Ziyang Liu

Zhizhi Yu, Di Jin, Jianguo Wei et al. · mit

Graph Neural Networks (GNNs) have gained great popularity in tackling various analytical tasks on graph-structured data (i.e., networks). Typical GNNs and their variants follow a message-passing manner that obtains network representations by the feature propagation process along network topology, which however ignore the rich textual semantics (e.g., local word-sequence) that exist in many real-world networks. Existing methods for text-rich networks integrate textual semantics by mainly utilizing internal information such as topics or phrases/words, which often suffer from an inability to comprehensively mine the text semantics, limiting the reciprocal guidance between network structure and text semantics. To address these problems, we propose a novel text-rich graph neural network with external knowledge (TeKo), in order to take full advantage of both structural and textual information within text-rich networks. Specifically, we first present a flexible heterogeneous semantic network that incorporates high-quality entities and interactions among documents and entities. We then introduce two types of external knowledge, that is, structured triplets and unstructured entity description, to gain a deeper insight into textual semantics. We further design a reciprocal convolutional mechanism for the constructed heterogeneous semantic network, enabling network structure and textual semantics to collaboratively enhance each other and learn high-level network representations. Extensive experimental results on four public text-rich networks as well as a large-scale e-commerce searching dataset illustrate the superior performance of TeKo over state-of-the-art baselines.

Fengbin Zhu, Chao Wang, Wenqiang Lei et al.

Key Information Extraction (KIE) is aimed at extracting structured information (e.g. key-value pairs) from form-style documents (e.g. invoices), which makes an important step towards intelligent document understanding. Previous approaches generally tackle KIE by sequence tagging, which faces difficulty to process non-flatten sequences, especially for table-text mixed documents. These approaches also suffer from the trouble of pre-defining a fixed set of labels for each type of documents, as well as the label imbalance issue. In this work, we assume Optical Character Recognition (OCR) has been applied to input documents, and reformulate the KIE task as a region prediction problem in the two-dimensional (2D) space given a target field. Following this new setup, we develop a new KIE model named Region-based Document Understanding (RDU) that takes as input the text content and corresponding coordinates of a document, and tries to predict the result by localizing a bounding-box-like region. Our RDU first applies a layout-aware BERT equipped with a soft layout attention masking and bias mechanism to incorporate layout information into the representations. Then, a list of candidate regions is generated from the representations via a Region Proposal Module inspired by computer vision models widely applied for object detection. Finally, a Region Categorization Module and a Region Selection Module are adopted to judge whether a proposed region is valid and select the one with the largest probability from all proposed regions respectively. Experiments on four types of form-style documents show that our proposed method can achieve impressive results. In addition, our RDU model can be trained with different document types seamlessly, which is especially helpful over low-resource documents.

Fengbin Zhu, Ziyang Liu, Xiang Yao Ng et al.

Large Vision-Language Models (LVLMs) have achieved remarkable performance in many vision-language tasks, yet their capabilities in fine-grained visual understanding remain insufficiently evaluated. Existing benchmarks either contain limited fine-grained evaluation samples that are mixed with other data, or are confined to object-level assessments in natural images. To holistically assess LVLMs' fine-grained visual understanding capabilities, we propose using document images with multi-granularity and multi-modal information to supplement natural images. In this light, we construct MMDocBench, a benchmark with various OCR-free document understanding tasks for the evaluation of fine-grained visual perception and reasoning abilities. MMDocBench defines 15 main tasks with 4,338 QA pairs and 11,353 supporting regions, covering various document images such as research papers, receipts, financial reports, Wikipedia tables, charts, and infographics. Based on MMDocBench, we conduct extensive experiments using 13 open-source and 3 proprietary advanced LVLMs, assessing their strengths and weaknesses across different tasks and document image types. The benchmark, task instructions, and evaluation code will be made publicly available.

Ziyang Liu, Jingmeng Liu, Weihai Chen et al.

Semi-supervised video object segmentation (VOS) aims to segment a few moving objects in a video sequence, where these objects are specified by annotation of first frame. The optical flow has been considered in many existing semi-supervised VOS methods to improve the segmentation accuracy. However, the optical flow-based semi-supervised VOS methods cannot run in real time due to high complexity of optical flow estimation. A FAMINet, which consists of a feature extraction network (F), an appearance network (A), a motion network (M), and an integration network (I), is proposed in this study to address the abovementioned problem. The appearance network outputs an initial segmentation result based on static appearances of objects. The motion network estimates the optical flow via very few parameters, which are optimized rapidly by an online memorizing algorithm named relaxed steepest descent. The integration network refines the initial segmentation result using the optical flow. Extensive experiments demonstrate that the FAMINet outperforms other state-of-the-art semi-supervised VOS methods on the DAVIS and YouTube-VOS benchmarks, and it achieves a good trade-off between accuracy and efficiency. Our code is available at https://github.com/liuziyang123/FAMINet.

