知识图谱(KG)在人工智能应用中发挥着重要作用。然而现有工作忽略了事实中的条件信息，限制了传统KG的表达能力。因此，条件知识图谱(CKG)被提出，CKG可以有效地表示条件信息，进一步加强知识图谱的表达能力。但现有CKG工作只侧重于从文本中提取条件知识，而较少关注对提取出的条件知识的管理。为有效管理CKG,该文提出CondGraph,一个可以支持从存储到查询整个CKG管理过程的系统。CondGraph可以将任何形式的用于表示条件知识图谱的嵌套三元组转换为规范形式，并将其存储在分层树状数据结构中。此外，该文提出了CKG上带有条件约束的查询定义并设计和实现了查询算法，以支持高效的CKG查询。实验结果表明，与现有的图数据库相比，CondGraph将CKG查询的性能平均提高了1～3个数量级。
In recent years,graph neural networks have shown great potential in knowledge graph link prediction,es-pecially progressive relational graph neural networks.This paper proposed an Entity Context-Aware Progressive Re-lational Graph Attention Network to better understand relational context during reasoning by using Relational Con-text Generation Network and Relational Context Aware Network,and fully exploits the entity's dependent context information through a buffer layer mechanism.Experiments on multiple knowledge graph datasets show excellent performance of the proposed method in both transductive and inductive link prediction tasks.
地址实体识别旨在基于给定的地址文本识别出地址相关的实体，并判断每个地址实体的类别。现有模型未考虑地址的多义性和歧义性，这会引发实体边界问题和类型问题。因此，该文提出了一种基于多层知识感知的地址实体识别方法。首先，编码层通过知识嵌入的方法将句子知识融入BERT Transformer的前馈神经网络中，得到融合序列和句子知识的联合特征表示，缓解了类型问题对模型性能的影响；其次，通过词汇知识融合层将词汇知识与联合特征表示融合，增强了地址实体边界的表示能力。此外，通过语义相似度计算和注意力机制避免了引入知识可能带来的噪声问题。在CCKS2021地址要素解析数据集上的实验结果表明，该文模型优于多个基线模型，验证了方法的有效性。
To alleviate the class imbalance issue in unlabeled data for zero-shot text classification task and fully utilize the reasoning ability of pre-trained language models,this paper proposes a Knowledge Enhanced Zero-shot Text Classification(KE0TC)method.This method uses prompt templates to guide large pre-trained language model to extend category labels and builds a knowledge graph,using the graph structure for denoising and self-supervised data generation.Paragraph sampling and other methods are used to map the extracted labeled data into the parameter space of the classifier,thus achieving modeling of the classification space without collecting unlabeled training data.Compared with four baseline models on three text classification datasets,the proposed method achieves higher classification performance with lower time-consuming and less corpus.
Zero-shot stance detection(ZSSD)aims to detecting the stance of previously unseen targets during the inference stage.It is generally believed that the stance expression in a sentence is closely related to the stance target and topics discussed.We divide stance expressions of speakers into two categories:target-invariant and target-specific categories.Target-invariant stance expressions carry the same stance polarity regardless of the targets they are associated with.On the contrary,target-specific stance expressions only co-occur with certain targets.In this paper,we develop an effective approach to distinguish the types of target-related stance expressions to better learn transferable stance features.To be specific,inspired by self-supervised learning,we frame the stance-feature-type identification as a pretext task in ZSSD.We apply prompt learning to predict changing relationship between stance polarity labels and topic information in pretext task.This essentially allows the model to learn transferable stance features.Experiments on two benchmark datasets show that the proposed method obtains an improved performance than the baseline in ZSSD.
This work presents the task of text polishing, which generates a sentence that is more graceful than the input sentence while retaining its semantic meaning. Text polishing has great value in real usage and is an important component in modern writing assistance systems. However, the task is still not well studied in the literature. Further research in this important direction requires more formal task definitions, benchmark datasets, and powerful baseline models. In this work, we formulate the task as a context-dependent text generation problem and conduct a case study on the text polishing with Chinese idiom. To circumvent the difficulties of task data annotation, we propose a semi-automatic data construction pipeline based on human-machine collaboration, and establish a large-scale text polishing dataset consisting of 1.5 million instances. We propose two types of task-specific pre-training objectives for the text polishing task and implement a series of Transformer based models pre-trained on a massive Chinese corpus as baselines. We conduct extensive experiments with the baseline models on the constructed text polishing datasets and have some major findings. The human evaluation further reveals the polishing ability of the final system.
