Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660590
M. Hamedani, Sang-Wook Kim
The link prediction task has attracted significant attention in the literature. Link-based similarity measures (in short, similarity measures) are the conventional methods for this task, while recently graph embedding methods (in short, embedding methods) are widely employed as well. In this paper, we extensively investigate the effectiveness of embedding methods and similarity measures (i.e., both non-recursive and recursive ones) in link prediction. Our experimental results with three real-world datasets demonstrate that 1) recursive similarity measures are not beneficial in this task than non-recursive one,2) increasing the number of dimensions in vectors may not help improve the accuracy of embedding methods, and 3) in comparison with embedding methods, Adamic/Adar, a non-recursive similarity measure, can be a useful method for link prediction since it shows promising results while being parameter-free.
{"title":"Embedding Methods or Link-based Similarity Measures, Which is Better for Link Prediction?","authors":"M. Hamedani, Sang-Wook Kim","doi":"10.1109/IC-NIDC54101.2021.9660590","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660590","url":null,"abstract":"The link prediction task has attracted significant attention in the literature. Link-based similarity measures (in short, similarity measures) are the conventional methods for this task, while recently graph embedding methods (in short, embedding methods) are widely employed as well. In this paper, we extensively investigate the effectiveness of embedding methods and similarity measures (i.e., both non-recursive and recursive ones) in link prediction. Our experimental results with three real-world datasets demonstrate that 1) recursive similarity measures are not beneficial in this task than non-recursive one,2) increasing the number of dimensions in vectors may not help improve the accuracy of embedding methods, and 3) in comparison with embedding methods, Adamic/Adar, a non-recursive similarity measure, can be a useful method for link prediction since it shows promising results while being parameter-free.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"44 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116917350","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660448
Bo-cheng Yu, Xin Zhang
The increasing construction of 5G dense network creates the conditions for the application of Massive MIMO system. However, with the continuous expansion of business requirements, users put forward higher requirements for the number of antennas in MIMO system. With the increase of the number of antennas, the cost of traditional MIMO beamforming algorithm for channel detection and feedback will increase rapidly, which consumes more wire-less resources and greatly increases the computational burden of the system. The use of computer vision aids provides convenience for the beamforming method to track the target accurately under LOS condition. Combined with image tracking algorithm, the position of the target in each image frame can be calculated so that the angle information of LOS path and the best beam-forming scheme can be determined directly, which can reduce the cost and calculation of the system through wireless resource measurement and feed-back. As a result, the operation speed and accuracy of the system are improved. In this paper, a beamforming method based on binocular positioning is studied. Compared with the traditional method, this method can reduce the number of codeword searches and improve the channel capacity in high-density 5G network.
{"title":"A Small Range Ergodic Beamforming Method Based on Binocular Vision Positioning","authors":"Bo-cheng Yu, Xin Zhang","doi":"10.1109/IC-NIDC54101.2021.9660448","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660448","url":null,"abstract":"The increasing construction of 5G dense network creates the conditions for the application of Massive MIMO system. However, with the continuous expansion of business requirements, users put forward higher requirements for the number of antennas in MIMO system. With the increase of the number of antennas, the cost of traditional MIMO beamforming algorithm for channel detection and feedback will increase rapidly, which consumes more wire-less resources and greatly increases the computational burden of the system. The use of computer vision aids provides convenience for the beamforming method to track the target accurately under LOS condition. Combined with image tracking algorithm, the position of the target in each image frame can be calculated so that the angle information of LOS path and the best beam-forming scheme can be determined directly, which can reduce the cost and calculation of the system through wireless resource measurement and feed-back. As a result, the operation speed and accuracy of the system are improved. In this paper, a beamforming method based on binocular positioning is studied. Compared with the traditional method, this method can reduce the number of codeword searches and improve the channel capacity in high-density 5G network.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"14 5-6 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116471315","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Although the academia has done a lot of research on DNS abnormal behavior, whether from the perspective of traffic or irregular domain name recognition, the mechanism behind DNS is ignored in the pre-processing of DNS logs and other data. In addition, most studies focus on traffic anomaly detection and unconventional domain name recognition, and lack of systematic research on the combination of the two, so the proposed algorithm has no practical application. This paper proposes a clustering method based on DNS client IP address traffic characteristics, which divides DNS logs into five access modes. Then, a DNS log preprocessing algorithm is designed to preprocess the logs that may exist in zombie hosts. Finally, a two-layer GRU network detection algorithm based on domain name text features is proposed. Experimental results show that this method can effectively identify zombie hosts in DNS logs.