Ziyang Liu, Kevin Valencia, Justin Cui

While Text-To-Video (T2V) models have advanced rapidly, they continue to struggle with generating legible and coherent text within videos. In particular, existing models often fail to render correctly even short phrases or words and previous attempts to address this problem are computationally expensive and not suitable for video generation. In this work, we investigate a lightweight approach to improve T2V diffusion models using synthetic supervision. We first generate text-rich images using a text-to-image (T2I) diffusion model, then animate them into short videos using a text-agnostic image-to-video (I2v) model. These synthetic video-prompt pairs are used to fine-tune Wan2.1, a pre-trained T2V model, without any architectural changes. Our results show improvement in short-text legibility and temporal consistency with emerging structural priors for longer text. These findings suggest that curated synthetic data and weak supervision offer a practical path toward improving textual fidelity in T2V generation.

Ziyang Liu, Pengjunfei Chu, Shuming Dong et al.

In recent years, Multimodal Sentiment Analysis (MSA) has become a research hotspot that aims to utilize multimodal data for human sentiment understanding. Previous MSA studies have mainly focused on performing interaction and fusion on complete multimodal data, ignoring the problem of missing modalities in real-world applications due to occlusion, personal privacy constraints, and device malfunctions, resulting in low generalizability. To this end, we propose a Factorization-guided Semantic Recovery Framework (FSRF) to mitigate the modality missing problem in the MSA task. Specifically, we propose a de-redundant homo-heterogeneous factorization module that factorizes modality into modality-homogeneous, modality-heterogeneous, and noisy representations and design elaborate constraint paradigms for representation learning. Furthermore, we design a distribution-aligned self-distillation module that fully recovers the missing semantics by utilizing bidirectional knowledge transfer. Comprehensive experiments on two datasets indicate that FSRF has a significant performance advantage over previous methods with uncertain missing modalities.

Fengbin Zhu, Ziyang Liu, Fuli Feng et al.

In this work, we address question answering (QA) over a hybrid of tabular and textual data that are very common content on the Web (e.g. SEC filings), where discrete reasoning capabilities are often required. Recently, large language models (LLMs) like GPT-4 have demonstrated strong multi-step reasoning capabilities. We then consider harnessing the amazing power of LLMs to solve our task. We abstract a Step-wise Pipeline for tabular and textual QA, which consists of three key steps, including Extractor, Reasoner and Executor, and initially design an instruction to instantiate the pipeline and validate that GPT-4 outperforms all existing methods. However, utilizing an online LLM like GPT-4 holds various challenges in terms of cost, latency, and data security risk, which motivates us to specialize smaller LLMs in this task. We develop a TAT-LLM language model by fine-tuning LLaMA 2 with the training data generated automatically from existing expert-annotated datasets following the Step-wise Pipeline. The experimental results have verified that our TAT-LLM model can outperform all baseline models, including the previous best fine-tuned models and very large-scale LLMs like GPT-4 on FinQA, TAT-QA and TAT-DQA benchmarks.

Ziyang Liu, Zhengguo Li, Xingming Wu et al.

The generative adversarial network (GAN) is successfully applied to study the perceptual single image superresolution (SISR). However, the GAN often tends to generate images with high frequency details being inconsistent with the real ones. Inspired by conventional detail enhancement algorithms, we propose a novel prior knowledge, the detail prior, to assist the GAN in alleviating this problem and restoring more realistic details. The proposed method, named DSRGAN, includes a well designed detail extraction algorithm to capture the most important high frequency information from images. Then, two discriminators are utilized for supervision on image-domain and detail-domain restorations, respectively. The DSRGAN merges the restored detail into the final output via a detail enhancement manner. The special design of DSRGAN takes advantages from both the model-based conventional algorithm and the data-driven deep learning network. Experimental results demonstrate that the DSRGAN outperforms the state-of-the-art SISR methods on perceptual metrics and achieves comparable results in terms of fidelity metrics simultaneously. Following the DSRGAN, it is feasible to incorporate other conventional image processing algorithms into a deep learning network to form a model-based deep SISR.