Text classification is a critical and foundational task in Tibetan natural language processing, it plays a crucial role in various applications, such as sentiment analysis and information extraction. However, the limited availability of annotated data poses a significant challenge to Tibetan natural language processing. This paper proposes a prompt learning-based method for low-resource Tibetan text classification to overcome this challenge. This method utilizes pre-trained language models to learn text representation and generation capabilities on a large-scale unsupervised Tibetan corpus, enabling few-shot Tibetan text classification. Experimental results demonstrate that the proposed method significantly improves the performance of Tibetan text classification in low-resource scenarios. This work provides a new research idea and method for low-resource language processing, such as Tibetan natural language processing. Hopefully, it will inspire subsequent work on low-resource language processing.
The objective of Entity Alignment (EA) is to identify equivalent entity pairs from multiple Knowledge Graphs (KGs) and create a more comprehensive and unified KG. The majority of EA methods have primarily focused on the structural modality of KGs, lacking exploration of multi-modal information. A few multi-modal EA methods have made good attempts in this field. Still, they have two shortcomings: (1) inconsistent and inefficient modality modeling that designs complex and distinct models for each modality; (2) ineffective modality fusion due to the heterogeneous nature of modalities in EA. To tackle these challenges, we propose PathFusion, consisting of two main components: (1) MSP, a unified modeling approach that simplifies the alignment process by constructing paths connecting entities and modality nodes to represent multiple modalities; (2) IRF, an iterative fusion method that effectively combines information from different modalities using the path as an information carrier. Experimental results on real-world datasets demonstrate the superiority of PathFusion over state-of-the-art methods, with 22.4%-28.9% absolute improvement on Hits@1, and 0.194-0.245 absolute improvement on MRR.
The unbalanced distribution of category labels and the correlation between these labels tend to cause over-learning issues in deep learning models. In fine-grained sentiment analysis datasets, the correlation between category labels and the heterogeneity of tag distribution are prominent. In the deep learning model, we use the adjusted circle-loss to introduce margin and gradient attenuation in the loss function to handle the challenges caused by unbalanced label distribution and non-independence between labels. This method can be well combined with pre-trained models and adapt to various learning models and algorithms. Compared with the state-of-the-art typical models, our loss function mechanism achieves significant improvement using SemEval18 and GoeEmotions by measure of Jaccard coefficient, micro-F1, and macro-F1. It implies that our solution could work efficiently for sentiment analysis and sentiment analysis tasks.
Electronic medical record(EMR)is an important source of medical information with rich medical knowl-edge.In light of the I2B2 entity and relationship classification,we establish the classification system of entity and entity relationship of the diabetes EMR,as well as the annotation scheme.With multiple rounds of manual proof-reading,the Diabetes Electronic Medical Record entity and relation Corpus(DEMRC)is finally completed.DEMRC contains 8899 entities,456 entity modifications and 16564 relationships.The consistency of entity and relationship annotation has reached 0.86 and 0.94,respectively.For the entity identification and entity relationship extraction tasks,The BILSTM-CRF model based on transfer learning and the Roberta model are trained on the corpus for pre-liminary experiments and various entities and relationships in the corpus are evaluated,which lay a foundation for the follow-up research on the Entity identification and relation extraction of diabetes EMR and the construction of the diabetes knowledge graph.
Entity boundary prediction is essential for Chinese named entity recognition.Most of multi-task learning methods consider only employing the word segmentation task.This paper presents a new Chinese named entity rec-ognition method based on the multi-task label consistency mechanism.The method integrates word segmentation and part-of-speech information into the named entity recognition model to establish a multi-task learning mode based on the label consistency mechanism,It enhances the boundary information learning by capturing label consistency relationships under the multi-task framework.The method is vilidated by the full sample experiment,simulated small sample experiment and real small sample experiment,resulting 10.28%,11.17%and 8.84%improvements o-ver the baseline model,respectively.
Humor plays an important role in human communication and is abundant in sitcoms.Punchline is one of a form to achieve humorous effects in sitcoms.The existing punchlines recognition methods only recognize the punch-line by modeling the contextual semantic relationship.In contrast,this paper proposes a new method based on multi-task learning model.First,we regard the transfer relationship between two tags as a manifestation of inconsistency in humor theory,and we use the conditional random field to learn this transfer relationship.Secondly,learning the transfer relationship between adjacent tags and the contextual semantic relationship can both capture the inconsisten-cy between the setup and punchline,and we introduce the multi-task learning method to learn the meaning of each sentence,the meaning of all the characters that make up each sentence,the label transfer relationship at the word level and the label transfer relationship at the sentence level.Experiments on the English data set of CCL2020"Mav-ericks Cup"humorous calculation-sitcom punchlines recognition and evaluation task.show that the proposed method is 3.2%higher than the current best method,achieving the best effect on the punchlines recognition task.
