Social Propagation: Boosting Social Annotations for Web Mining

This paper is concerned with the problem of boosting social annotations using propagation, which is also called social propagation. In particular, we focus on propagating social annotations of web pages (e.g., annotations in Del.icio.us). Social annotations are novel resources and valuable in many web applications, including web search and browsing. Although they are developing fast, social annotations of web pages cover only a small proportion (<0.1%) of the World Wide Web. To alleviate the low coverage of annotations, a general propagation model based on Random Surfer is proposed. Specifically, four steps are included, namely basic propagation, multiple-annotation propagation, multiple-link-type propagation, and constraint-guided propagation. The model is evaluated on a dataset of 40,422 web pages randomly sampled from 100 most popular English sites and ten famous academic sites. Each page’s annotations are obtained by querying the history interface of Del.icio.us. Experimental results show that the proposed model is very effective in increasing the coverage of annotations while still preserving novel properties of social annotations. Applications of propagated annotations on web search and classification further verify the effectiveness of the model.     

Notes and annotations as information resources in a social networking platform

Taking notes and annotations contributes in the learning process. Many platforms are developed as Computer Supported Collaborative Learning (CSCL) thanks to advancements in new technologies. A common limitation of these platforms is the restricted ability to share/retrieve notes and annotations (Su, Yang, Hwang, & Zhang, 2010a). This is because the annotations in these platforms are disconnected from the information system and they are only accessible in the annotation system. As a result, the annotations could not be indexed as any other information resources (e.g., a document). This means that the annotations are not accessible/visible like other resources. In this paper, we present an original semantic model in which notes and annotations are modeled as information resources. The semantic model is used within the MEMORAe web platform. Then we detail an experiment of collaborative learning made within a university course using the MEMORAe web platform. The feedback of this experiment shows us that students are satisfied with the use of the MEMORAe web platform for helping them to index and retrieve notes and annotations as any information resources.     

Image annotation via graph learning

Image annotation has been an active research topic in recent years due to its potential impact on both image understanding and web image search. In this paper, we propose a graph learning framework for image annotation. First, the image-based graph learning is performed to obtain the candidate annotations for each image. In order to capture the complex distribution of image data, we propose a Nearest Spanning Chain (NSC) method to construct the image-based graph, whose edge-weights are derived from the chain-wise statistical information instead of the traditional pairwise similarities. Second, the word-based graph learning is developed to refine the relationships between images and words to get final annotations for each image. To enrich the representation of the word-based graph, we design two types of word correlations based on web search results besides the word co-occurrence in the training set. The effectiveness of the proposed solution is demonstrated from the experiments on the Corel dataset and a web image dataset.     

Adapting web pages using graph partitioning algorithms for user-centric multi-device web browsing

In the era of ubiquitous computing, applications are emerging to benefit from using devices of different users and different capabilities together. This paper focuses on user-centric web browsing using multiple devices, where content of a web page is partitioned, adapted and allocated to devices in the vicinity. We contribute two novel web page partitioning algorithms. They differ from existing approaches by allowing for both, automatic and semi-automatic partitioning. On the one hand, this provides good automatic, web page independent results by utilizing sophisticated structural pre- and postprocessing of the web page. On the other hand, these results can be improved by considering additional semantic information provided through user-generated web page annotations. We further present a performance evaluation of our algorithms. Moreover, we contribute the results of a user study. These clearly show that (1) our algorithms provide good automatic results and (2) the application of user-centric, annotation-based semantic information leads to a significantly higher user satisfaction.     

Detection of malicious and non-malicious website visitors using unsupervised neural network learning

Distributed denials of service (DDoS) attacks are recognized as one of the most damaging attacks on the Internet security today. Recently, malicious web crawlers have been used to execute automated DDoS attacks on web sites across the WWW. In this study, we examine the use of two unsupervised neural network (NN) learning algorithms for the purpose web-log analysis: the Self-Organizing Map (SOM) and Modified Adaptive Resonance Theory 2 (Modified ART2). In particular, through the use of SOM and modified ART2, our work aims to obtain a better insight into the types and distribution of visitors to a public web-site based on their browsing behavior, as well as to investigate the relative differences and/or similarities between malicious web crawlers and other non-malicious visitor groups. The results of our study show that, even though there is a pretty clear separation between malicious web-crawlers and other visitor groups, 52% of malicious crawlers exhibit very ‘human-like’ browsing behavior and as such pose a particular challenge for future web-site security systems. Also, we show that some of the feature values of malicious crawlers that exhibit very ‘human-like’ browsing behavior are not significantly different than the features values of human visitors. Additionally, we show that Google, MSN and Yahoo crawlers exhibit distinct crawling behavior.     

