Graph Convolutional Network (GCN) has achieved extraordinary success in learning effective task-specific representations of nodes graphs. However, regarding Heterogeneous Information (HIN), existing HIN-oriented GCN methods still suffer from two deficiencies: (1) they cannot flexibly explore all possible meta-paths and extract the most useful ones for a target object, which hinders both effectiveness interpretability; (2) often need to generate intermediate meta-path based dense graphs,...

To alleviate data sparsity and cold-start problems of traditional recommender systems (RSs), incorporating knowledge graphs (KGs) to supplement auxiliary information has attracted considerable attention recently. However, simply integrating KGs in current KG-based RS models is not necessarily a guarantee improve the recommendation performance, which may even weaken holistic model capability. This because construction these independent collection historical user-item interactions; hence,...

Session-based recommendation aims to predict a user's next action based on previous actions in the current session. The major challenge is capture authentic and complete user preferences entire Recent work utilizes graph structure represent session adopts Graph Neural Network (GNN) encode information. This modeling choice has been proved be effective achieved remarkable results. However, most of existing studies only consider each item within independently do not semantics from high-level...

Graph Neural Networks (GNNs) have become mainstream methods for solving the semi-supervised node classification problem. However, due to uneven location distribution of labeled nodes in graph, are only accessible a small portion unlabeled nodes, leading under-reaching issue. In this study, we firstly reveal by conducting an empirical investigation on various well-known graphs. Then, demonstrate that results unsatisfactory alignment between and through systematic experimental analysis,...

Learning text representation is crucial for classification and other language related tasks. There are a diverse set of networks in the literature, how to find optimal one non-trivial problem. Recently, emerging Neural Architecture Search (NAS) techniques have demonstrated good potential solve Nevertheless, most existing works NAS focus on search algorithms pay little attention space. In this paper, we argue that space also an important human prior success different applications. Thus,...

BERT is a cutting-edge language representation model pre-trained by large corpus, which achieves superior performances on various natural understanding tasks. However, major blocking issue of applying to online services that it memory-intensive and leads unsatisfactory latency user requests, raising the necessity compression. Existing solutions leverage knowledge distillation framework learn smaller imitates behaviors BERT. training procedure expensive itself as requires sufficient data...

Jiangang Bai, Yujing Wang, Yiren Chen, Yaming Yang, Jing Yu, Yunhai Tong. Proceedings of the 16th Conference European Chapter Association for Computational Linguistics: Main Volume. 2021.

Recent Weak Supervision (WS) approaches have had widespread success in easing the bottleneck of labeling training data for machine learning by synthesizing labels from multiple potentially noisy supervision sources. However, proper measurement and analysis these remain a challenge. First, datasets used existing works are often private and/or custom, limiting standardization. Second, WS with same name base vary terms weak sources used, significant "hidden" source evaluation variance. Finally,...

Qingfeng Sun, Yujing Wang, Can Xu, Kai Zheng, Yaming Yang, Huang Hu, Fei Jessica Zhang, Xiubo Geng, Daxin Jiang. Proceedings of the 60th Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2022.

Jiazhan Feng, Qingfeng Sun, Can Xu, Pu Zhao, Yaming Yang, Chongyang Tao, Dongyan Qingwei Lin. Proceedings of the 61st Annual Meeting Association for Computational Linguistics (Volume 1: Long Papers). 2023.

Data-rich documents are ubiquitous in various applications, yet they often rely solely on textual descriptions to convey data insights. Prior research primarily focused providing visualization-centric augmentation data-rich documents. However, few have explored using automatically generated word-scale visualizations enhance the document-centric reading process. As an exploratory step, we propose GistVis, automatic pipeline that extracts and visualizes insight from text descriptions. GistVis...

Graph Neural Networks (GNNs) have shown advantages in various graph-based applications. Most existing GNNs assume strong homophily of graph structure and apply permutation-invariant local aggregation neighbors to learn a representation for each node. However, they fail generalize heterophilic graphs, where most neighboring nodes different labels or features, the relevant are distant. Few recent studies attempt address this problem by combining multiple hops hidden representations central...

