Adapter与Prompt Tuning微调方法研究综述

文本挖掘是数据挖掘的一个分支学科,涵盖多种技术,其中自然语言处理技术是文本挖掘的核心工具之一,旨在帮助用户从海量数据中获取有用的信息。近年来,预训练模型对自然语言处理的研究和发展有重要的推动作用,预训练模型的微调方法也成为重要的研究领域。根据近年来预训练模型微调方法的相关文献,选择目前主流的Adapter与Prompt微调方法进行介绍。对自然语言处理的发展脉络进行简要梳理,分析目前预训练模型微调存在的问题与不足;介绍Adapter与Prompt两类微调方法,对两个研究方向中经典方法进行介绍,并从优缺点和性能等方面进行详细分析;进行总结归纳,阐述目前预训练模型的微调方法存在的局限性并讨论未来发展方向。     

跨模态视觉问答与推理研究进展

随着社交媒体和人机交互技术的快速发展,视频、图像以及文本等多模态数据在互联网中呈爆炸式增长,因此多模态智能研究受到关注。其中,视觉问答与推理任务是跨模态智能研究的一个重要组成部分,也是人类实现人工智能的重要基础,已成功应用于人机交互、智能医疗以及无人驾驶等领域。本文对视觉问答与推理的相关算法进行了全面概括和归类分析。首先,介绍了视觉问答与推理的定义,并简述了当前该任务面临的挑战;其次,从基于注意力机制、基于图网络、基于预训练、基于外部知识库和基于可解释推理机制5个方面对现有方法进行总结和归纳;然后,全面介绍了视觉问答与推理常用公开数据集,并对相关数据集上的已有算法进行详细分析;最后,对视觉问答与推理任务的未来方向进行了展望。     

题目难度评估方法研究综述

题目难度是保证试卷合理性及考试公平性的关键信息,也是智能教学系统(ITS)中的关键参数,有效支撑着包括智能组卷、题目自动生成和个性化习题推荐在内的多项智能教学功能.因此,题目难度评估已成为教育数据挖掘领域的一个重要研究方向,拥有大量研究工作.全面回顾了近十年题目难度评估研究领域的研究进展,将题目难度分为题目绝对难度和题...     

选择式机器阅读理解综述

机器阅读理解是自然语言处理领域的研究热点之一,对提升机器阅读能力和智能水平有着重要意义,为跟进相关领域的研究进展对其进行综述。首先,介绍机器阅读理解的发展历程及主要任务;其次,重点梳理当前选择式机器阅读理解基于深度学习方法的相关工作,并从语义匹配、预训练模型、语义推理、外部知识四个方面展开叙述;归纳总结了相关数据集以及评价指标;最后,对选择式机器阅读理解的未来发展趋势进行了展望。     

抽取式机器阅读理解研究综述

机器阅读理解要求机器能够理解自然语言文本并回答相关问题,是自然语言处理领域的核心技术,也是自然语言处理领域最具挑战性的任务之一。抽取式机器阅读理解是机器阅读理解任务中一个重要的分支,因其更贴合实际情况,更能够反映机器的理解能力,成为当前学术界和工业界的研究热点。对抽取式机器阅读理解从以下四个方面进行了全面地综述：介绍了机器阅读理解任务及其发展历程;介绍了抽取式机器阅读理解任务以及其现阶段存在的难点;对抽取式机器阅读理解任务的主要数据集及方法进行了梳理总结;讨论了抽取式机器阅读理解的未来发展方向。     

汉语零形回指消解研究综述

关于零形回指的研究一直是语言学研究中的一个热点,零形回指消解是自然语言处理中一项十分重要的任务。20多年来,学者们基于语言学规则、机器学习、深度学习等方面,提出了各种研究方法,并取得了大量研究成果。该文首先介绍零形回指的相关概念;接着介绍目前国际上汉语零形回指消解的公开评测资源OntoNotes 5.0数据集及评价指标;其次,系统梳理和对比了国内外汉语零形回指消解所采用的方法;最后,总结和分析了目前零形回指消解研究的主要制约因素,这些因素也正是未来可能的研究方向。     

