Web Page Segmentation Revisited

Content-oriented XML retrieval systems support access to XML repositories by retrieving, in response to user queries, XML document components (XML elements) instead of whole documents. The retrieved XML elements should not only contain information relevant to the query, but also should be specific to the given query (i.e. do not discuss other irrelevant topics). To score XML elements according to how relevant and specific they are given a query, the content and logical structure of XML documents have been widely used. This thesis aims to examine a new source of evidence deriving from the semantic decomposition of XML documents. We consider that XML documents can be semantically decomposed through the application of a topic segmentation algorithm. Using the semantic decomposition and the logical structure of XML documents, we define the notion of topic shifts in an XML element. We then formalise the number of topic shifts to reflect the element's relevance, and more particularly its specificness, to the given user's query. This thesis investigates the use of topic shifts in content-oriented XML retrieval, which is mainly involved in retrieving information from semi-structured (XML) documents. First, we examine the characteristics of XML elements reflected by their number of topic shifts. Second, we use the number of topic shifts to estimate the relevance of the elements in the collection. Finally, we use topic shifts to provide a focused access to XML documents, which aims to determine not only relevant elements, but those at the right level of granularity. The main contributions of this thesis are the introduction of topic shifts in the context of content-oriented XML retrieval and the extensive evaluation of the ways this evidence can be employed in retrieving XML elements. This thesis demonstrates that topic shifts in XML elements constitute a useful source of evidence for both improving the ranking of XML elements, and determining elements at the right level of granularity in content-oriented XML retrieval. The thesis is available online at http://elham.ashoori.org/publications/phd-thesis.pdf.

Comics, comic books, and graphic novels are increasingly the target of seriously scholarly attention in the humanities. Moreover, comic books are exceptionally complex documents, with intricate relationships between pictorial and textual elements and a wide variety of content types within a single comic book publication. The complexity of these documents, their combination of textual and pictorial elements, and the collaborative nature of their production shares much in common with other complex documents studied by humanists — illuminated manuscripts, artists’ books, illustrated poems like those of William Blake, letterpress productions like those of the Kelmscott Press, illustrated children’s books, and even Web pages and other born-digital media. Comic Book Markup Language, or CBML, is a TEI-based XML vocabulary for encoding and analyzing comic books, comics, graphic novels, and related documents. This article discusses the goals and motivations for developing CBML, reviews the various content types found in comic book publications, provides an overview and examples of the key features of the CBML XML vocabulary, explores some of the problems and challenges in the encoding and digital representation of comic books, and outlines plans for future work. The structural, textual, visual, and bibliographic complexity of comic books make them an excellent subject for the general study of complex documents, especially documents combining pictorial and textual elements.

Mobile devices have already been widely used to access the Web. However, because most available web pages are designed for desktop PC, it is inconvenient to browse these large web pages on a mobile device with a small screen. In this paper, we propose a new web page transforming scheme to facilitate navigation on a small-form-factor device based on cloud computing. Different with other existing schemes, our approach is based on page segmentation and template-rendering. The semantic structure of a web page is extracted by DOM (Document Object Model) analyzing. Such semantic structure is hierarchical. Each node is corresponded to a block area of one web page and is in charge of reconstructing the whole web page. According to different content styles, the template-rendering scheme generates different display modes. Experimental results show good experience.

My PhD research aims to develop novel solutions to the challenge of identifying web page structure through the visual analysis of web pages as images. The intention is to then combine this back end design with various front end applications in order to provide improved web experiences for users with assistive needs (e.g. assisting visually impaired users by supporting more selective screenreader output, or improving experiences of users with cognitive deficits by allowing reduction of clutter or zooming in on selected web page content). I propose to build a comprehensive computer vision-based system to analyse the semantic structure of web pages based purely on an image of the rendered page, which will produce a rich representation of the page as a tree of regions labelled according to their semantic role. Most research into web page segmentation has focused on the use of the structure of the DOM tree and visual features derived from properties specified in the DOM tree. I argue, however, that the image of the rendered page may be a better representation to use, since it is created by the page designer to convey the structure of the page to the user, while the source code and DOM tree are simply intended to cause the browser's rendering engine to produce the correct appearance, and treat many types of content as black boxes. Additionally, my proposed system uses exactly the information seen by a user regardless of implementation method; this gives advantages in implementation-independence and versatility.

