远程监督实体关系抽取研究

【目的/意义】实体关系抽取是构建领域本体、知识图谱、开发问答系统的基础工作。远程监督方法将大规 模非结构化文本与已有的知识库实体对齐,自动标注训练样本,解决了有监督机器学习方法人工标注训练语料耗 时费力的问题,但也带来了数据噪声。【方法/过程】本文详细梳理了近些年远程监督结合深度学习技术,降低训练 样本噪声,提升实体关系抽取性能的方法。【结果/结论】卷积神经网络能更好的捕获句子局部、关键特征、长短时记 忆网络能更好的处理句子实体对远距离依赖关系,模型自动抽取句子词法、句法特征,注意力机制给予句子关键上 下文、单词更大的权重,在神经网络模型中融入先验知识能丰富句子实体对的语义信息,显著提升关系抽取性能。 【创新/局限】下一步的研究应考虑实体对重叠关系、实体对长尾语义关系的处理方法,更加全面的解决实体对关系 噪声问题。     

金融领域文本序列标注与实体关系联合抽取研究

【目的/意义】金融领域实体关系抽取是构造金融知识库的基础,对金融领域的文本信息利用具有重要作 用。本文提出金融领域实体关系联合抽取模型,增加了对金融文本复杂重叠关系的识别,可以有效避免传统的流 水线模型中识别错误在不同任务之间的传递。【方法/过程】本文构建了高质量金融文本语料,提出一种新的序列 标注模式和实体关系匹配规则,在预训练语言模型BERT（Bidirectional Encoder Representations from Transformers） 的基础上结合双向门控循环单元 BiGRU（Bidirectional Gated Recurrent Units）与条件随机场 CRF（Conditional Random Field）构建了端到端的序列标注模型,实现了实体关系的联合抽取。【结果/结论】针对金融领域文本数据 进行实验,实验结果表明本文提出的联合抽取模型在关系抽取以及重叠关系抽取上的F1值分别达到了0.627和 0.543,初步验证了中文语境下本文模型对金融领域实体关系抽取的有效性。【创新/局限】结合金融文本特征提出 了新的序列标注模式并构建了基于BERT的金融领域实体关系联合抽取模型,实现了对金融文本中实体间重叠关 系的识别。     

Planarized sentence representation for nested named entity recognition

One strategy to recognize nested entities is to enumerate overlapped entity spans for classification. However, current models independently verify every entity span, which ignores the semantic dependency between spans. In this paper, we first propose a planarized sentence representation to represent nested named entities. Then, a bi-directional two-dimensional recurrent operation is implemented to learn semantic dependencies between spans. Our method is evaluated on seven public datasets for named entity recognition. It achieves competitive performance in named entity recognition. The experimental results show that our method is effective to resolve nested named entities and learn semantic dependencies between them.     

Relevance-based entity selection for ad hoc retrieval

Recent developments have shown that entity-based models that rely on information from the knowledge graph can improve document retrieval performance. However, given the non-transitive nature of relatedness between entities on the knowledge graph, the use of semantic relatedness measures can lead to topic drift. To address this issue, we propose a relevance-based model for entity selection based on pseudo-relevance feedback, which is then used to systematically expand the input query leading to improved retrieval performance. We perform our experiments on the widely used TREC Web corpora and empirically show that our proposed approach to entity selection significantly improves ad hoc document retrieval compared to strong baselines. More concretely, the contributions of this work are as follows: (1) We introduce a graphical probability model that captures dependencies between entities within the query and documents. (2) We propose an unsupervised entity selection method based on the graphical model for query entity expansion and then for ad hoc retrieval. (3) We thoroughly evaluate our method and compare it with the state-of-the-art keyword and entity based retrieval methods. We demonstrate that the proposed retrieval model shows improved performance over all the other baselines on ClueWeb09B and ClueWeb12B, two widely used Web corpora, on the [email protected], and [email protected] metrics. We also show that the proposed method is most effective on the difficult queries. In addition, We compare our proposed entity selection with a state-of-the-art entity selection technique within the context of ad hoc retrieval using a basic query expansion method and illustrate that it provides more effective retrieval for all expansion weights and different number of expansion entities.     