中文命名实体识别研究综述

随着自然语言处理领域相关技术的快速发展,作为自然语言处理的上游任务,提高命名实体识别的准确率对于后续的文本处理任务而言具有重要的意义。然而,中文和英文语系之间存在差异,导致英文的命名实体识别研究成果难以有效地迁移到中文研究中。因此从以下四方面分析了当前中文命名实体识别研究中的关键问题：首先以命名实体识别的发展历程作为主要线索,从各阶段存在的优缺点、常用方法和研究成果等角度进行了综合论述;其次从序列标注、评价指标、中文分词方法及数据集的角度出发,对中文文本预处理方法进行了总结;接着针对中文字词特征融合方法,从字融合和词融合的角度对当前的研究进行了总结,并对当前中文命名实体识别模型的优化方向进行了论述;最后分析了当前中文命名实体识别在各领域的实际应用。对当前中文命名实体识别的研究进行论述,旨在帮助科研工作者更为全面地了解该任务的研究方向和研究意义,从而为新方法和新改进的提出提供一定的参考。     

基于深度学习的口语理解联合建模算法综述

口语理解是自然语言处理领域的研究热点之一,应用在个人助理、智能客服、人机对话、医疗等多个领域.口语理解技术指的是将机器接收到的用户输入的自然语言转换为语义表示,主要包含意图识别、槽位填充这两个子任务.现阶段,使用深度学习对口语理解中意图识别和槽位填充任务的联合建模方法已成为主流,并且获得了很好的效果.因此,对基于深度学习的口语理解联合建模算法进行总结分析具有十分重要的意义.首先介绍了深度学习技术应用到口语理解的相关工作,然后从意图识别和槽位填充的关联关系对现有的研究工作进行剖析,并对不同模型的实验结果进行了对比分析和总结,最后给出了未来的研究方向及展望.     

基于深度学习的自然语言处理鲁棒性研究综述

近年来,基于深度神经网络的模型在几乎所有自然语言处理任务上都取得了非常好的效果,在很多任务上甚至超越了人类.展现了极强能力的大规模语言模型也为自然语言处理模型的发展与落地提供了新的机遇和方向.然而,这些在基准测试集合上取得很好结果的模型在实际应用中的效果却经常大打折扣.近期的一些研究还发现,在测试数据上替换一个相似词语、增加一个标点符号,甚至只是修改一个字母都可能使得这些模型的预测结果发生改变,效果大幅度下降.即使是大型语言模型,也会因输入中的微小扰动而改变其预测结果.什么原因导致了这种现象的发生?深度神经网络模型真的如此脆弱吗?如何才能避免这种问题的出现?这些问题近年来受到了越来越多的关注,诸多有影响力的工作都不约而同地从不同方面讨论了自然语言处理的鲁棒性问题.在本文中,我们从自然语言处理任务的典型范式出发,从数据构建、模型表示、对抗攻防以及评估评价等四个方面对自然语言处理鲁棒性相关研究进行了总结和归纳,并对最新进展进行了介绍,最后探讨了未来的可能研究方向以及我们对自然语言处理鲁棒性问题的一些思考.     

视线跟踪及其在智能教育中的应用研究综述

与教育相结合是当前人工智能研究的热点方向之一,获得学习状态信息是智能教育中非常重要的一个环节。视线变化可以直接或间接反映心理以及状态的变化,所以视线跟踪在智能教育领域起到重要作用。首先,对智能教育的发展进行了介绍;然后,对视线跟踪技术的发展、当前的研究工作以及研究现状进行了归纳与分析,总结了近3年视线跟踪技术在教育领域的相关应用与研究工作;最后,对视线跟踪技术在教育领域的发展趋势进行了总结与展望。     

面向在线智慧学习的教育数据挖掘技术研究

随着教育信息化进程的深入,学生在线学习数据得到不断积累,为数据驱动的教育评估和智能辅助教学提供良好条件.然而,已有的面向在线智慧学习的教育数据挖掘模型很难从海量、稀疏、高噪的数据中准确分析试题特征和学生学业水平,也较少考虑学生及教师的个性化需求.文中针对上述问题开展若干面向在线智慧学习的教育数据挖掘技术研究工作,以教育学习所涉及的试题、学生、教师为对象,以个性化推荐等技术同教育领域知识相结合为手段,以提高学生学业水平为目标.具体介绍用于试题分析和检索的试题文本表征模型、基于认知诊断的个性化学习资源推荐方法、针对教师的教学建议和指导等方法,以及这些技术所依托的应用平台——科大讯飞在线教育系统“智学网”.最后简单讨论面向在线智慧学习的教育数据挖掘技术未来可能的研究方向.     