Recent self-supervised pre-training methods on Heterogeneous Information Networks (HINs) have shown promising competitiveness over traditional semi-supervised Graph Neural (HGNNs). Unfortunately, their performance heavily depends careful customization of various strategies for generating high-quality positive examples and negative examples, which notably limits flexibility generalization ability. In this work, we present SHGP, a novel Self-supervised Pre-training approach, does not need to...

In this paper, a theoretical scheme for bidirectional quantum teleportation by using four-qubit GHZ state and two Bell states as channel is proposed. scheme, arbitrary single-qubit an unknown three-qubit can be mutually transmitted between the communication participants Alice Bob. The information if performs four measurement operations on her qubits Bob carries out joint his qubits, respectively. Then both reconstruct target means of appropriate unitary operations. we take comparisons with...

BERT is a cutting-edge language representation model pre-trained by large corpus, which achieves superior performances on various natural understanding tasks. However, major blocking issue of applying to online services that it memory-intensive and leads unsatisfactory latency user requests, raising the necessity compression. Existing solutions leverage knowledge distillation framework learn smaller imitates behaviors BERT. training procedure expensive itself as requires sufficient data...

There has been a surge of researchers' interest in building predictive models over graphs. However, the overwhelming complexity graph space often makes it challenging to extract interpretable and discriminative structural features for classification. In this work, we propose new neural network model called Substructure Assembling Network (SAN) learn representations The key innovation is unified Unit (SAU), which variant Recurrent Neural (RNN) designed hierarchically assemble useful pieces...

Multi-view Comprehensive Representation Learning (MCRL) aims to synthesize information from multiple views learn comprehensive representations of data items. Prevalent deep MCRL methods typically concatenate synergistic view-specific or average aligned in the fusion stage. However, performance inevitably degenerate even fail when partial are missing real-world applications; based usually cannot fully exploit complementarity multi-view data. To eliminate all these drawbacks, this work we...

Abstract The discrimination of seafloor substrate type is an extremely important part science, and the information great significance for development marine science protection environment. Current sonar equipment can efficiently generate images present visually, so classification technology based on has become a hot research topic. Convolutional neural network, as one most algorithms in seabed image classification, excellent performance cases. However, size convolutional kernel network...

The reservoir brittleness index can characterize the relative of shale oil and gas reservoirs, which provides guidance for hydraulic fracturing in later stage exploration development. It is a significant to evaluate sweet spot reservoirs. Most research on involves laboratory measurements petrophysical quantitative analysis, but few studies directly predict from prestack seismic data. Markov chain Monte Carlo (MCMC) probabilistic algorithm incorporating delayed rejection adaptive Metropolis...

As the main revenue source of Internet companies, online advertising is always a significant topic, where click-through rate (CTR) prediction plays central role. In systems, there are often many advertisement products. Due to competition in bidding mechanism, some products may get lots data train CTR model while lack high-quality data. However, predict accurate CTR, large amount needed. Therefore, transfer knowledge from product (source) small (target) necessary. We propose learning method...

Transformer is a ubiquitous model for natural language processing and has attracted wide attentions in computer vision. The attention maps are indispensable transformer to encode the dependencies among input tokens. However, they learned independently each layer sometimes fail capture precise patterns. In this paper, we propose novel generic mechanism based on evolving improve performance of transformers. On one hand, different layers share common knowledge, thus ones preceding can instruct...

Stance Detection Task (SDT) aims at identifying the stance of sentence towards a specific target and is usually modeled as classification problem. Backgound knowledge often necessary for detection with respect to target, especially when there no explicitly mentioned in text. This paper focuses on stimulation low-resource tasks. We firstly explore formalize prompt based contrastive learning task. At same time, make suit detection, we design template mechanism incorporate corresponding into...