基于荔枝和龙眼病虫害知识图谱的问答系统

病虫害一直以来都是影响农作物生产的重要原因,在该领域构建基于知识图谱的问答系统有助于智慧农业的发展,是人工智能时代下农业信息化的重要举措.该研究针对专家内部作物病虫害数据,编写程序对其进行清洗、知识提取等操作,完成该领域的知识图谱构建任务.对病虫害文本进行序列标注,完成实体识别模型构建用于问句实体识别,并构建文本分类模型对问句进行意图识别,编写模版规则匹配可执行Cypher语句,从知识图谱中得到答案.结合Flask与React前后端分离的Web框架,最终形成完整的基于图谱的问答系统.     

基于知识表示的联合问答模型

基于知识库的问答系统旨在通过解析用户的自然语言问句直接在知识库中提取出答案.目前,大多数知识库问答模型都遵循实体检测和关系识别这两个步骤,但是此类方法忽略了知识库本身所蕴含的结构信息以及这两个步骤之间的联系.文中提出了一种基于知识表示的联合问答模型.首先应用知识表示模型将知识库中的实体与关系映射到低维的向量空间,然后通过神经网络将问句也嵌入相同的向量空间,同时检测出问句中的实体,并在此向量空间内度量知识库三元组与问句的语义相似度,从而实现将知识库嵌入和多任务学习引入知识库问答.实验结果表明,所提模型可以极大地提高训练速度,在实体检测和关系识别任务上的准确率达到了主流水平,证明了知识库嵌入及多任务学习可以提升知识库问答任务的性能.     

基于预训练模型与知识蒸馏的法律判决预测算法

针对法律判决预测中罪名预测和法条推荐子任务,提出基于BERT （bidirectional encoder representation from transformers）预训练模型与知识蒸馏策略的多任务多标签文本分类模型.为挖掘子任务间的关联,提高预测准确率,运用BERT预训练模型进行多任务学习,建立BERT₁₂multi文本分类模型;针对罪名、法条类别中的样本不均衡问题,采用分组的焦点损失（focal loss）以增强模型对于罕见罪名及法条的辨别能力;为降低模型计算复杂度并且提高模型推理速度,提出一种以教师模型评价为参考的知识蒸馏策略,通过动态平衡蒸馏中的蒸馏损失和分类损失,将BERT₁₂multi压缩为浅层结构的学生模型.综上,构建出可以处理不均衡样本且具有较高推理速度的多任务多标签文本分类模型BERT₆multi.在CAIL2018数据集上的实验表明:采用预训练模型及分组focal loss可显著提高法律判决预测的性能;通过融入教师模型评价,知识蒸馏得到的学生模型推理速度提高近一倍,并且在罪名预测及法条推荐任务...     

基于Bi-LSTM和CNN并包含注意力机制的社区问答问句分类方法

问句分类的目标是将用户提出的自然语言问句分到预先设定的类别.在社区问答中,如何准确高效的对问句进行分类是一项重要任务.本文提出了一种基于深度神经网络的问句分类方法,该方法首先将问句用词向量进行表示,然后用融合双向长短时记忆网络（Bi-LSTM）和卷积神经网络（CNN）结构并包含注意力机制的深度学习模型提取问句特征进行分类.该方法的特色在于利用Bi-LSTM和CNN在句子级文本表示的优点,充分捕捉问句特征,并结合问句的对应答案来表示问句,丰富了问句信息.实验表明,该问句分类方法准确率较高,在多个数据集上取得不错结果.     