Automatically generating effective search queries directly from community question-answering questions for finding related questions

Community Question-Answering platforms are massive knowledge bases of questions and answers pairs produced by their members. In other to provide a vibrant service, they are compelled to provide answers to new posted questions as soon as possible. However, since their dynamic requires their own users to answer questions, there is an inherent delay between posting time and the arrival of good answers. In fact, many of these new questions might be already asked and satisfactorily answered in the past. Ergo, one of the pressing needs of these services is capitalizing on good answers given to related resolved questions across their large-scale knowledge base. To that end, current approaches have studied the effectiveness of human-generated web queries across search logs in fetching related questions and potential good answers from these community archives. However, this kind of strategy is not suitable for questions without click-through data, in particular those recently posted, limiting their capability of providing them with real-time answers.In this paper, we propose an approach to find related questions across the cQA knowledge base, which automatically generate effective search strings directly from question titles and bodies. In so doing, we automatically construct a massive corpus of related questions on top of the relationships yielded by their click-through graph, and generated candidate queries by inspecting dependency paths across the title and body of each question afterwards. Then, we utilize this corpus for automatically annotating the retrieval power of each of these candidates. With this labelled corpus, we study the effectiveness of several learning to rank models enriched with assorted linguistically-motivated properties. Thus deducing the linguistic structure of automatically generated search strings that are effective in finding related questions. Since these models are inferred solely from each question itself, they can be used when search log data (i.e., web queries) is unavailable.Overall, our experiments underline the effectiveness of our approach, in particular our outcomes indicate that named entity recognition is instrumental in structuring and recognizing 2–5 terms effective queries. Furthermore, we carry out experiments considering and ignoring question bodies, and we show that profiting only from question titles is more promising, but most effective queries are harder to detect. Conversely, adding question bodies makes the retrieval of past related questions noisier, but their content helps to generalize models capable of identifying more effective candidates.     

Old Communication, New Literacies: Social Network Sites as Social Learning Resources

This study examined the role of a social network site (SNS) in the lives of 11 high school teenagers from low-income families in the U.S. We conducted interviews, talk-alouds and content analysis of MySpace profiles. Qualitative analysis of these data revealed three themes. First, SNSs facilitated emotional support, helped maintain relationships, and provided a platform for self-presentation. Second, students used their online social network to fulfill essential social learning functions. Third, within their SNS, students engaged in a complex array of communicative and creative endeavors. In several instances, students' use of social network sites demonstrated the new literacy practices currently being discussed within education reform efforts. Based on our findings, we suggest additional directions for related research and educational practices.     

Assessing and improving quality of QVTo model transformations

We investigate quality improvement in QVT operational mappings (QVTo) model transformations, one of the languages defined in the OMG standard on model-to-model transformations. Two research questions are addressed. First, how can we assess quality of QVTo model transformations? Second, how can we develop higher-quality QVTo transformations? To address the first question, we utilize a bottom–up approach, starting with a broad exploratory study including QVTo expert interviews, a review of existing material, and introspection. We then formalize QVTo transformation quality into a QVTo quality model. The quality model is validated through a survey of a broader group of QVTo developers. We find that although many quality properties recognized as important for QVTo do have counterparts in general purpose languages, a number of them are specific to QVTo or model transformation languages. To address the second research question, we leverage the quality model to identify developer support tooling for QVTo. We then implemented and evaluated one of the tools, namely a code test coverage tool. In designing the tool, code coverage criteria for QVTo model transformations are also identified. The primary contributions of this paper are a QVTo quality model relevant to QVTo practitioners and an open-source code coverage tool already usable by QVTo transformation developers. Secondary contributions are a bottom–up approach to building a quality model, a validation approach leveraging developer perceptions to evaluate quality properties, code test coverage criteria for QVTo, and numerous directions for future research and tooling related to QVTo quality.     

A Tutor Assisting Novel Electronic Framework for Qualitative Analysis of a Question Bank

The traditional approach followed by tutors to assess the students is through a set of questions. The quality of a question bank has an impact on the effectiveness of evaluation in educational institutions. Determining the coverage of these questions with respect to a set of prescribed text/reference books helps in evaluating students efficiently. In this paper, we describe a Tutor Assisting e-Framework (TAeF) that enables the tutors to analyze the quality of a question bank. Initially, it clusters all individual topics of each of the input text/reference books according to their dependencies. Later, the questions are classified into these topics. The result is a set of topics, each containing the topic title and the probability by which the question is related to it. Lower the accuracy of the predicted topics, higher is the quality of the question. In other words, if question contains the topic title unaltered, it has a higher probability of being related to the topic; this degrades the quality of question. Furthermore, the congruence relation between the questions and the set of topics is found. This gives the question coverage of each topic. Finally, with this relation, the percentage of understanding the students have developed in each of these topics is computed. The Tutor Assisting e-Framework (TAeF) helps to improve the quality of a question bank, to check the topics covered by each question and the knowledge gained.     