Contrastive learning, a successful form of representational learning, has shown promising results in pretraining deep learning (DL) models for downstream tasks. When working with limited annotation data, as in medical image segmentation tasks, learning domain-specific local representations can further improve the performance of DL models. In this work, we extend the contrastive learning framework to utilize domain-specific contrast information from unlabeled Magnetic Resonance (MR) images to improve the performance of downstream MR image segmentation tasks in the presence of limited labeled data. The contrast in MR images is controlled by underlying tissue properties (e.g., T1 or T2) and image acquisition parameters. We hypothesize that learning to discriminate local representations based on underlying tissue properties should improve subsequent segmentation tasks on MR images. We propose a novel constrained contrastive learning (CCL) strategy that uses tissue-specific information via a constraint map to define positive and negative local neighborhoods for contrastive learning, embedding this information in the representational space during pretraining. For a given MR contrast image, the proposed strategy uses local signal characteristics (constraint map) across a set of related multi-contrast MR images as a surrogate for underlying tissue information. We demonstrate the utility of the approach for downstream: (1) multi-organ segmentation tasks in T2-weighted images where a DL model learns T2 information with constraint maps from a set of 2D multi-echo T2-weighted images (n=101) and (2) tumor segmentation tasks in multi-parametric images from the public brain tumor segmentation (BraTS) (n=80) dataset where DL models learn T1 and T2 information from multi-parametric BraTS images. Performance is evaluated on downstream multi-label segmentation tasks with limited data in (1) T2-weighted images of the abdomen from an in-house Radial-T2 (Train/Test=30/20), (2) public Cartesian-T2 (Train/Test=6/12) dataset, and (3) multi-parametric MR images from the public brain tumor segmentation dataset (BraTS) (Train/Test=40/50). The performance of the proposed CCL strategy is compared to state-of-the-art self-supervised contrastive learning techniques. In each task, a model is also trained using all available labeled data for supervised baseline performance. The proposed CCL strategy consistently yielded improved Dice scores, Precision, and Recall metrics, and reduced HD95 values across all segmentation tasks. We also observed performance comparable to the baseline with reduced annotation effort. The t-SNE visualization of features for T2-weighted images demonstrates its ability to embed T2 information in the representational space. On the BraTS dataset, we also observed that using an appropriate multi-contrast space to learn T1+T2, T1, or T2 information during pretraining further improved the performance of tumor segmentation tasks. Learning to embed tissue-specific information that controls MR image contrast with the proposed constrained contrastive learning improved the performance of DL models on subsequent segmentation tasks compared to conventional self-supervised contrastive learning techniques. The use of such domain-specific local representations could help understand, improve performance, and mitigate the scarcity of labeled data in MR image segmentation tasks.

DrasCLR: A self-supervised framework of learning disease-related and anatomy-specific representation for 3D lung CT images

CHAPTER 1 - Cross-Site Scripting

Due to the explosive growth of the Internet technology, children commonly search information using a Web search engine for their homework and satisfy their curiosity. However, there are few Web search engines considering children’s inherent characteristics, e.g., children prefer to view images on a Web page rather than difficult texts. Therefore, general search results are neither friendly nor satisfactory to children. In this paper, to support children to obtain suitable information for them, we propose a method to re-rank a general search engine’s ranking according to the children-friendly score. Our method determines the score based on the structure of a Web page and its text. We conduct an experiment to verify the re-ranked results match children’s preferences. As a ground-truth, we chose 300 Web pages and asked 34 elementary school students whether these Web pages are preferable for them. The result shows that our method can re-rank children-friendly pages highly.KeywordsWeb searchChildrenRe-ranking

