| 书目名称 | Knowledge Science, Engineering and Management | | 副标题 | 14th International C | | 编辑 | Han Qiu,Cheng Zhang,Sun-Yuan Kung | | 视频video | | | 丛书名称 | Lecture Notes in Computer Science | | 图书封面 |  | | 描述 | This three-volume set constitutes the refereed proceedings of the 14th International Conference on Knowledge Science, Engineering and Management, KSEM 2021, held in Tokyo, Japan, in August 2021..The 164 revised full papers were carefully reviewed and selected from 492 submissions. The contributions are organized in the following topical sections: knowledge science with learning and AI; knowledge engineering research and applications; knowledge management with optimization and security.. | | 出版日期 | Conference proceedings 2021 | | 关键词 | artificial intelligence; computational linguistics; computer networks; computer science; computer system | | 版次 | 1 | | doi | https://doi.org/10.1007/978-3-030-82147-0 | | isbn_softcover | 978-3-030-82146-3 | | isbn_ebook | 978-3-030-82147-0Series ISSN 0302-9743 Series E-ISSN 1611-3349 | | issn_series | 0302-9743 | | copyright | Springer Nature Switzerland AG 2021 |
| 1 |
Front Matter |
|
|
Abstract
|
| 2 |
|
|
|
Abstract
|
| 3 |
A Semantic Textual Similarity Calculation Model Based on Pre-training Model |
Zhaoyun Ding,Kai Liu,Wenhao Wang,Bin Liu |
|
Abstract
As a basic research topic in natural language processing, the calculation of text similarity is widely used in the fields of plagiarism checker and sentence search. The traditional calculation of text similarity constructed text vectors only based on TF-IDF, and used the cosine of the angle between vectors to measure the similarity between two texts. However, this method cannot solve the similar text detection task with different text representation but similar semantic representation. In response to the above-mentioned problems, we proposed the pre-training of text based on the ERNIE semantic model of PaddleHub, and constructed similar text detection into a classification problem; in view of the problem that most of the similar texts in the data set led to the imbalance of categories in the training set, an oversampling method for confusion sampling, OSConfusion, was proposed. The experimental results showed that the method proposed in this paper was able to solve the problem of paper comparison well, and could identify the repetitive paragraphs with different text representations. And the ERNIE-SIM with OSConfusion was better than the ERNIE-SIM without OSConfusion in the predicti
|
| 4 |
Representation Learning of Knowledge Graph with Semantic Vectors |
Tianyu Gao,Yuanming Zhang,Mengni Li,Jiawei Lu,Zhenbo Cheng,Gang Xiao |
|
Abstract
Knowledge graph (.) is a structured semantic knowledge base, which is widely used in the fields of semantic search, such as intelligent Q&A and intelligent recommendation. Representation learning, as a key issue of ., aims to vectorize entities and relations in . to reduce data sparseness and improve computational efficiency. Translation-based representation learning model shows great knowledge representation ability, but there also are limitations in complex relations modeling and representation accuracy. To address these problems, this paper proposes a novel representation learning model with semantic vectors, called TransV, which makes full use of external text corpus and .’s context to accurately represent entities and complex relations. Entity semantic vectors and relation semantic vectors are constructed, which can not only deeply extend semantic structure of ., but also transform complex relations into precise simple relations from a semantic perspective. Link prediction and triple classification tasks are performed on TransV with public datasets. Experimental results show that TransV can outperform other translation-based models. . is reduced by 66 and . is increased by 20%
|
| 5 |
Chinese Relation Extraction with Flat-Lattice Encoding and Pretrain-Transfer Strategy |
Xiuyue Zeng,Jiang Zhong,Chen Wang,Cong Hu |
|
Abstract
Relation Extraction (RE) aims to assign a correct relation class holding between entity pairs in context. However, many existing methods suffer from segmentation errors, especially for Chinese RE. In this paper, an improved lattice encoding is introduced. Our structure is a variant of the flat-lattice Transformer. The lattice framework can combine character-level and word-level information to avoid segmentation errors. We optimize the position encoding scheme to embrace the relative distance between spans and target entities. Moreover, to reduce the classification errors between positive instances and negative instances in our model, we propose a pretrain-transfer strategy. Specifically, the main idea is to migrate the classification ability of the binary classifier to multi-class identification. Experiments on SanWen, FinRE, and ACE-2005 corpora demonstrate that our methods are effective and outperform other relation extraction models in performance.