Knowledge graph question generation task is to generate related questions from a given knowledge graph.In recent years,knowledge graph problem generation models mainly use RNN or Transformer to encode the knowl-edge graph subgraph,ignoring the explicit graph structure.We propose an Iterative Message Passing graph encoder to encode subgraph and capture the explicit structured information of subgraph.In addition,we use the Sliding-Win-dow Hierarchical Attention mechanism to learn the importance of the local information of the subgraph to nodes.Experiments on WebQuestions and PathQuestions datasets indicate that our model outperforms KTG model by 2.16 and 15.44 respectively,in terms of BLEU-4 metric,which verify the effectiveness of proposed model.
Testset for text proofreading evaluation is the basis of spell checking research,including traditional and standard text proofreading testset.The traditional testset for text proofreading is obtained by artificially forging the correct data through subjective experience.The standard testset for text proofreading is obtained from the real data-set with strong reliability.Based on the analysis of the construction methods of English and Chinese text proofread-ing testsets,combined with the characteristics of Tibetan language,this paper studies the testset construction for Tibetan text proofreading,and completes a standard text proofreading testset with statistical analysis of the types and distribution of errors.The validity and usability of the testset are verified.
从非结构化文本中提取关系三元组对于大规模知识图谱的构建至关重要.目前,大部分研究集中于从书面文本中抽取实体关系,从对话中抽取实体关系的研究还很少.和书面文本中的实体关系相比,对话中的实体关系更强调"人"的关系且更口语化.为此,该文提出了一种使用GCN(图卷积神经网络)建模对话情景的对话实体关系识别方法.该方法根据多人对话的特点,将对话句子看作节点,根据句子距离为句子间分配有权重的边,从而构建出一张对话情景图,然后使用 GCN来建模对话之间的关系.在 DialogRE 数据集上的实验证明,该文方法优于本文研究同时期性能最好的模型.
2023年6月30日-7月2日,由中国中文信息学会情感计算专委会主办,西安交通大学承办的第三届中国情感计算大会(The Third Chinese Conference on Affective Computing,CCAC 2023)暨中国中文信息学会情感计算专委会工作会议在陕西省西安市举行.来自全国五十多所高校、科研机构和企业的近四百位代表齐聚西安,共同探讨情感计算领域的最新进展和发展方向.大会由西安交通大学管晓宏院士、哈尔滨工业大学李生教授担任大会荣誉主席,清华大学陶建华教授、西安交通大学龚怡宏教授担任大会主席.山西大学王素格教授、复旦大学魏忠钰副教授担任程序委员会主席.
近年来,提示学习方法由于可以充分激发预训练语言模型的潜能而得到了研究者越来越多的关注,特别是在知识抽取任务中取得了较好进展.为了提升提示学习性能,研究者也开展了基于知识的提示学习模板工程、答案工程优化等多项研究.该文对提示学习与知识相结合的相关研究进行了系统综述,包括知识抽取中的提示学习方法以及基于知识约束的提示学习相关进展.在此基础上,该文还探讨了目前方法存在的局限性,展望了提示学习与知识相结合的发展趋势.
多文档机器阅读理解是利用计算机同时将多个文档的语义信息进行阅读理解、筛选糅合后进行问题作答的过程.与传统的单文档机器阅读理解相比,其更关注问题约束下的文档间语义逻辑关系理解,是自然语言处理领域中新兴的研究方向之一.该文首先选取了多文档机器阅读理解的重点数据集,涵盖多源信息型数据集、线性推理型数据集和细粒度半结构化型数据集等以分析任务发展趋势;其次调研了深度学习领域中主流的评价指标和研究方法,划分了具有代表性的四层阅读架构并分别展开了详细的讨论,特别在阅读理解层中选取了文档选择、Transformer语言模型、图卷积神经网络和外部知识融合等阅读方法以分析任务的研究现状;最后对多文档阅读理解的研究进行了总结与展望,提出了后续研究面临的问题与挑战.
对话推荐旨在通过与用户对话来获取用户偏好并向其推荐高质量的商品,现有的对话推荐系统大多忽略了用户记忆中的潜在兴趣,导致难以在短时间内准确获取用户偏好.针对这一问题,该文提出了基于用户记忆的对话推荐模型,用户记忆包括用户的历史行为序列和评论、对话记录.首先,通过图神经网络学习评论和对话记录中用户、商品和属性之间的关系信息,保证系统能够提出与用户偏好最相关的问题来尽快了解用户当前需求;其次,利用改进的Transformer建模用户多类型行为序列来学习用户潜在兴趣;最后,与学习到的关系信息融合来做出推荐.在包含多个领域的对话数据集上的实验结果表明,该文提出的模型既能获得更高的推荐准确性又能以更少的对话次数成功推荐商品.
2023年7月6-8日,由中国中文信息学会、未来数据论坛指导委员会主办,北京理工大学、鹏城实验室、奇安信共同承办的第十一届网络空间智慧搜索暨未来数据高峰论坛暨BCS2023北京网络安全大会分论坛在北京召开.