AHAB: Aligning heterogeneous knowledge bases via iterative blocking

With the development of information extraction, there have been an increasing number of large-scale knowledge bases available in different domains. In recent years, a great deal of approaches have been proposed for large-scale knowledge base alignment. Most of them are based on iterative matching. If a pair of entities has been aligned, their compatible neighbors are selected as candidate entity pairs. The limitation of these methods is that they discover candidate entity pairs depending on aligned relations, which cannot be used for aligning heterogeneous knowledge bases. Only few existing methods focus on aligning heterogeneous knowledge bases, which discover candidate entity pairs just for once by traditional blocking methods. However, the performance of these methods depends on blocking keys heavily, which are hard to select. In this paper, we present an approach for aligning heterogeneous knowledge bases via iterative blocking (AHAB) to improve the discovery and refinement of candidate entity pairs. AHAB iteratively utilizes different relations for blocking, and then matches block pairs based on matched entity pairs. The Cartesian product of unmatched entities in matched block pairs forms candidate entity pairs. By filtering out dissimilar candidate entity pairs, matched entity pairs will be found. The number of matched entity pairs proliferates with iterations, which in turn helps match block pairs in each iteration. Experiments on real-world heterogeneous knowledge bases demonstrate that AHAB is able to yield a competitive performance.     

A domain-independent approach to finding related entities

We propose an approach to the retrieval of entities that have a specific relationship with the entity given in a query. Our research goal is to investigate whether related entity finding problem can be addressed by combining a measure of relatedness of candidate answer entities to the query, and likelihood that the candidate answer entity belongs to the target entity category specified in the query. An initial list of candidate entities, extracted from top ranked documents retrieved for the query, is refined using a number of statistical and linguistic methods. The proposed method extracts the category of the target entity from the query, identifies instances of this category as seed entities, and computes similarity between candidate and seed entities. The evaluation was conducted on the Related Entity Finding task of the Entity Track of TREC 2010, as well as the QA list questions from TREC 2005 and 2006. Evaluation results demonstrate that the proposed methods are effective in finding related entities.     

A Cross-Media Deep Relationship Classification Method Using Discrimination Information

Relation classification is one of the most fundamental tasks in the area of cross-media, which is essential for many practical applications such as information extraction, question&answer system, and knowledge base construction. In the cross-media semantic retrieval task, in order to meet the needs of cross-media uniform representation and semantic analysis, it is necessary to analyze the semantic potential relationship and construct semantic-related cross-media knowledge graph. The relationship classification technology is an important part of solving semantic correlation classification. Most of existing methods regard relation classification as a multi-classification task, without considering the correlation between different relationships. However, two relationships in the opposite directions are usually not independent of each other. Hence, this kind of relationships are easily confused in the traditional way. In order to solve the problem of confusing the relationships of the same semantic with different entity directions, this paper proposes a neural network fusing discrimination information for relation classification. In the proposed model, discrimination information is used to distinguish the relationship of the same semantic with different entity directions, the direction of entity in space is transformed into the direction of vector in mathematics by the method of entity vector subtraction, and the result of entity vector subtraction is used as discrimination information. The model consists of three modules: sentence representation module, relation discrimination module and discrimination fusion module. Moreover, two fusion methods are used for feature fusion. One is a Cascade-based feature fusion method, and another is a feature fusion method based on convolution neural network. In addition, this paper uses the new function added by cross-entropy function and deformed Max-Margin function as the loss function of the model. The experimental results show that the proposed discriminant feature is effective in distinguishing confusing relationships, and the proposed loss function can improve the performance of the model to a certain extent. Finally, the proposed model achieves 84.8% of the F1 value without any additional features or NLP analysis tools. Hence, the proposed method has a promising prospect of being incorporated in various cross-media systems.     

TSVFN: Two-Stage Visual Fusion Network for multimodal relation extraction

Multimodal relation extraction is a critical task in information extraction, aiming to predict the class of relations between head and tail entities from linguistic sequences and related images. However, the current works are vulnerable to less relevant visual objects detected from images and are not able to sufficiently fuse visual information into text pre-trained models. To overcome these problems, we propose a Two-Stage Visual Fusion Network (TSVFN) that employs the multimodal fusion approach in vision-enhanced entity relation extraction. In the first stage, we design multimodal graphs, whose novelty lies mainly in transforming the sequence learning into the graph learning. In the second stage, we merge the transformer-based visual representation into the text pre-trained model by a multi-scale cross-model projector. Specifically, two multimodal fusion operations are implemented inside the pre-trained model respectively. We finally accomplish deep interaction of multimodal multi-structured data in two fusion stages. Extensive experiments are conducted on a dataset (MNRE), our model outperforms the current state-of-the-art method by 1.76%, 1.52%, 1.29%, and 1.17% in terms of accuracy, precision, recall, and F1 score, respectively. Moreover, our model also achieves excellent results under the condition of fewer samples.     