|
| 6 |
English Cloze Test Based on BERT |
Minjie Ding,Mingang Chen,Wenjie Chen,Lizhi Cai |
|
Abstract
Cloze test is a common test in language examinations. It is also a research direction of natural language processing, which is an important field of artificial intelligence. In general, some words in a complete article are hidden, and several candidates are given to let the student choose the correct hidden word. To explore whether machine can do cloze test, we have done some research to build down-stream tasks of BERT for cloze test. In this paper, we consider the compound words in articles and make an improvement to help the model handling these kind of words. The experimental results show that our model performs well on questions of compound words and has better accuracy on CLOTH dataset.
|
| 7 |
An Automatic Method for Understanding Political Polarization Through Social Media |
Yihong Zhang,Masumi Shirakawa,Takahiro Hara |
|
Abstract
Understanding political polarization is an important problem when one studies the culture of a democratic country. As a platform for discussing social issues, social media such as Twitter contains rich information about political polarization. In this paper, we propose an automatic method for discovering information from social media that can help people understand political polarization of the country. Previous researches have answered the “who” question, as they proposed methods for identifying ideal points of social media users. In our work, we make a step forward by answering the “what” question. Our method consists of two main techniques, namely, ideal point estimation and discriminative natural language processing. The inputs of our method are raw social media data, and the outputs are representative phrases for different political sides. Using real-world data from Twitter, we also verify that the representative phrases our method generates are consistent with our general knowledge of political polarization in Japan.
|
| 8 |
An Improved Convolutional Neural Network Based on Noise Layer |
Zhaoyang Wang,Shaowei Pan |
|
Abstract
In order to solve the over-fitting problem in Convolutional Neural Networks (CNN), a new method to improve the performance of CNN with noise layer on the basis of the previous studies has been proposed. This method improves the generalization performance of the CNN model by adding corresponding noise to the feature image obtained after convolution operation. The constructed noise layer can be flexibly embedded in a certain position of the CNN structure, and with each iteration of training, the added noise is also constantly changing, which makes the interference to the CNN model more profound, thus the more essential features of the input image are obtained. The experimental results show that the improved CNN model based on the noise layer has better recognition effect on some test images than the CNN model without any improvement; for different CNN models, the position of the noise layer which can improve the recognition accuracy is different; as the number of layers deepens, to improve the generalization performance of the CNN model, the position of the noise layer needs to be moved back. The improved CNN model based on the noise layer proposed in this paper solves the overfittin
|
| 9 |
Syntactic Enhanced Projection Network for Few-Shot Chinese Event Extraction |
Linhui Feng,Linbo Qiao,Yi Han,Zhigang Kan,Yifu Gao,Dongsheng Li |
|
Abstract
Few-shot learning event extraction methods gain more and more attention due to their ability to handle new event types. Current few-shot learning studies mainly focus on English event detection, which suffering from error propagation due to the identify-then-classify paradigm. And these methods could not be applied to Chinese event extraction directly, because they suffer from the Chinese word-trigger mismatch problem. In this work, we explore the Chinese event extraction with limited labeled data and reformulate it as a few-shot sequence tagging task. To this end, we propose a novel and practical few-shot syntactic enhanced projection network (SEPN), which exploits a syntactic learner to not only integrate the semantics of the characters and the words by Graph Convolution Networks, but also make the extracted feature more discriminative through a cross attention mechanism. Differing from prototypical networks which may lead to poor performance due to the prototype of each class could be closely distributed in the embedding space, SEPN learns to project embedding to space where different labels are well-separated. Furthermore, we deliberately construct an adaptive max-margin loss t
|
| 10 |
A Framework of Data Augmentation While Active Learning for Chinese Named Entity Recognition |
Qingqing Li,Zhen Huang,Yong Dou,Ziwen Zhang |
|
Abstract
Named entity recognition (NER) is a basic task to construct knowledge graph. The training performance is limited with few labelled data. One solution is active learning, which can achieve ideal results by multi-round sampling strategy to augment unlabelled data. However, there is very few labelled data in the early rounds, which leads to slow improvement on training performance. We thus propose a framework of data augmentation while active learning. To validate our claims, we focus on Chinese NER task and carry out extensive experiments on two public datasets. Experimental results show that our framework is effective for a series of classical query strategy. We can achieve 99% of the best deep model trained on full data using only 22% of the data on Resume, 63% labelled data is reduced as compared to pure active learning (PAL).
|
| 11 |
Traffic Route Planning in Partially Observable Environment Using Actions Group Representation |
Minzhong Luo,Shan Yu |
|
Abstract
We investigate the problem of optimal route planning formulized as Partially Observable Markov Decision Process (POMDP) [.]: Given a partially traffic-aware road network, we aim to find a route for agent vehicle such that the global travel time cost is minimized. In this paper, we show that the theory of group representation with its ability to make mechanism of . (actions acting on states) computable efficiently, which is able to provide significant advantages in multi-step planning with information partially observable. Using the action group Representation, we build a more “visionary” system. Extensive experiments offer insight into the efficiency of proposed algorithms.