{"title":"Zombie Hosts Identification Based on DNS Log","authors":"Renjie Wang, Yangsen Zhang, Ruixue Duan, Zhuofan Huang","doi":"10.1109/IC-NIDC54101.2021.9660578","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660578","url":null,"abstract":"Although the academia has done a lot of research on DNS abnormal behavior, whether from the perspective of traffic or irregular domain name recognition, the mechanism behind DNS is ignored in the pre-processing of DNS logs and other data. In addition, most studies focus on traffic anomaly detection and unconventional domain name recognition, and lack of systematic research on the combination of the two, so the proposed algorithm has no practical application. This paper proposes a clustering method based on DNS client IP address traffic characteristics, which divides DNS logs into five access modes. Then, a DNS log preprocessing algorithm is designed to preprocess the logs that may exist in zombie hosts. Finally, a two-layer GRU network detection algorithm based on domain name text features is proposed. Experimental results show that this method can effectively identify zombie hosts in DNS logs.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123986820","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660574
Hyunmok Park, Kijung Yoon
Graph neural network (GNN) is a general framework for using deep neural networks on graph data. The defining feature of a GNN is that it uses a form of neural message passing where vector messages are exchanged between nodes and updated using neural networks. The message passing operation that underlies GNNs has recently been applied to develop neural approximate inference algorithms, but little work has been done on understanding under what conditions GNNs can be used as a core module for building general inference models. To study this question, we consider the task of out-of-distribution generalization where training and test data have different distributions, by systematically investigating how the graph size and structural properties affect the inferential performance of GNNs. We find that (1) the average unique node degree is one of the key features in predicting whether GNNs can generalize to unseen graphs; (2) the graph size is not a fundamental limiting factor of the generalization in GNNs when the average node degree remains invariant across training and test distributions; (3) despite the size-invariant generalization, training GNNs on graphs of high degree (and of large size consequently) is not trivial (4) neural inference by GNNs outperforms algorithmic inferences especially when the pairwise potentials are strong, which naturally makes the inference problem harder.
{"title":"Degree Matters: Assessing the Generalization of Graph Neural Network","authors":"Hyunmok Park, Kijung Yoon","doi":"10.1109/IC-NIDC54101.2021.9660574","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660574","url":null,"abstract":"Graph neural network (GNN) is a general framework for using deep neural networks on graph data. The defining feature of a GNN is that it uses a form of neural message passing where vector messages are exchanged between nodes and updated using neural networks. The message passing operation that underlies GNNs has recently been applied to develop neural approximate inference algorithms, but little work has been done on understanding under what conditions GNNs can be used as a core module for building general inference models. To study this question, we consider the task of out-of-distribution generalization where training and test data have different distributions, by systematically investigating how the graph size and structural properties affect the inferential performance of GNNs. We find that (1) the average unique node degree is one of the key features in predicting whether GNNs can generalize to unseen graphs; (2) the graph size is not a fundamental limiting factor of the generalization in GNNs when the average node degree remains invariant across training and test distributions; (3) despite the size-invariant generalization, training GNNs on graphs of high degree (and of large size consequently) is not trivial (4) neural inference by GNNs outperforms algorithmic inferences especially when the pairwise potentials are strong, which naturally makes the inference problem harder.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"63 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126771654","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660407
Sheng Zhuang, Lin Cao, Zongmin Zhao, Dongfeng Wang
The method based on the fusion of radar and video in this paper is oriented to detecting surrounding objects while driving. This is usually a method of improving robustness and accuracy by using several senses, which makes sensor fusion a key part of the perception system. We propose a new fusion method called CT-EPNP, which uses radar and camera data for fast detection. Adding a central fusion algorithm on the basis of EPNP, and use the truncated cone method to compensate the radar information on the associated image when mapping. CT-EPNP returns to the object attributes depth, rotation, speed and other attributes. Based on this, simulation verification and related derivation of mathematical formulas are proved. We combined the improved algorithm with the RetinaNet model to ensure that the model is satisfied with the normal driving of the vehicle while gaining a certain increase in the detection rate. We have also made a certain improvement in ensuring repeated detection without using any additional time information.