Yilun Xu, Ziyang Liu, Xingming Wu et al.

Spatially varying exposure (SVE) is a promising choice for high-dynamic-range (HDR) imaging (HDRI). The SVE-based HDRI, which is called single-shot HDRI, is an efficient solution to avoid ghosting artifacts. However, it is very challenging to restore a full-resolution HDR image from a real-world image with SVE because: a) only one-third of pixels with varying exposures are captured by camera in a Bayer pattern, b) some of the captured pixels are over- and under-exposed. For the former challenge, a spatially varying convolution (SVC) is designed to process the Bayer images carried with varying exposures. For the latter one, an exposure-guidance method is proposed against the interference from over- and under-exposed pixels. Finally, a joint demosaicing and HDRI deep learning framework is formalized to include the two novel components and to realize an end-to-end single-shot HDRI. Experiments indicate that the proposed end-to-end framework avoids the problem of cumulative errors and surpasses the related state-of-the-art methods.

Ziyang Liu, Zhaomeng Cheng, Yunjiang Jiang et al.

Result relevance prediction is an essential task of e-commerce search engines to boost the utility of search engines and ensure smooth user experience. The last few years eyewitnessed a flurry of research on the use of Transformer-style models and deep text-match models to improve relevance. However, these two types of models ignored the inherent bipartite network structures that are ubiquitous in e-commerce search logs, making these models ineffective. We propose in this paper a novel Second-order Relevance, which is fundamentally different from the previous First-order Relevance, to improve result relevance prediction. We design, for the first time, an end-to-end First-and-Second-order Relevance prediction model for e-commerce item relevance. The model is augmented by the neighborhood structures of bipartite networks that are built using the information of user behavioral feedback, including clicks and purchases. To ensure that edges accurately encode relevance information, we introduce external knowledge generated from BERT to refine the network of user behaviors. This allows the new model to integrate information from neighboring items and queries, which are highly relevant to the focus query-item pair under consideration. Results of offline experiments showed that the new model significantly improved the prediction accuracy in terms of human relevance judgment. An ablation study showed that the First-and-Second-order model gained a 4.3% average gain over the First-order model. Results of an online A/B test revealed that the new model derived more commercial benefits compared to the base model.

Di Jin, Xiangchen Song, Zhizhi Yu et al.

Graph convolutional networks (GCNs), aiming to integrate high-order neighborhood information through stacked graph convolution layers, have demonstrated remarkable power in many network analysis tasks. However, topological limitations, including over-smoothing and local topology homophily, limit its capability to represent networks. Existing studies only perform feature convolution on network topology, which inevitably introduces unbalance between topology and features. Considering that in real world, the information network consists of not only the node-level citation information but also the local text-sequence information. We propose BiTe-GCN, a novel GCN architecture with bidirectional convolution of both topology and features on text-rich networks to solve these limitations. We first transform the original text-rich network into an augmented bi-typed heterogeneous network, capturing both the global node-level information and the local text-sequence information from texts. We then introduce discriminative convolution mechanisms to performs convolutions of both topology and features simultaneously. Extensive experiments on text-rich networks demonstrate that our new architecture outperforms state-of-the-art by a breakout improvement. Moreover, this architecture can also be applied to several e-commerce searching scenes such as JD searching. The experiments on the JD dataset validate the superiority of the proposed architecture over the related methods.

Yunjiang Jiang, Yue Shang, Ziyang Liu et al.

Relevance has significant impact on user experience and business profit for e-commerce search platform. In this work, we propose a data-driven framework for search relevance prediction, by distilling knowledge from BERT and related multi-layer Transformer teacher models into simple feed-forward networks with large amount of unlabeled data. The distillation process produces a student model that recovers more than 97\% test accuracy of teacher models on new queries, at a serving cost that's several magnitude lower (latency 150x lower than BERT-Base and 15x lower than the most efficient BERT variant, TinyBERT). The applications of temperature rescaling and teacher model stacking further boost model accuracy, without increasing the student model complexity. We present experimental results on both in-house e-commerce search relevance data as well as a public data set on sentiment analysis from the GLUE benchmark. The latter takes advantage of another related public data set of much larger scale, while disregarding its potentially noisy labels. Embedding analysis and case study on the in-house data further highlight the strength of the resulting model. By making the data processing and model training source code public, we hope the techniques presented here can help reduce energy consumption of the state of the art Transformer models and also level the playing field for small organizations lacking access to cutting edge machine learning hardwares.