|
| 12 |
Bayesian Belief Network Model Using Sematic Concept for Expert Finding |
Wei Zheng,Hongxu Hou,Nier Wu,Shuo Sun |
|
Abstract
The Expert finding is a research hotspot in the area of entity retrieval. However, due to the small number of search terms, the retrieval effect will be poor due to the mechanical text matching. In view of the above shortcomings, we use Bayesian belief network as a model frame, and two expert finding models are proposed. One is a basic semantic belief network retrieval model, in which BERT and LDA models are used, and the other is a compound semantic belief network model. The compound model uses an effective data fusion technique to integrate the retrieval results of the two sub-models in this paper. The paper presents the topology and retrieval algorithm of two models proposed. The experiments verify the validity of the research content on Amine platform. Experimental results show that the semantic model can improve the MAP value, and the compound semantic model is better than the existing expert finding model on multiple evaluation indicators such as P@N, MAP and MRR, and it can improve the performance of expert retrieval.
|
| 13 |
Towards Solving the Winograd Schema Challenge: Model-Free, Model-Based and a Spectrum in Between |
Weinan He,Zhanhao Xiao |
|
Abstract
The Winograd Schema Challenge (WSC) has attracted much attention recently as . is recognized to be not only the key to human-level intelligence but also a bottleneck faced by recent progress. Although neural language models (LMs) have achieved state-of-the-art (SOTA) performance on WSC, they fall short on interpretability and robustness against adversarial attacks. Contrarily, methods with structured representation and explicit reasoning suffer from the difficulty of knowledge acquisition and the rigidness of representation. In this paper, we look back on the current model-free and model-based approaches, pointing out the missing ingredients towards solving the WSC. We report our preliminary exploration of formalizing the WSC problems using a variant of first-order language and our first-hand findings of indispensable capabilities of human-level commonsense reasoning. The issues we encounter suggest that a full spectrum of representation tools and reasoning abilities are called for.
|
| 14 |
The Novel Efficient Transformer for NLP |
Benjamin Mensa-Bonsu,Tao Cai,Tresor Y. Koffi,Dejiao Niu |
|
Abstract
Reducing the numerical precision of weights and activations of deep neural networks have proven to be a stunningly efficient way of deploying deep networks on edge devices with limited resources. With the advent of the Transformer model, several quantization techniques have been proposed to reduce the computation and model size. However, these existing quantization techniques use fixed bit-width assignments, which result in a significant degradation in the accuracy of the model. We present in this work an efficient Transformer based on our novel multi-layer quantization technique, which reduces the precision of data based on the characteristics of weights and activations in each layer of the Transformer architecture while at the same time preserving the model’s structure. The WMT2014 DE-EN and WMT2014 FR-EN datasets are used to evaluate. The results show that our efficient Transformer achieves 4x compression with improved accuracy and an overall reduction in the training time overhead. By comparing with existing state-of-the-art techniques, we further proved that with a minimum of 3-bit and a maximum of 8-bit quantization, comparable state-of-the-art BLEU scores can be obtained.
|
| 15 |
Joint Entity and Relation Extraction for Long Text |
Dong Cheng,Hui Song,Xianglong He,Bo Xu |
|
Abstract
Extracting relation triplets from unstructured text has been well studied in recent years. However, previous works focus on solving the relation overlapping problem, and few of them deal with the long text relation extraction. In this work, we introduce a novel end-to-end joint entity and relation extraction model, namely, LTRel, which is capable of extracting relation triplets from long text based on a cross-sentence relation classification algorithm. On the other hand, due to the importance of entity recognition to the entire end-to-end model, we refine the entity tagging scheme and the feature representation of TPLinker, which save the memory space and computation, and also improve the accuracy. We evaluate our model on two public datasets: the English dataset NYT and the Chinese dataset DuIE2.0 proposed by Baidu, both of which are better than state-of-the-art on F1 score, especially significant on the Chinese dataset with a higher proportion of long text samples.