融合法律文本结构信息的刑事案件判决预测

近年来,法律领域的智能化引起了学界的广泛关注。选取法律领域中十分重要的法律判决预测任务作为研究重点,法律判决预测包含推荐相关法条、定罪和刑期预测等三个子任务。随着深度学习在各个领域的广泛应用,一些研究者将深度学习方法引入法律判决预测任务并取得了较好的效果。现有基于深度学习的法律判决预测方法通常是通过构建案情描述和法条之间的注意力来提升模型预测能力,或者利用三个法律判决预测子任务间的关系来提升整体的性能。但是这些工作未考虑法律文本中的多层层次化信息,如刑法第三百九十七条包含职务侵占罪和玩忽职守罪,其法条大类是渎职罪,并且每个罪行有不同的刑期。针对该问题,考虑引入法律文本的多层层次化信息用于法律判决预测任务。具体来说,对法律文本的多层结构信息进行预处理,并利用协同注意力机制将法条的多层信息融入到案情描述中,得到每个子任务的融合不同层次的法律信息的案情描述表示,从而提升司法判决预测任务的性能。在真实的法律判决预测任务公开的数据集上进行了实验,结果显示提出的融合法律文本多层结构信息的模型在法律判决预测任务上优于当前最好的模型。对法律智能化的未来和发展进行了展望。     

基于自然语言理解的受限领域自动应答系统

提出了一种在受限领域内进行自动应答的系统总体结构，通过构建领域词汇库、领域本体库、领域真实问句语料库及领域知识文本库等领域资源库，并借助于这些领域资源来理解问句和知识文本，在一定程度上降低了理解的难度。文章以银行业务咨询服务为例，分析了系统结构及实现过程，对问句分析、答案析取、问句语料库构建、文本知识标注等功能模块的实现进行了详细介绍。     

基于本体的网络入侵知识库模型研究

在信息安全领域,网络入侵知识库对有效分析和防御网络非法入侵起着重要作用,然而网络入侵知识库的构建是研究的难点之一.本体作为一种能为特定领域提供知识共享的概念模型建模工具,已经在各领域得到广泛应用.针对当前还没有一个完善的网络入侵知识本体,研究基于本体的网络入侵知识库模型,构建了网络入侵知识本体.首先,在深入分析网络入侵技术的基础上,形式化定义了各类网络入侵行为,给出了多层次、多维度的网络入侵知识库分类体系.接着,结合本体建模原则,构建了由网络入侵知识领域本体、任务本体、应用本体和原子本体组成的网络入侵知识本体,并给出它们之间的逻辑关系和组织结构.最后,通过两个网络场景,验证了模型用于获取网络入侵知识的有效性.     

基于知识库的开放领域问答系统

问答系统能够理解用户问题,并直接返回答案。现有问答系统大多是面向领域的,仅能回答特定领域的问题。文中提出了基于大规模知识库的开放领域问答系统实现方法。该系统首先采用自定义词典分词和CRF模型相结合的方法识别问句中的主体;其次,采用模糊匹配方法将问句中的主体与知识库中实体建立链接;然后,通过相似度计算以及规则匹配等多种方法识别问句中的谓词并与知识库实体的属性建立关联;最后,进行实体消歧和答案获取。该系统平均F-Measure值为0.695 6,表明所提方法在基于知识库的开放领域问答上具有可行性。     

基于表示学习的知识库问答研究进展与展望

面向知识库的问答(Question answering over knowledge base, KBQA)是问答系统的重要组成. 近些年, 随着以深度学习为代表的表示学习技术在多个领域的成功应用, 许多研究者开始着手研究基于表示学习的知识库问答技术. 其基本假设是把知识库问答看做是一个语义匹配的过程. 通过表示学习知识库以及用户问题的语义表示, 将知识库中的实体、关系以及问句文本转换为一个低维语义空间中的数值向量, 在此基础上, 利用数值计算, 直接匹配与用户问句语义最相似的答案. 从目前的结果看, 基于表示学习的知识库问答系统在性能上已经超过传统知识库问答方法. 本文将对现有基于表示学习的知识库问答的研究进展进行综述, 包括知识库表示学习和问句(文本)表示学习的代表性工作, 同时对于其中存在难点以及仍存在的研究问题进行分析和讨论.     

