of CS knowledge. And the DialoGPT+CS\_Adapter response model adapted from CommonGen training can generate underlying CS triplets that fits better to dialogue context.
Scene segmentation and classification (SSC) serve as a critical step towards the field of video structuring analysis. Intuitively, jointly learning of these two tasks can promote each other by sharing common information. However, scene segmentation concerns more on the local difference between adjacent shots while classification needs the global representation of scene segments, which probably leads to the model dominated by one of the two tasks in the training phase. In this paper, from an alternate perspective to overcome the above challenges, we unite these two tasks into one task by a new form of predicting shots link: a link connects two adjacent shots, indicating that they belong to the same scene or category. To the end, we propose a general One Stage Multimodal Sequential Link Framework (OS-MSL) to both distinguish and leverage the two-fold semantics by reforming the two learning tasks into a unified one. Furthermore, we tailor a specific module called DiffCorrNet to explicitly extract the information of differences and correlations among shots. Extensive experiments on a brand-new large scale dataset collected from real-world applications, and MovieScenes are conducted. Both the results demonstrate the effectiveness of our proposed method against strong baselines.
The extraction of text information in videos serves as a critical step towards semantic understanding of videos. It usually involved in two steps: (1) text recognition and (2) text classification. To localize texts in videos, we can resort to large numbers of text recognition methods based on OCR technology. However, to our knowledge, there is no existing work focused on the second step of video text classification, which will limit the guidance to downstream tasks such as video indexing and browsing. In this paper, we are the first to address this new task of video text classification by fusing multimodal information to deal with the challenging scenario where different types of video texts may be confused with various colors, unknown fonts and complex layouts. In addition, we tailor a specific module called CorrelationNet to reinforce feature representation by explicitly extracting layout information. Furthermore, contrastive learning is utilized to explore inherent connections between samples using plentiful unlabeled videos. Finally, we construct a new well-defined industrial dataset from the news domain, called TI-News, which is dedicated to building and evaluating video text recognition and classification applications. Extensive experiments on TI-News demonstrate the effectiveness of our method.
Massive open online courses (MOOCs), which provide a large-scale interactive participation and open access via the web, are becoming a modish way for online and distance education. To help users have a better study experience, many MOOC platforms have provided the services of recommending courses to users. However, we argue that directly recommending a course to users will ignore the expertise levels of different users. To fill this gap, this paper studies the problem of concept recommendation in a more fine-grained view. We propose a novel Heterogeneous Information Networks based Concept Recommender with Reinforcement Learning (HinCRec-RL) incorporated for concept recommendation in MOOCs. Specifically, we first formulate the concept recommendation in MOOCs as a reinforcement learning problem to better model the dynamic interaction among users and knowledge concepts. In addition, to mitigate the data sparsity issue which also exists in many other recommendation tasks, we consider a heterogeneous information network (HIN) among users, courses, videos and concepts, to better learn the semantic representation of users. In particular, we use the meta-paths on HIN to guide the propagation of users' preferences and propose a heterogeneous graph attention network to represent the meta-paths. To validate the effectiveness of our proposed approach, we conduct comprehensive experiments on a real-world dataset from XuetangX, a popular MOOC platform from China. The promising results show that our proposed approach can outperform other baselines.
Defense models against adversarial attacks have grown significantly, but the lack of practical evaluation methods has hindered progress. Evaluation can be defined as looking for defense models' lower bound of robustness given a budget number of iterations and a test dataset. A practical evaluation method should be convenient (i.e., parameter-free), efficient (i.e., fewer iterations) and reliable (i.e., approaching the lower bound of robustness). Towards this target, we propose a parameter-free Adaptive Auto Attack (A$^3$) evaluation method which addresses the efficiency and reliability in a test-time-training fashion. Specifically, by observing that adversarial examples to a specific defense model follow some regularities in their starting points, we design an Adaptive Direction Initialization strategy to speed up the evaluation. Furthermore, to approach the lower bound of robustness under the budget number of iterations, we propose an online statistics-based discarding strategy that automatically identifies and abandons hard-to-attack images. Extensive experiments demonstrate the effectiveness of our A$^3$. Particularly, we apply A$^3$ to nearly 50 widely-used defense models. By consuming much fewer iterations than existing methods, i.e., $1/10$ on average (10$\times$ speed up), we achieve lower robust accuracy in all cases. Notably, we won $\textbf{first place}$ out of 1681 teams in CVPR 2021 White-box Adversarial Attacks on Defense Models competitions with this method. Code is available at: $\href{https://github.com/liuye6666/adaptive_auto_attack}{https://github.com/liuye6666/adaptive\_auto\_attack}$
Finding relevant moments and highlights in videos according to natural language queries is a natural and highly valuable common need in the current video content explosion era. Nevertheless, jointly conducting moment retrieval and highlight detection is an emerging research topic, even though its component problems and some related tasks have already been studied for a while. In this paper, we present the first unified framework, named Unified Multi-modal Transformers (UMT), capable of realizing such joint optimization while can also be easily degenerated for solving individual problems. As far as we are aware, this is the first scheme to integrate multi-modal (visual-audio) learning for either joint optimization or the individual moment retrieval task, and tackles moment retrieval as a keypoint detection problem using a novel query generator and query decoder. Extensive comparisons with existing methods and ablation studies on QVHighlights, Charades-STA, YouTube Highlights, and TVSum datasets demonstrate the effectiveness, superiority, and flexibility of the proposed method under various settings. Source code and pre-trained models are available at https://github.com/TencentARC/UMT.
Few-shot table-to-text generation is a task of composing fluent and faithful sentences to convey table content using limited data. Despite many efforts having been made towards generating impressive fluent sentences by fine-tuning powerful pre-trained language models, the faithfulness of generated content still needs to be improved. To this end, this paper proposes a novel approach Attend, Memorize and Generate (called AMG), inspired by the text generation process of humans. In particular, AMG (1) attends over the multi-granularity of context using a novel strategy based on table slot level and traditional token-by-token level attention to exploit both the table structure and natural linguistic information; (2) dynamically memorizes the table slot allocation states; and (3) generates faithful sentences according to both the context and memory allocation states. Comprehensive experiments with human evaluation on three domains (i.e., humans, songs, and books) of the Wiki dataset show that our model can generate higher qualified texts when compared with several state-of-the-art baselines, in both fluency and faithfulness.
This work combines information about the dialogue history encoded by pre-trained model with a meaning representation of the current system utterance to realize contextual language generation in task-oriented dialogues. We utilize the pre-trained multi-context ConveRT model for context representation in a model trained from scratch; and leverage the immediate preceding user utterance for context generation in a model adapted from the pre-trained GPT-2. Both experiments with the MultiWOZ dataset show that contextual information encoded by pre-trained model improves the performance of response generation both in automatic metrics and human evaluation. Our presented contextual generator enables higher variety of generated responses that fit better to the ongoing dialogue. Analysing the context size shows that longer context does not automatically lead to better performance, but the immediate preceding user utterance plays an essential role for contextual generation. In addition, we also propose a re-ranker for the GPT-based generation model. The experiments show that the response selected by the re-ranker has a significant improvement on automatic metrics.