An exploration of ranking models and feedback method for related entity finding

Most existing search engines focus on document retrieval. However, information needs are certainly not limited to finding relevant documents. Instead, a user may want to find relevant entities such as persons and organizations. In this paper, we study the problem of related entity finding. Our goal is to rank entities based on their relevance to a structured query, which specifies an input entity, the type of related entities and the relation between the input and related entities. We first discuss a general probabilistic framework, derive six possible retrieval models to rank the related entities, and then compare these models both analytically and empirically. To further improve performance, we study the problem of feedback in the context of related entity finding. Specifically, we propose a mixture model based feedback method that can utilize the pseudo feedback entities to estimate an enriched model for the relation between the input and related entities. Experimental results over two standard TREC collections show that the derived relation generation model combined with a relation feedback method performs better than other models.     

An end-to-end joint model for evidence information extraction from court record document

Information extraction is one of the important tasks in the field of Natural Language Processing (NLP). Most of the existing methods focus on general texts and little attention is paid to information extraction in specialized domains such as legal texts. This paper explores the task of information extraction in the legal field, which aims to extract evidence information from court record documents (CRDs). In the general domain, entities and relations are mostly words and phrases, indicating that they do not span multiple sentences. In contrast, evidence information in CRDs may span multiple sentences, while existing models cannot handle this situation. To address this issue, we first add a classification task in addition to the extraction task. We then formulate the two tasks as a multi-task learning problem and present a novel end-to-end model to jointly address the two tasks. The joint model adopts a shared encoder followed by separate decoders for the two tasks. The experimental results on the dataset show the effectiveness of the proposed model, which can obtain 72.36% F1 score, outperforming previous methods and strong baselines by a large margin.     

Multi-heterogeneous neighborhood-aware for Knowledge Graphs alignment

Entity alignment is an important task for the Knowledge Graph (KG) completion, which aims to identify the same entities in different KGs. Most of previous works only utilize the relation structures of KGs, but ignore the heterogeneity of relations and attributes of KGs. However, these information can provide more feature information and improve the accuracy of entity alignment. In this paper, we propose a novel Multi-Heterogeneous Neighborhood-Aware model (MHNA) for KGs alignment. MHNA aggregates multi-heterogeneous information of aligned entities, including the entity name, relations, attributes and attribute values. An important contribution is to design a variant attention mechanism, which adds the feature information of relations and attributes to the calculation of attention coefficients. Extensive experiments on three well-known benchmark datasets show that MHNA significantly outperforms 12 state-of-the-art approaches, demonstrating that our approach has good scalability and superiority in both cross-language and monolingual KGs. An ablation study further supports the effectiveness of our variant attention mechanism.     

Cascade embedding model for knowledge graph inference and retrieval

Knowledge graphs are widely used in retrieval systems, question answering systems (QA), hypothesis generation systems, etc. Representation learning provides a way to mine knowledge graphs to detect missing relations; and translation-based embedding models are a popular form of representation model. Shortcomings of translation-based models however, limits their practicability as knowledge completion algorithms. The proposed model helps to address some of these shortcomings.The similarity between graph structural features of two entities was found to be correlated to the relations of those entities. This correlation can help to solve the problem caused by unbalanced relations and reciprocal relations. We used Node2vec, a graph embedding algorithm, to represent information related to an entity's graph structure, and we introduce a cascade model to incorporate graph embedding with knowledge embedding into a unified framework. The cascade model first refines feature representation in the first two stages (Local Optimization Stage), and then uses backward propagation to optimize parameters of all the stages (Global Optimization Stage). This helps to enhance the knowledge representation of existing translation-based algorithms by taking into account both semantic features and graph features and fusing them to extract more useful information. Besides, different cascade structures are designed to find the optimal solution to the problem of knowledge inference and retrieval.The proposed model was verified using three mainstream knowledge graphs: WIN18, FB15K and BioChem. Experimental results were validated using the hit@10 rate entity prediction task. The proposed model performed better than TransE, giving an average improvement of 2.7% on WN18, 2.3% on FB15k and 28% on BioChem. Improvements were particularly marked where there were problems with unbalanced relations and reciprocal relations. Furthermore, the stepwise-cascade structure is proved to be more effective and significantly outperforms other baselines.     