<p><strong>Because of its wide range of applications, generative artificial intelligence (Generative AI) has received a lot of attention in academia and industry. Despite significant advances in computer vision and natural language processing, the use of Generative AI in tabular data is still underexplored. This gap is especially significant given the prevalence of tabular data as the primary data modality. This thesis seeks to close this gap by focusing on the efficient generation, evaluation, refinement, and application of tabular data, while addressing the challenges inherent in its heterogeneous nature.</strong></p><p>The primary goal of this thesis is to develop efficient algorithms for tabular data synthesis, advancing the field of Generative AI in the tabular domain. To achieve this goal, four specific research objectives have been outlined.</p><p>First, this thesis addresses gaps in evaluation metrics by unifying a framework for the comprehensive and consistent assessment of synthetic data. The proposed framework is designed to meet diverse business requirements across various downstream tasks by incorporating a wide range of advanced metrics. These metrics cover various types of evaluations, including univariate, bivariate, multivariate, cluster, and record-level evaluations. Additionally, standardized visualizations are provided to facilitate qualitative assessments. The results indicate that the proposed evaluation framework not only enables consistent comparisons, rankings, and the selection of different data synthesis approaches but also acts as a valuable tool to promptly assess the reliability of results obtained from synthetic data.</p><p>Second, this thesis presents innovative algorithms for the generation of tabular data, aiming to significantly enhance the quality of data synthesis. One of the contributions is the development of a reversible feature engineering pipeline designed to automatically represent tabular data in an efficient format while also ensuring that the transformed data can be easily converted back to its original format. Additionally, the thesis proposes novel deep learning-based tabular data generation models that are capable of learning the joint distribution of multivariate datasets without relying on predefined distribution assumptions. Experimental results highlight the effectiveness of the proposed algorithms in handling heterogeneous data, demonstrating superior performance in most scenarios within the same training duration when compared to similar alternatives.</p><p>Third, this thesis introduces innovative synthetic data prototype selection algorithms with the goal of refining the generated samples. This approach aims to leverage the advantages of Deep Generative Models, which, once trained, can generate unlimited and diverse synthetic data. By carefully identifying and selecting high-quality samples or removing unrealistic ones, the quality of the synthetic data can be enhanced from a post-processing perspective. Building on this hypothesis, we recognize the iterative nature inherent in the data synthesis procedure, where the processes of data generation, evaluation, and refinement should operate repeatedly in a cyclical flow. During the data generation phase, the evaluation framework assists in identifying potential issues and risks based on use cases, while the refinement (post-processing) step iteratively improves synthetic data in alignment with the evaluation outcomes.</p><p>Last, this thesis applies and validates the proposed models across various domains, including business, healthcare, and government, each requiring distinct downstream tasks. Specifically, we assess our algorithms from the standpoint of data balancing using churn data, evaluate our models focusing on data augmentation with health data, and test our algorithms from the perspective of data representation considering data privacy concerns within sensitive citizen data. These applications serve as practical illustrations, demonstrating the effectiveness and utility of the proposed Generative AI models in real-world scenarios.</p><p>In summary, this thesis makes a substantial contribution to the advancement of Generative AI within the domain of tabular data. It not only presents innovative algorithms and evaluation methods but also introduces practical frameworks applicable to real-world scenarios. The findings demonstrate the considerable capacity of Generative AI to fundamentally transform tabular data in various fields, ultimately leading to improved data availability, quality, and quantity.</p>

Web pages contain a large amount of valuable information and resources, meanwhile may update at any time. However, the current Web-data extraction algorithms are generally targeted at specific web page structure. When web pages update, the problem which is caused by the changes of web pages may be encountered, leading to the inability to extract web page information or wrong information. In order to solve this problem, this paper proposes a new method to extract the feature values of each area in the web page through page rendering, and then combine the DOM tree structure of the page, semantic similarity and other information, so that it can still extract the target data correctly after the structure of the web page changes.