{"title":"A Target Detection Method Based on the Fusion Algorithm of Radar and Camera","authors":"Sheng Zhuang, Lin Cao, Zongmin Zhao, Dongfeng Wang","doi":"10.1109/IC-NIDC54101.2021.9660407","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660407","url":null,"abstract":"The method based on the fusion of radar and video in this paper is oriented to detecting surrounding objects while driving. This is usually a method of improving robustness and accuracy by using several senses, which makes sensor fusion a key part of the perception system. We propose a new fusion method called CT-EPNP, which uses radar and camera data for fast detection. Adding a central fusion algorithm on the basis of EPNP, and use the truncated cone method to compensate the radar information on the associated image when mapping. CT-EPNP returns to the object attributes depth, rotation, speed and other attributes. Based on this, simulation verification and related derivation of mathematical formulas are proved. We combined the improved algorithm with the RetinaNet model to ensure that the model is satisfied with the normal driving of the vehicle while gaining a certain increase in the detection rate. We have also made a certain improvement in ensuring repeated detection without using any additional time information.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129564232","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660508
Conghui Ma, Bin Zhang, Lutao Yan, Haiyuan Li
This paper presents a data-driven optimal approach based on differential dynamic programming (DDP) for two-person differential game of nonlinear affine systems. Using test data, the Hamilton-Jacobi-Isaacs (HJI) equation is expanded into a set of high-order differential equations. Basis functions is adopted to approximate the unknown system function and value function. Based on the approximation, a data-driven optimal approach is proposed to obtain the unknown coefficients of the basis functions. A numerical example is proposed to demonstrate the effectiveness of this method.
{"title":"Data-driven Differential Games for Affine Nonlinear Systems","authors":"Conghui Ma, Bin Zhang, Lutao Yan, Haiyuan Li","doi":"10.1109/IC-NIDC54101.2021.9660508","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660508","url":null,"abstract":"This paper presents a data-driven optimal approach based on differential dynamic programming (DDP) for two-person differential game of nonlinear affine systems. Using test data, the Hamilton-Jacobi-Isaacs (HJI) equation is expanded into a set of high-order differential equations. Basis functions is adopted to approximate the unknown system function and value function. Based on the approximation, a data-driven optimal approach is proposed to obtain the unknown coefficients of the basis functions. A numerical example is proposed to demonstrate the effectiveness of this method.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131926997","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660447
Wei Liu, Hanwen Xu, Cheng Fang, Lei Yang, Weidong Jiao
Aiming at the problem of adjusting the penalty term coefficient of feature enhancement in high-resolution synthetic aperture radar (SAR) imaging, a marginal estimation Bayes (MEB) algorithm is proposed, so that the prior features of the target can be fitted properly to improve the accuracy of image feature extraction. Firstly, the alternating direction method of multipliers (ADMM) convex optimization framework is modeled based on the echoed data, and least absolute shrinkage and selection operator (Lasso) model and sparse group Lasso (SG-Lasso) model are introduced, then the maximum marginal likelihood distribution of the regularization parameters is derived. Moreover, the Moreau Yoshida unadjusted Langevin algorithm (MYULA) is used to realize target posteriori sampling solution. Because the posterior distribution is difficult to solve, the gradient projection method is introduced to estimate the regularization parameters. Finally, auto-learning parameters are used to optimize the imaging. The proposed algorithm can not only estimate the parameters of a single regularization term, but also estimate the parameters of multiple regularization terms. Aiming at non-differentiable part in the prior, MYULA is adopted to calculate the subgradient of the non-differentiable posterior distribution. Therefore, the proposed algorithm is capable of auto-leaning parameters even regularization function is non-differentiable. In the experimental part, compared with the optimal value of manual debugging, the error between the proposed method and the optimal value is within 15%, and the effectiveness of the algorithm are verified by phase transition diagram (PTD).