关系抽取技术的发展与应用——以生物信息学为例

关系抽取是文本挖掘的一项重要研究内容,它能够反映命名实体之间的关系,有助于发现隐含在大量数据和文本中的知识。以生物信息学为例,重点论述了国内外关系抽取技术的研究进展、常用技术与方法及应用,并对未来关系抽取技术的发展进行了展望。     

Learning entity-centric document representations using an entity facet topic model

Learning semantic representations of documents is essential for various downstream applications, including text classification and information retrieval. Entities, as important sources of information, have been playing a crucial role in assisting latent representations of documents. In this work, we hypothesize that entities are not monolithic concepts; instead they have multiple aspects, and different documents may be discussing different aspects of a given entity. Given that, we argue that from an entity-centric point of view, a document related to multiple entities shall be (a) represented differently for different entities (multiple entity-centric representations), and (b) each entity-centric representation should reflect the specific aspects of the entity discussed in the document.In this work, we devise the following research questions: (1) Can we confirm that entities have multiple aspects, with different aspects reflected in different documents, (2) can we learn a representation of entity aspects from a collection of documents, and a representation of document based on the multiple entities and their aspects as reflected in the documents, (3) does this novel representation improves algorithm performance in downstream applications, and (4) what is a reasonable number of aspects per entity? To answer these questions we model each entity using multiple aspects (entity facets¹), where each entity facet is represented as a mixture of latent topics. Then, given a document associated with multiple entities, we assume multiple entity-centric representations, where each entity-centric representation is a mixture of entity facets for each entity. Finally, a novel graphical model, the Entity Facet Topic Model (EFTM), is proposed in order to learn entity-centric document representations, entity facets, and latent topics.Through experimentation we confirm that (1) entities are multi-faceted concepts which we can model and learn, (2) a multi-faceted entity-centric modeling of documents can lead to effective representations, which (3) can have an impact in downstream application, and (4) considering a small number of facets is effective enough. In particular, we visualize entity facets within a set of documents, and demonstrate that indeed different sets of documents reflect different facets of entities. Further, we demonstrate that the proposed entity facet topic model generates better document representations in terms of perplexity, compared to state-of-the-art document representation methods. Moreover, we show that the proposed model outperforms baseline methods in the application of multi-label classification. Finally, we study the impact of EFTM’s parameters and find that a small number of facets better captures entity specific topics, which confirms the intuition that on average an entity has a small number of facets reflected in documents.     

Boosting question answering over knowledge graph with reward integration and policy evaluation under weak supervision

Among existing knowledge graph based question answering (KGQA) methods, relation supervision methods require labeled intermediate relations for stepwise reasoning. To avoid this enormous cost of labeling on large-scale knowledge graphs, weak supervision methods, which use only the answer entity to evaluate rewards as supervision, have been introduced. However, lacking intermediate supervision raises the issue of sparse rewards, which may result in two types of incorrect reasoning path: (1) incorrectly reasoned relations, even when the final answer entity may be correct; (2) correctly reasoned relations in a wrong order, which leads to an incorrect answer entity. To address these issues, this paper considers the multi-hop KGQA task as a Markov decision process, and proposes a model based on Reward Integration and Policy Evaluation (RIPE). In this model, an integrated reward function is designed to evaluate the reasoning process by leveraging both terminal and instant rewards. The intermediate supervision for each single reasoning hop is constructed with regard to both the fitness of the taken action and the evaluation of the unreasoned information remained in the updated question embeddings. In addition, to lead the agent to the answer entity along the correct reasoning path, an evaluation network is designed to evaluate the taken action in each hop. Extensive ablation studies and comparative experiments are conducted on four KGQA benchmark datasets. The results demonstrate that the proposed model outperforms the state-of-the-art approaches in terms of answering accuracy.     

Exploring syntactic structured features over parse trees for relation extraction using kernel methods

Extracting semantic relationships between entities from text documents is challenging in information extraction and important for deep information processing and management. This paper proposes to use the convolution kernel over parse trees together with support vector machines to model syntactic structured information for relation extraction. Compared with linear kernels, tree kernels can effectively explore implicitly huge syntactic structured features embedded in a parse tree. Our study reveals that the syntactic structured features embedded in a parse tree are very effective in relation extraction and can be well captured by the convolution tree kernel. Evaluation on the ACE benchmark corpora shows that using the convolution tree kernel only can achieve comparable performance with previous best-reported feature-based methods. It also shows that our method significantly outperforms previous two dependency tree kernels for relation extraction. Moreover, this paper proposes a composite kernel for relation extraction by combining the convolution tree kernel with a simple linear kernel. Our study reveals that the composite kernel can effectively capture both flat and structured features without extensive feature engineering, and easily scale to include more features. Evaluation on the ACE benchmark corpora shows that the composite kernel outperforms previous best-reported methods in relation extraction.     