面向领域知识复用的C4ISR系统需求建模方法

为了准确地获取和复用C4ISR系统需求,提出基于多本体的需求获取与建模框架。详细介绍了在此框架下进行需求建模的过程,首先明确用户的目标模型,然后使用目标对已有的领域知识库进行查询,如果发现匹配的目标,则利用目标-任务的映射关系抽取相关知识,如果没有匹配的目标,则生成概念描述框架引导用户提出需求,最后将两部分需求合并形成最终的需求模型。需求建模过程一方面能够复用已有需求,另一方面能够在本体的引导下提出新的需求,保证了需求模型的质量和建模的效率。     

Legal Ontologies in Knowledge Engineering and Information Management

In this article we describe two core ontologies of law that specify knowledge that is common to all domains of law. The first one, FOLaw describes and explains dependencies between types of knowledge in legal reasoning; the second one, LRI-Core ontology, captures the main concepts in legal information processing. Although FOLaw has shown to be of high practical value in various applied European ICT projects, its reuse is rather limited as it is rather concerned with the structure of legal reasoning than with legal knowledge itself: as many other “legal core ontologies”, FOLaw is therefore rather an epistemological framework than an ontology. Therefore, we also developed LRI-Core. As we argue here that legal knowledge is based to a large extend on common-sense knowledge, LRI-Core is particularly inspired by research on abstract common-sense concepts. The main categories of LRI-Core are: physical, mental and abstract concepts. Roles cover in particular social worlds. Another special category are occurrences; terms that denote events and situations. We illustrate the use of LRI-Core with an ontology for Dutch criminal law, developed in the e-Court European project.     

Understanding the law: improving legal knowledge dissemination by translating the contents of formal sources of law

Considerable attention has been given to the accessibility of legal documents, such as legislation and case law, both in legal information retrieval (query formulation, search algorithms), in legal information dissemination practice (numerous examples of on-line access to formal sources of law), and in legal knowledge-based systems (by translating the contents of those documents to ready-to-use rule and case-based systems). However, within AI & law, it has hardly ever been tried to make the contents of sources of law, and the relations among them, more accessible to those without a legal education. This article presents a theory about translating sources of law into information accessible to persons without a legal education. It illustrates the theory by providing two elaborated examples of such translation ventures. In the first example, formal sources of law in the domain of exchanging police information are translated into rules of thumb useful for policemen. In the second example, the goal of providing non-legal professionals with insight into legislative procedures is translated into a framework for making available sources of law through an integrated legislative calendar. Although the theory itself does not support automating the several stages described, in this article some hints are given as to what such automation would have to look like.

The structuring of legal knowledge in LOIS

Legal information retrieval is in need of the provision of legal knowledge for the improvement of search strategies. For this purpose, the LOIS project is concerned with the construction of a multilingual WordNet for cross-lingual information retrieval in the legal domain. In this article, we set out how a hybrid approach, featuring lexically and legally grounded conceptual representations, can fit the cross-lingual information retrieval needs of both legal professionals and laymen     

Out of their minds: legal theory in neural networks

This paper examines the use of connectionism (neural networks) in modelling legal reasoning. I discuss how the implementations of neural networks have failed to account for legal theoretical perspectives on adjudication. I criticise the use of neural networks in law, not because connectionism is inherently unsuitable in law, but rather because it has been done so poorly to date. The paper reviews a number of legal theories which provide a grounding for the use of neural networks in law. It then examines some implementations undertaken in law and criticises their legal theoretical naïvete. It then presents a lessons from the implementations which researchers must bear in mind if they wish to build neural networks which are justified by legal theories.     

The application of AI to law

There is much interest in moving AI out into real world applications, a move which has been encouraged by recent funding which has attempted to show industry and commerce can benefit from the Fifth Generation of computing. In this article I suggest that the legal application area is one which is very much more complex than it might — at first sight — seem. I use arguments from the sociology of law to indicate that the viewing of the legal system as simply a rule-bound discipline is inherently nave. This, while not new in jurisprudence, is — as the literature of AI and law indicates — certainly novel to the field of artificial intelligence. The socio-legal argument provided is set within the context of AI as one more example of the failure of scientific success and method to easily transmit itself over into the social sciences.     