针对高分辨率合成孔径雷达(SAR)成像中特征增强惩罚项系数的调整问题，提出了一种边缘估计贝叶斯(MEB)算法，使目标的先验特征得到适当拟合，提高图像特征提取的精度。首先，基于回波数据建立了交替方向乘子法(ADMM)凸优化框架模型，引入了最小绝对收缩和选择算子(Lasso)模型和稀疏群Lasso (SG-Lasso)模型，推导了正则化参数的最大边际似然分布;此外，采用Moreau - Yoshida unadjusted Langevin算法(MYULA)实现目标后验采样解。由于后验分布难以求解，引入梯度投影法估计正则化参数。最后，利用自学习参数对图像进行优化。该算法不仅可以估计单个正则化项的参数，还可以估计多个正则化项的参数。针对先验中不可微的部分，采用MYULA计算不可微后验分布的次梯度。因此，即使正则化函数不可微，该算法也能自动学习参数。在实验部分，与人工调试的最优值相比，所提方法与最优值的误差在15%以内，并通过相变图(PTD)验证了算法的有效性。
{"title":"Auto-Learning of Parameters for High Resolution Sparse Group Lasso SAR Imagery","authors":"Wei Liu, Hanwen Xu, Cheng Fang, Lei Yang, Weidong Jiao","doi":"10.1109/IC-NIDC54101.2021.9660447","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660447","url":null,"abstract":"Aiming at the problem of adjusting the penalty term coefficient of feature enhancement in high-resolution synthetic aperture radar (SAR) imaging, a marginal estimation Bayes (MEB) algorithm is proposed, so that the prior features of the target can be fitted properly to improve the accuracy of image feature extraction. Firstly, the alternating direction method of multipliers (ADMM) convex optimization framework is modeled based on the echoed data, and least absolute shrinkage and selection operator (Lasso) model and sparse group Lasso (SG-Lasso) model are introduced, then the maximum marginal likelihood distribution of the regularization parameters is derived. Moreover, the Moreau Yoshida unadjusted Langevin algorithm (MYULA) is used to realize target posteriori sampling solution. Because the posterior distribution is difficult to solve, the gradient projection method is introduced to estimate the regularization parameters. Finally, auto-learning parameters are used to optimize the imaging. The proposed algorithm can not only estimate the parameters of a single regularization term, but also estimate the parameters of multiple regularization terms. Aiming at non-differentiable part in the prior, MYULA is adopted to calculate the subgradient of the non-differentiable posterior distribution. Therefore, the proposed algorithm is capable of auto-leaning parameters even regularization function is non-differentiable. In the experimental part, compared with the optimal value of manual debugging, the error between the proposed method and the optimal value is within 15%, and the effectiveness of the algorithm are verified by phase transition diagram (PTD).","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"276 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134366693","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660599
Jaeuk Lee, Jiye G. Kim, Joon‐Hyuk Chang
Many voice cloning studies based on multi-speaker text-to-speech (TTS) have been conducted. Among the techniques of voice cloning, we focus on zero-shot voice cloning. The most important aspect of zero-shot voice cloning is which speaker embedding is used. In this study, two types of speaker embeddings are used. One is extracted from the mel spectrogram using a speaker encoder and the other is stored in an embedding dictionary, such as a vector quantized-variational autoencoder (VQ-VAE). To extract embedding from the embedding dictionary, an attention mechanism is applied, which we call attention- V AE (AT - V AE). By employing the embedding extracted by the speaker encoder as a query in the attention mechanism, the attention weights are calculated in the embedding dictionary. This mechanism allows the extraction of speaker embedding, which represents unseen speakers. In addition, training is applied to make our model robust to unseen speakers. Through the training stage, our system has developed further. The performance of the proposed method was validated in terms of various metrics, and it was demonstrated that the proposed method enables voice cloning without adaptation training.
许多基于多说话人文本到语音(TTS)的语音克隆研究已经开展。在语音克隆技术中，我们重点研究了零采样语音克隆技术。零射击语音克隆最重要的方面是使用哪一个扬声器嵌入。在本研究中，使用了两种类型的说话人嵌入。一种是使用扬声器编码器从mel频谱图中提取，另一种是存储在嵌入字典中，例如矢量量化变分自编码器(VQ-VAE)。为了从嵌入字典中提取嵌入，采用了一种注意机制，我们称之为注意- V AE (AT - V AE)。利用说话人编码器提取的嵌入作为注意机制的查询，在嵌入字典中计算注意权值。这种机制允许提取说话人嵌入，它代表看不见的说话人。此外，还进行了训练，使我们的模型对未见的说话者具有鲁棒性。经过培训阶段，我们的系统得到了进一步的发展。从多个指标对所提方法的性能进行了验证，结果表明所提方法无需自适应训练即可实现语音克隆。
{"title":"Zero-Shot Voice Cloning Using Variational Embedding with Attention Mechanism","authors":"Jaeuk Lee, Jiye G. Kim, Joon‐Hyuk Chang","doi":"10.1109/IC-NIDC54101.2021.9660599","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660599","url":null,"abstract":"Many voice cloning studies based on multi-speaker text-to-speech (TTS) have been conducted. Among the techniques of voice cloning, we focus on zero-shot voice cloning. The most important aspect of zero-shot voice cloning is which speaker embedding is used. In this study, two types of speaker embeddings are used. One is extracted from the mel spectrogram using a speaker encoder and the other is stored in an embedding dictionary, such as a vector quantized-variational autoencoder (VQ-VAE). To extract embedding from the embedding dictionary, an attention mechanism is applied, which we call attention- V AE (AT - V AE). By employing the embedding extracted by the speaker encoder as a query in the attention mechanism, the attention weights are calculated in the embedding dictionary. This mechanism allows the extraction of speaker embedding, which represents unseen speakers. In addition, training is applied to make our model robust to unseen speakers. Through the training stage, our system has developed further. The performance of the proposed method was validated in terms of various metrics, and it was demonstrated that the proposed method enables voice cloning without adaptation training.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114319105","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660465
Yaping Wang, Huiqin Shao, Zhen Li, Yan Zhu, Zhe Liu
Sequence to sequence (Seq2Seq) model together with pointer network (Ptr-Net) has recently show promising results in slot filling task, in the situation where only sentence-level annotations are available, while the model's prediction contains repetition of slot values. In this paper, we add a coverage mechanism to alleviate issues of repeating prediction in slot filling task. We use a coverage vector to record attention history, and then add to the computation of attention, which can force model to consider more about un-predicted slot values. Experiments show that the proposed model significantly improves slot value prediction F1 with 8.5% relative improvement compare to the baseline models on benchmark DSTC2 (Dialog State Tracking Challenge 2) datasets.