Using semantic similarity to reduce wrong labels in distant supervision for relation extraction

Distant supervision (DS) has the advantage of automatically generating large amounts of labelled training data and has been widely used for relation extraction. However, there are usually many wrong labels in the automatically labelled data in distant supervision (Riedel, Yao, & McCallum, 2010). This paper presents a novel method to reduce the wrong labels. The proposed method uses the semantic Jaccard with word embedding to measure the semantic similarity between the relation phrase in the knowledge base and the dependency phrases between two entities in a sentence to filter the wrong labels. In the process of reducing wrong labels, the semantic Jaccard algorithm selects a core dependency phrase to represent the candidate relation in a sentence, which can capture features for relation classification and avoid the negative impact from irrelevant term sequences that previous neural network models of relation extraction often suffer. In the process of relation classification, the core dependency phrases are also used as the input of a convolutional neural network (CNN) for relation classification. The experimental results show that compared with the methods using original DS data, the methods using filtered DS data performed much better in relation extraction. It indicates that the semantic similarity based method is effective in reducing wrong labels. The relation extraction performance of the CNN model using the core dependency phrases as input is the best of all, which indicates that using the core dependency phrases as input of CNN is enough to capture the features for relation classification and could avoid negative impact from irrelevant terms.     

A local tree alignment approach to relation extraction of multiple arguments

In this paper, we address the problem of relation extraction of multiple arguments where the relation of entities is framed by multiple attributes. Such complex relations are successfully extracted using a syntactic tree-based pattern matching method. While induced subtree patterns are typically used to model the relations of multiple entities, we argue that hard pattern matching between a pattern database and instance trees cannot allow us to examine similar tree structures. Thus, we explore a tree alignment-based soft pattern matching approach to improve the coverage of induced patterns. Our pattern learning algorithm iteratively searches the most influential dependency tree patterns as well as a control parameter for each pattern. The resulting method outperforms two baselines, a pairwise approach with the tree-kernel support vector machine and a hard pattern matching method, on two standard datasets for a complex relation extraction task.     

An efficiency relation-specific graph transformation network for knowledge graph representation learning

Knowledge graph representation learning (KGRL) aims to infer the missing links between target entities based on existing triples. Graph neural networks (GNNs) have been introduced recently as one of the latest trendy architectures serves KGRL task using aggregations of neighborhood information. However, current GNN-based methods have fundamental limitations in both modelling the multi-hop distant neighbors and selecting relation-specific neighborhood information from vast neighbors. In this study, we propose a new relation-specific graph transformation network (RGTN) for the KGRL task. Specifically, the proposed RGTN is the first pioneer model that transforms a relation-based graph into a new path-based graph by generating useful paths that connect heterogeneous relations and multi-hop neighbors. Unlike the existing GNN-based methods, our approach is able to adaptively select the most useful paths for each specific relation and to effectively build path-based connections between unconnected distant entities. The transformed new graph structure opens a new way to model the arbitrary lengths of multi-hop neighbors which leads to more effective embedding learning. In order to verify the effectiveness of our proposed model, we conduct extensive experiments on three standard benchmark datasets, e.g., WN18RR, FB15k-237 and YAGO-10-DR. Experimental results show that the proposed RGTN achieves the promising results and even outperforms other state-of-the-art models on the KGRL task (e.g., compared to other state-of-the-art GNN-based methods, our model achieves 2.5% improvement using H@10 on WN18RR, 1.2% improvement using H@10 on FB15k-237 and 6% improvement using H@10 on YAGO3-10-DR).     

实体语义关系分类及应用研究

[目的/意义]实体语义关系分类是信息抽取重要任务之一，将非结构化文本转化成结构化知识，是构建领域本体、知识图谱、开发问答系统、信息检索系统的基础工作。[方法/过程]本文详细梳理了实体语义关系分类的发展历程，从技术方法、应用领域两方面回顾和总结了近5年国内外的最新研究成果，并指出了研究的不足及未来的研究方向。[结果/结论]热门的深度学习方法抛弃了传统浅层机器学习方法繁琐的特征工程，自动学习文本特征，实验发现，在神经网络模型中融入词法、句法特征、引入注意力机制能有效提升关系分类性能。