Instructional support for enhancing students’ information problem solving ability

This special issue discusses European research on instructional support to foster students’ ability to solve information-based problems. In this introduction, the concept of information problem solving (IPS) and research in this field of interest will be placed in the broader perspective, which is called information behavior. The focus of this special issue is an educational one and the papers all go into a specific kind of instructional support. The main research questions, findings and conclusions of the six contributions will be outlined. It is concluded that the most important directions for future research deal with how instructional support for different aspect of the process, like for instance how to regulated the process, best can be designed in order to make the instruction adaptive and fit to the learners needs.     

Dynamic epistemic logic of belief change in legal judgments

This study realizes belief/reliability change of a judge in a legal judgment by dynamic epistemic logic (DEL). A key feature of DEL is that possibilities in an agent’s belief can be represented by a Kripke model. This study addresses two difficulties in applying DEL to a legal case. First, since there are several methods for constructing a Kripke model, our question is how we can construct the model from a legal case. Second, since this study employs several dynamic operators, our question is how we can decide which operators are to be applied for belief/reliability change of a judge. In order to solve these difficulties, we have implemented a computer system which provides two functions. First, the system can generate a Kripke model from a legal case. Second, the system provides an inconsistency solving algorithm which can automatically perform several operations in order to reduce the effort needed to decide which operators are to be applied. By our implementation, the above questions can be adequately solved. With our analysis method, six legal cases are analyzed to demonstrate our implementation.     

Similarity/dissimilarity calculation methods of DNA sequences: A survey

DNA sequence similarity/dissimilarity analysis is a fundamental task in computational biology, which is used to analyze the similarity of different DNA sequences for learning their evolutionary relationships. In past decades, a large number of similarity analysis methods for DNA sequence have been proposed due to the ever-growing demands. In order to learn the advances of DNA sequence similarity analysis, we make a survey and try to promote the development of this field. In this paper, we first introduce the related knowledge of DNA similarities analysis, including the data sets, similarities distance and output data. Then, we review recent algorithmic developments for DNA similarity analysis to represent a survey of the art in this field. At last, we summarize the corresponding tendencies and challenges in this research field. This survey concludes that although various DNA similarity analysis methods have been proposed, there still exist several further improvements or potential research directions in this field.     

How do humans inspect BPMN models: an exploratory study

Even though considerable progress regarding the technical perspective on modeling and supporting business processes has been achieved, it appears that the human perspective is still often left aside. In particular, we do not have an in-depth understanding of how process models are inspected by humans, what strategies are taken, what challenges arise, and what cognitive processes are involved. This paper contributes toward such an understanding and reports an exploratory study investigating how humans identify and classify quality issues in BPMN process models. Providing preliminary answers to initial research questions, we also indicate other research questions that can be investigated using this approach. Our qualitative analysis shows that humans adapt different strategies on how to identify quality issues. In addition, we observed several challenges appearing when humans inspect process models. Finally, we present different manners in which classification of quality issues was addressed.     

What happens where during disasters? A Workflow for the multifaceted characterization of crisis events based on Twitter data

Twitter data are a valuable source of information for rescue and helping activities in case of natural disasters and technical accidents. Several methods for disaster‐ and event‐related tweet filtering and classification are available to analyse social media streams. Rather than processing single tweets, taking into account space and time is likely to reveal even more insights regarding local event dynamics and impacts on population and environment. This study focuses on the design and evaluation of a generic workflow for Twitter data analysis that leverages that additional information to characterize crisis events more comprehensively. The workflow covers data acquisition, analysis and visualization, and aims at the provision of a multifaceted and detailed picture of events that happen in affected areas. This is approached by utilizing agile and flexible analysis methods providing different and complementary views on the data. Utilizing state‐of‐the‐art deep learning and clustering methods, we are interested in the question, whether our workflow is suitable to reconstruct and picture the course of events during major natural disasters from Twitter data. Experimental results obtained with a data set acquired during hurricane Florence in September 2018 demonstrate the effectiveness of the applied methods but also indicate further interesting research questions and directions.