Reflect: A practical approach to web semantics

To date, adding semantic capabilities to web content usually requires considerable server-side re-engineering, thus only a tiny fraction of all web content currently has semantic annotations. Recently, we announced Reflect (http://reflect.ws), a free service that takes a more practical approach: Reflect uses augmented browsing to allow end-users to add systematic semantic annotations to any web-page in real-time, typically within seconds. In this paper we describe the tagging process in detail and show how further entity types can be added to Reflect; we also describe how publishers and content providers can access Reflect programmatically using SOAP, REST (HTTP post), and JavaScript. Usage of Reflect has grown rapidly within the life sciences, and while currently only genes, protein and small molecule names are tagged, we plan to soon expand the scope to include a much broader range of terms (e.g., Wikipedia entries). The popularity of Reflect demonstrates the use and feasibility of letting end-users decide how and when to add semantic annotations. Ultimately, ‘semantics is in the eye of the end-user’, hence we believe end-user approaches such as Reflect will become increasingly important in semantic web technologies.     

Toward a real-time model-based training system

This article describes the development of a real-time model-based training system that provides adaptive “over-the-shoulder” (OTS) instructions to trainees as they learn to perform an Anti-Air Warfare Coordinator (AAWC) task. The long-term goal is to develop a system that will provide real-time instructional materials based on learners’ actions, so that eventually the initial set of instructions on a task can be strengthened, complemented, or overridden at different stages of training. The training system is based on the ACT-R architecture, which serves as the theoretical background for the cognitive model that monitors the learning process of the trainee. An experiment was designed to study the impact of OTS instructions on learning. Results showed that while OTS instructions facilitated short-term learning, (a) they took time away from the processing of current information, (b) their effects tended to decay rapidly in initial stages of training, and (c) their effects on training diminished when the OTS instructions were proceduralized in later stages of training. A cognitive model that learned from both the upfront and OTS instructions was created and provided good fits to the learning and performance data collected from human participants. Our results suggest that to fully capture the symbiotic performance between humans and intelligent training systems, it is important to closely monitor the learning process of the trainee so that instructional interventions can be delivered effectively at different stages of training. We proposed that such a flexible system can be developed based on an adaptive cognitive model that provides real-time predictions on learning and performance.     

Browsing shortcuts as a means to improve information seeking of blind people in the WWW

This paper presents a “Semantic Web application framework” which allows different applications to be designed and developed for improving the accessibility of the World Wide Web (WWW). The framework promotes the idea of creating a community of people federating into groups (ontology creators, annotators, user-agent developers, end-users) each playing a specific role, without the coordination of any central authority. The use of a specialised voice web browser for blind people, called SeEBrowser, is presented and discussed as an example of an accessibility tool developed based on the framework. SeEBrowser utilises annotations of web pages and provides browsing shortcuts. Browsing shortcuts are mechanisms, which facilitate blind people in moving efficiently through various elements of a web page (e.g. functional elements such as forms, navigational aids etc.) during the information-seeking process, hence operating effectively as a vital counterbalance to low accessibility. Finally, an experimental user study is presented and discussed which evaluates SeEBrowser with and without the use of browsing shortcuts.     

Designing an adaptive web-based learning system based on students’ cognitive styles identified online

This study developed an adaptive web-based learning system focusing on students’ cognitive styles. The system is composed of a student model and an adaptation model. It collected students’ browsing behaviors to update the student model for unobtrusively identifying student cognitive styles through a multi-layer feed-forward neural network (MLFF). The MLFF was adopted because of its ability on imprecise or incompletely understood data, ability to generalize and learn from specific examples, ability to be quickly updated with extra parameters, and speed in execution making them ideal for real time applications. The system then adaptively recommended learning content presented with a variety of content and interactive components through the adaptation model based on the student cognitive style identified in the student model. The adaptive web interfaces were designed by investigating the relationships between students’ cognitive styles and browsing patterns of content and interactive components. Training of the MLFF and an experiment were conducted to examine the accuracy of identifying students’ cognitive styles during browsing with the proposed MLFF and the impact of the proposed adaptive web-based system on students’ engagement in learning. The training results of the MLFF showed that the proposed system could identify students’ cognitive styles with high accuracy and the temporal effects should be considered while identifying students’ cognitive styles during browsing. Two factors, the acknowledgment of students’ cognitive styles while browsing and the existence of adaptive web interfaces, were used to assign three classes of college freshmen into three groups. The experimental results revealed that the proposed system could have significant impacts on temporal effects on students’ engagement in learning, not only for students with cognitive styles known before browsing, but also for students with cognitive styles identified during browsing. The results provide evidence of the effectiveness of the adaptive web-based learning system with students’ cognitive styles dynamically identified during browsing, thus validating the research purposes of this study.     