AI and the Conquest of Complexity in Law

The paper identifies some of the problems with legal systems and outlines the potential of AI technology for overcoming them. For expository purposes, this outline is based on a simplified epistemology of the primary functions of law. Social and philosophical impediments from the side of the legal community to taking advantage of the potential of this technology are discussed and strategic recommendations are given. What is possible is not independent of what we believe to be possible. The possibilities of such developments in the practical world depends upon their being grasped imaginatively by the people who make the practical world work. Neil MacCormick (1993, p.18)     

Development of an Internal Web Site for a Law Firm

Abstract

The Information Services Department at the Washington, DC law firm, Dickstein Shapiro Morin & Oshinsky, decided in 1995 to provide firmwide access to the Internet. Two separate World Wide Web home pages would be created to meet the divergent needs of the Marketing Department and the information resource requirements of the attorneys, who would be using the Internet for research. In the course of developing Internet pages for the firm, the Library began working in conjunction with attorneys in various practice sections to locate and create Internet resources. The cooperation between attorneys and the Libraiy has led to Internet pages of hypertext linked resources for each specific practice section within the firm. Furthermore, some of the pages of resources have evolved into home pages for the various sections.     

Legal informatics research in Italy: The Istituto per la Documentazione Giuridica of the Italian National Research Council

The IDG was originally founded to carry out research into the collection and processing of documentation relating to Italian legislation, case law and legal authority. The Institute has since concentrated on automated documentation and legal informatics, as well as the application of artificial intelligence to the law. This article describes the many projects undertaken at the Institute. Elio Fameli has, since 1971, been a researcher at the Instituto per la Documentazione Giuridica of Florence of the Italian National Research Council. He has worked in legal informatics generally but is mainly interested in problems specifically related to information systems and the application of artificial intelligence to the law. Currently, his research is focused on legal expert systems and the methodology and technology for their construction. He has, since 1975, been an editor of the international journal, Informatica e Dirrito and a member of the editorial board of the International Bibliography on Computers and the Law, published as a regular issue of the aformentioned journal.     

Preserving and ensuring long-term access to digitally born legal information

Written laws, records and legal materials form the very foundation of a democratic society. Lawmakers, legal scholars and everyday citizens alike need, and are entitled, to access the current and historic materials that comprise, explain, define, critique and contextualize their laws and legal institutions. The preservation of legal information in all formats is imperative. Thus far, the twenty-first century has witnessed unprecedented mass-scale acceptance and adoption of digital culture, which has resulted in an explosion in digital information. However, digitally born materials, especially those that are published directly and independently to the Web, are presently at an extremely high risk of permanent loss. Our legal heritage is no exception to this phenomenon, and efforts must be put forth to ensure that our current body of digital legal information is not lost. The authors explored the role of the United States law library community in the preservation of digital legal information. Through an online survey of state and academic law library directors, it was determined that those represented in the sample recognize that digitally born legal materials are at high risk for loss, yet their own digital preservation projects have primarily focused upon the preservation of digitized print materials, rather than digitally born materials. Digital preservation activities among surveyed libraries have been largely limited by a lack of funding, staffing and expertise; however, these barriers could be overcome by collaboration with other institutions, as well as participation in a large-scale regional or national digital preservation movement, which would allow for resource-sharing among participants. One such collaborative digital preservation program, the Chesapeake Project, is profiled in the article and explored as a collaborative effort that may be expanded upon or replicated by other institutions and libraries tackling the challenges of digital preservation.     

Computers and legal reasoning: Developments in Germany

The importance of reasoning in law is pointed out. Law and jurisprudence belong to the reasoning-conscious disciplines. Accordingly, there is a long tradition of logic in law. The specific methods of professional work in law are to be seen in close connection with legal reasoning. The advent of computers at first did not touch upon legal reasoning (or the professional work in law). At first computers could be used only for general auxiliary functions (e.g., numerical calculations in tax law). Gradually, the use of computers for auxiliary functions in law has become more specific and more sophisticated (e.g., legal information retrieval), touching more closely upon professional legal work. Moreover, renewed interest in AI has also fostered interest in AI in law, especially for legal expert systems. AI techniques can be used in support of legal reasoning. Yet until now legal expert systems have remained in the research and development stage and have hardly succeeded in becoming a profitable tool for the profession. Therefore it is hoped that the two lines of computer support, for auxiliary functions in law and for immediate support of legal reasoning, may unite in the future.Herbert Fiedler is professor of Legal Informatics, general theory of law and penal law in the Department of Economics and Law at the University of Bonn.