|
| 16 |
Evaluating Dataset Creation Heuristics for Concept Detection in Web Pages Using BERT |
Michael Paris,Robert Jäschke |
|
Abstract
Dataset creation for the purpose of training natural language processing (NLP) algorithms is often accompanied by an uncertainty about how the target concept is represented in the data. Extracting such data from web pages and verifying its quality is a non-trivial task, due to the Web’s unstructured and heterogeneous nature and the cost of annotation. In that situation, annotation heuristics can be employed to create a dataset that captures the target concept, but in turn may lead to an unstable downstream performance. On the one hand, a trade-off exists between cost, quality, and magnitude for annotation heuristics in tasks such as classification, leading to fluctuations in trained models’ performance. On the other hand, general-purpose NLP tools like BERT are now commonly used to benchmark new models on a range of tasks on static datasets. We utilize this standardization as a means to assess dataset quality, as most applications are dataset specific. In this study, we investigate and evaluate the performance of three annotation heuristics for a classification task on extracted web data using BERT. We present multiple datasets, from which the classifier shall learn to identify web
|
| 17 |
Panoptic-DLA: Document Layout Analysis of Historical Newspapers Based on Proposal-Free Panoptic Segm |
Min Lu,Feilong Bao,Guanglai Gao |
|
Abstract
In this paper, we introduce a novel historical newspaper layout analysis model named Panoptic-DLA. Different from the previous works regarding layout analysis as a separate object detection or semantic segmentation problem, we define the layout analysis task as the proposal-free panoptic segmentation to assign a unique value to each pixel in the document image, encoding both semantic label and instance id. The model consists of two branches: the semantic segmentation branch and the instance segmentation branch. Firstly, the pixels are separated to “things” and “stuff” by semantic classification taking the background as “stuff”, and content objects such as images, paragraphs, etc., as “things”. Then the predicted “things” are grouped further to their instance ids by instance segmentation. The semantic segmentation branch adopted DeepLabV3+ to predict pixel-wise class labels. In order to split adjacent regions well, the instance segmentation branch produce a mountain-like soft score-map and a center-direction map to represent content objects. The method is trained and tested on a dataset of historical European newspapers with complex content layout. The experiment shows that the prop
|
| 18 |
Improving Answer Type Classification Quality Through Combined Question Answering Datasets |
Aleksandr Perevalov,Andreas Both |
|
Abstract
Understanding what a person is asking via a question is one of the first steps that humans use to find the corresponding answer. The same is true for Question Answering (QA) systems. Hence, the quality of the expected answer type classifier (EAT) has a direct influence on QA quality. Many research papers are aiming at improving short text classification quality, however, there is a lack of focus on the impact of training data characteristics on the classification quality as well as effective reuse of datasets through their augmentation and combination. In this work, we propose an approach of analyzing and improving the EAT classification quality via a combination of existing QA datasets. We provide 4 new question classification datasets based on several well-known QA datasets as well as the approach to unify its class taxonomy. We made a sufficient amount of experiments to demonstrate several valuable insights related to the impact of training data characteristics on the classification quality. Additionally, an embedding-based approach for automatic data labeling error detection is demonstrated.
|
| 19 |
FOBA: Flight Operation Behavior Analysis Based on Hierarchical Encoding |
Tongyu Zhu,Zhiwei Tong |
|
Abstract
Through analyzing flight data we can detect pilot’s improper operation, which effectively improves flight safety. This paper proposes an approach to convert multivariate flight data into symbol series and an auto-regressive semantic understanding model. Our model can predict what kind of pilot operation or aircraft status should appear at the next time step according to data at the current time step. Furthermore, we proposed a prediction model for unsafe event predicting based on our semantic understanding model. The experiment results show that our prediction model outperforms well known classifiers. Finally, experiments show that our model has the application value of correcting pilot operation.
|
| 20 |
An Event Detection Method Combining Temporal Dimension and Position Dimension |
Zehao Yu |
|
Abstract
Increasing numbers of tweets streaming data present both challenges and opportunities to improve event detection approach. So, it is important to propose a method that can solve some challenges. One such challenge is to automatically detect a set of events from large texts dynamically. The unique features of tweets, such as short and noisy content, diverse and fast changing topics, and large data volume, make event detection a challenge. Many previous works on event detection focused on supervised methods that did not take temporal information of the text and the position information of the words into account simultaneously. In this paper, we propose an unsupervised approach for event detection from tweets or texts that incorporates information from all positions of a word’s occurrences into a biased PageRank and the temporal information into the tweets or texts. Our proposed model obtains remarkable improvements in performance over three event detection methods called Joint Model, Globe Vector- Latent Dirichlet Allocation, and Language Independent Neural Network that do not take into account word positions and temporal information for this task. Specifically, on three datasets of
|
书目名称Knowledge Science, Engineering and Management影响因子(影响力)
书目名称Knowledge Science, Engineering and Management影响因子(影响力)学科排名
书目名称Knowledge Science, Engineering and Management网络公开度
书目名称Knowledge Science, Engineering and Management网络公开度学科排名
书目名称Knowledge Science, Engineering and Management被引频次
书目名称Knowledge Science, Engineering and Management被引频次学科排名
书目名称Knowledge Science, Engineering and Management年度引用
书目名称Knowledge Science, Engineering and Management年度引用学科排名
书目名称Knowledge Science, Engineering and Management读者反馈
书目名称Knowledge Science, Engineering and Management读者反馈学科排名
|
|
|
发表于 2025-3-21 16:09:56