Video Surveillance Online Repository (ViSOR): an integrated framework

The availability of new techniques and tools for Video Surveillance and the capability of storing huge amounts of visual data acquired by hundreds of cameras every day call for a convergence between pattern recognition, computer vision and multimedia paradigms. A clear need for this convergence is shown by new research projects which attempt to exploit both ontology-based retrieval and video analysis techniques also in the field of surveillance. This paper presents the ViSOR (Video Surveillance Online Repository) framework, designed with the aim of establishing an open platform for collecting, annotating, retrieving, and sharing surveillance videos, as well as evaluating the performance of automatic surveillance systems. Annotations are based on a reference ontology which has been defined integrating hundreds of concepts, some of them coming from the LSCOM and MediaMill ontologies. A new annotation classification schema is also provided, which is aimed at identifying the spatial, temporal and domain detail level used. The ViSOR web interface allows video browsing, querying by annotated concepts or by keywords, compressed video previewing, media downloading and uploading. Finally, ViSOR includes a performance evaluation desk which can be used to compare different annotations.     

Procedures for Obtaining and Testing User-Selected Terminologies

Four experiments were conducted to assess procedures for obtaining and testing user-selected terms for task-specific concepts in complex, unfamiliar word-processing instructions. Experiment 1 tested user-selected terms against both user-nominated and the original technical terms. The final three experiments employed only the user-selected and technical terms. The effect of terminology on subjects' abilities to follow the instructions was evaluated by measuring errors and task completion times during the practice period. Comprehension of the instructions was assessed by performance on a transfer task. Extensive practice produced acceptable and comparable performance for all term types. However, instruction comprehension, as measured by the transfer task, was clearly influenced by terminology. User-selected word-processing terms were more understandable than both user-nominated and the original technical terms. In addition, the present study demonstrated that transfer tasks can be more sensitive (and often more appropriate) evaluations of the goodness of a term than are learning measures.     

基于模糊模拟的加权偏爱浏览模式的挖掘

每个网页由不同的专家给出语义上的重要性评估,这些语义评估再被刻画成相应的模糊语言变量,通过模糊模拟的方法,这些模糊语言变量被转化成表示网页重要性的权重。此外,简单地认为用户的访问频度反映了用户的访问兴趣是不准确的,因此在提出的加权支持度和偏爱度概念的基础上,从建立的包含了所有用户浏览信息的FLaAT(Frequent Link and Access Tree)上,挖掘用户偏爱的加权浏览模式。试验证明该算法是行之有效的。     

基于遗传算法的Web用户聚类

按照访问兴趣对用户进行聚类分析是Web挖掘的一项重要内容。在用户访问相似度量中着重考虑浏览路径次序因素;在聚类分析中采用了遗传算法并且依据新的路径相似度计算方法定义适应度函数;遗传算法的全局寻优性可以提高用户聚类的准确性,实验结果验证此算法是有效的。     

基于web日志的连续频繁路径挖掘算法

频繁模式挖掘已成为web使用挖掘的研究热点,本文基于web日志提出一种新的频繁路径的挖掘算法.首先以线性回归方法求解兴趣度,其次将此兴趣度和页面名称作为最基本要素,建立的web浏览树,此浏览树可以完整地表现出web日志中连续、重复的浏览路径,最后在web浏览树上进行分析挖掘频繁浏览路径.该算法经实验证明能更全面地反映用户兴趣所在,挖掘的频繁浏览路径准确、合理.     

Relevance Feedback and Learning in Content-Based Image Search

A major bottleneck in content-based image retrieval (CBIR) systems or search engines is the large gap between low-level image features used to index images and high-level semantic contents of images. One solution to this bottleneck is to apply relevance feedback to refine the query or similarity measures in image search process. In this paper, we first address the key issues involved in relevance feedback of CBIR systems and present a brief overview of a set of commonly used relevance feedback algorithms. Almost all of the previously proposed methods fall well into such framework. We present a framework of relevance feedback and semantic learning in CBIR. In this framework, low-level features and keyword annotations are integrated in image retrieval and in feedback processes to improve the retrieval performance. We have also extended framework to a content-based web image search engine in which hosting web pages are used to collect relevant annotations for images and users' feedback logs are used to refine annotations. A prototype system has developed to evaluate our proposed schemes, and our experimental results indicated that our approach outperforms traditional CBIR system and relevance feedback approaches.     