{"title":"Combined Coverage, Attention and Pointer Networks for Improving Slot Filling in Spoken Language Understanding","authors":"Yaping Wang, Huiqin Shao, Zhen Li, Yan Zhu, Zhe Liu","doi":"10.1109/IC-NIDC54101.2021.9660465","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660465","url":null,"abstract":"Sequence to sequence (Seq2Seq) model together with pointer network (Ptr-Net) has recently show promising results in slot filling task, in the situation where only sentence-level annotations are available, while the model's prediction contains repetition of slot values. In this paper, we add a coverage mechanism to alleviate issues of repeating prediction in slot filling task. We use a coverage vector to record attention history, and then add to the computation of attention, which can force model to consider more about un-predicted slot values. Experiments show that the proposed model significantly improves slot value prediction F1 with 8.5% relative improvement compare to the baseline models on benchmark DSTC2 (Dialog State Tracking Challenge 2) datasets.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"40 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132321008","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-11-17DOI: 10.1109/IC-NIDC54101.2021.9660453
Tong Liu, K. Yu, Lu Wang, Xuanyu Zhang, Xiaofei Wu
In online social medias, there is a large amount of clickbait using various tricks such as curious words and well-designed sentence structures, to attract users to click on hyperlinks for unknown benefits. Clickbait detection aims to detect these hyperlinks through automated algorithms. Most of the previous clickbait datasets are based on English online social media corpus. Detection models based on these datasets usually cannot be well generalized to Chinese social media scenarios. In this paper, we construct a WeChat based Chinese clickbait dataset, i.e., WCD. Based on the WCD, we conduct a detailed analysis of the clickbait features from three aspects: behavior features, headline text features, and content text features. Finally, we use popular methods for clickbait detection on our dataset. We also respectively propose a machine learning detection model using feature fusion and a deep learning detection model combining headline semantic and POS tag information, both of which achieve excellent detection performance. The results of clickbait analysis and detection show that the dataset we constructed is of high quality.
{"title":"WCD: A New Chinese Online Social Media Dataset for Clickbait Analysis and Detection","authors":"Tong Liu, K. Yu, Lu Wang, Xuanyu Zhang, Xiaofei Wu","doi":"10.1109/IC-NIDC54101.2021.9660453","DOIUrl":"https://doi.org/10.1109/IC-NIDC54101.2021.9660453","url":null,"abstract":"In online social medias, there is a large amount of clickbait using various tricks such as curious words and well-designed sentence structures, to attract users to click on hyperlinks for unknown benefits. Clickbait detection aims to detect these hyperlinks through automated algorithms. Most of the previous clickbait datasets are based on English online social media corpus. Detection models based on these datasets usually cannot be well generalized to Chinese social media scenarios. In this paper, we construct a WeChat based Chinese clickbait dataset, i.e., WCD. Based on the WCD, we conduct a detailed analysis of the clickbait features from three aspects: behavior features, headline text features, and content text features. Finally, we use popular methods for clickbait detection on our dataset. We also respectively propose a machine learning detection model using feature fusion and a deep learning detection model combining headline semantic and POS tag information, both of which achieve excellent detection performance. The results of clickbait analysis and detection show that the dataset we constructed is of high quality.","PeriodicalId":264468,"journal":{"name":"2021 7th IEEE International Conference on Network Intelligence and Digital Content (IC-NIDC)","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-11-17","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114417224","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：