Learning automatic concept detectors from online video

Concept detection is targeted at automatically labeling video content with semantic concepts appearing in it, like objects, locations, or activities. While concept detectors have become key components in many research prototypes for content-based video retrieval, their practical use is limited by the need for large-scale annotated training sets. To overcome this problem, we propose to train concept detectors on material downloaded from web-based video sharing portals like YouTube, such that training is based on tags given by users during upload, no manual annotation is required, and concept detection can scale up to thousands of concepts. On the downside, web video as training material is a complex domain, and the tags associated with it are weak and unreliable. Consequently, performance loss is to be expected when replacing high-quality state-of-the-art training sets with web video content.This paper presents a concept detection prototype named TubeTagger that utilizes YouTube content for an autonomous training. In quantitative experiments, we compare the performance when training on web video and on standard datasets from the literature. It is demonstrated that concept detection in web video is feasible, and that – when testing on YouTube videos – the YouTube-based detector outperforms the ones trained on standard training sets. By applying the YouTube-based prototype to datasets from the literature, we further demonstrate that: (1) If training annotations on the target domain are available, the resulting detectors significantly outperform the YouTube-based tagger. (2) If no annotations are available, the YouTube-based detector achieves comparable performance to the ones trained on standard datasets (moderate relative performance losses of 11.4% is measured) while offering the advantage of a fully automatic, scalable learning. (3) By enriching conventional training sets with online video material, performance improvements of 11.7% can be achieved when generalizing to domains unseen in training.     

A novel machine learning approach to rank web forum posts

Since the user generated contents in Web forums are rich but vary in quality, ranging from excellent detailed opinions to simple repetition of the content of previous, or even spams, it is difficult to find high quality information in the process of post browsing, retrieval and other Web forum applications. In this paper, we propose a novel machine learning approach named LGPRank to evaluate the web forum posts, where a genetic programming architecture is used to rank Web forum posts according to the qualities of their contents. In order to address the shortcomings of current studies, we take both the semantic-free and semantic-specific information of a post into account. We propose a set of new features named Latent Dirichlet Allocation (LDA) semantic features which are computed in LDA topic space. The proposed features as well as content surface features and forum specific features are used in the learning process. Experiments are conducted on three web forum datasets in comparison with methods used in prior ranking research. LGPRank outperforms all the other methods in terms of P@N, NDCG@N and MAP measures. Furthermore, the experimental results also indicate that the proposed LDA semantic features have a positive effect in improving the ranking performance.     

Digital game-based learning: Impact of instructions and feedback on motivation and learning effectiveness

Although many studies have investigated the effects of digital game-based learning (DGBL) on learning and motivation, its benefits have never been systematically demonstrated. In our first experiment, we sought to identify the conditions under which DGBL is most effective, by analyzing the effects of two different types of instructions (learning instruction vs. entertainment instruction). Results showed that the learning instruction elicited deeper learning than the entertainment one, without impacting negatively on motivation. In our second experiment, we showed that if learners are given regular feedback about their performance, the entertainment instruction results in deep learning. These two experiments demonstrate that a serious game environment can promote learning and motivation, providing it includes features that prompt learners to actively process the educational content.     

Evaluation of key frame-based retrieval techniques for video

We investigate the application of a variety of content-based image retrieval techniques to the problem of video retrieval. We generate large numbers of features for each of the key frames selected by a highly effective shot boundary detection algorithm to facilitate a query by example type search. The retrieval performance of two learning methods, boosting and k-nearest neighbours, is compared against a vector space model. We carry out a novel and extensive evaluation to demonstrate and compare the usefulness of these algorithms for video retrieval tasks using a carefully created test collection of over 6000 still images, where performance is measured against relevance judgements based on human image annotations. Three types of experiment are carried out: classification tasks, category searches (both related to automated annotation and summarisation of video material) and real world searches (for navigation and entry point finding). We also show graphical results of real video search tasks using the algorithms, which have not previously been applied to video material in this way.     

视觉弱监督学习研究进展

视觉理解,如物体检测、语义和实例分割以及动作识别等,在人机交互和自动驾驶等领域中有着广泛的应用并发挥着至关重要的作用。近年来,基于全监督学习的深度视觉理解网络取得了显著的性能提升。然而,物体检测、语义和实例分割以及视频动作识别等任务的数据标注往往需要耗费大量的人力和时间成本,已成为限制其广泛应用的一个关键因素。弱监督学习作为一种降低数据标注成本的有效方式,有望对缓解这一问题提供可行的解决方案,因而获得了较多的关注。围绕视觉弱监督学习,本文将以物体检测、语义和实例分割以及动作识别为例综述国内外研究进展,并对其发展方向和应用前景加以讨论分析。在简单回顾通用弱监督学习模型,如多示例学习(multiple instance learning, MIL)和期望—最大化(expectation-maximization, EM)算法的基础上,针对物体检测和定位,从多示例学习、类注意力图机制等方面分别进行总结,并重点回顾了自训练和监督形式转换等方法;针对语义分割任务,根据不同粒度的弱监督形式,如边界框标注、图像级类别标注、线标注或点标注等,对语义分割研究进展进行总结分析,并主要回顾了基于图像级别类别...