Advanced Natural Language Processing Tools for Web Information Retrieval, Content Analysis, and Synthesis. The goal of this SBIR was to implement and evaluate several advanced Natural Language Processing (NLP) tools and techniques to enhance the precision and relevance of search results by analyzing and augmenting search queries and by helping to organize the search output obtained from heterogeneous databases and web pages containing textual information of interest to DOE and the scientific-technical user communities in general. The SBIR investigated 1) the incorporation of spelling checkers in search applications, 2) identification of significant phrases and concepts using a combination of linguistic and statistical techniques, and 3) enhancement of the query interface and search retrieval results through the use of semantic resources, such as thesauri. A search program with a flexible query interface was developed to search reference databases with the objective of enhancing search results from web queries or queries of specialized search systems such as DOE's Information Bridge. The DOE ETDE/INIS Joint Thesaurus was processed to create a searchable database. Term frequencies and term co-occurrences were used to enhance the web information retrieval by providing algorithmically-derived objective criteria to organize relevant documents into clusters containing significant terms. A thesaurus provides an authoritative overview and classification of a field of knowledge. By organizing the results of a search using the thesaurus terminology, the output is more meaningful than when the results are just organized based on the terms that co-occur in the retrieved documents, some of which may not be significant. An attempt was made to take advantage of the hierarchy provided by broader and narrower terms, as well as other field-specific information in the thesauri. The search program uses linguistic morphological routines to find relevant entries regardless of whether terms are stored in singular or plural form. Implementation of additional inflectional morphology processes for verbs can enhance retrieval further, but this has to be balanced by the possibility of broadening the results too much. In addition to the DOE energy thesaurus, other sources of specialized organized knowledge such as the Medical Subject Headings (MeSH), the Unified Medical Language System (UMLS), and Wikipedia were investigated. The supporting role of the NLP thesaurus search program was enhanced by incorporating spelling aid and a part-of-speech tagger to cope with misspellings in the queries and to determine the grammatical roles of the query words and identify nouns for special processing. To improve precision, multiple modes of searching were implemented including Boolean operators, and field-specific searches. Programs to convert a thesaurus or reference file into searchable support files can be deployed easily, and the resulting files are immediately searchable to produce relevance-ranked results with builtin spelling aid, morphological processing, and advanced search logic. Demonstration systems were built for several databases, including the DOE energy thesaurus.

In our research, we focus on ways to crawl and scrape structured data embedded in web pages and collect it as an available data set. To realize large-scale structured data collection, we develop SDC, specialized structured data extraction crawler.SDC integrates the three modules required for extracting structured data: crawling, scraping, and output generation. In particular, we are paying attention to the fact that only simple syntax and semantics are used for structured data. In this way, information extraction from web pages that exist in multiple domains is realized with a simple single setting. It is not necessary to consider the hierarchical structure of websites and the DOM structure of web pages. The users can also display a part of it as a preview to see if the extracted data is what they need.However, even if structured data is used, there are cases where how elements are written differently for each domain. For example, in an e-book sales website, one element is given to each author of the book, but in another site, the list of multiple authors is described as a comma-separated string. Also, there is a website that implements navigation such as pagination using Ajax, and it is necessary to execute the corresponding JavaScript to obtain hyperlink from this site. To cope with the difference between these domains, it is possible for users to quickly specify the data extraction mechanism corresponding to individual websites complementary.As the experiment, we extracted 500 top domains with a large number of links on the websites by our system, that 254 sites where structured data are present, and we have been successfully extracted from 243 sites.

This paper describes a dynamic rearrangement mechanism of Web page layouts. The aim of this research is to provide webmasters or administrators with a means of determining easily the best suited content layout of their Web pages in a short period of time by automatically testing and relocating the contents with a trend of user actions. We will become able to modify dynamically the structure and layout of Web pages on the client browsers using the information of user behaviors. Our method provides the means of changing the contents inside or between Web pages and updating the contents on Web browsers, and it shorten the period of time for optimization.KeywordsUser BehaviorCellular PhoneContent DeliveryContent Management SystemPage ViewThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.

As web pages are becoming increasingly complex, with images and different page layouts, understanding how users examine the page is important. We used eye-tracking analysis of user behavior and performance in the use of web. Using GP3 Eye Tracker we created tracking system, which take into account syntactic elements and content structure of the tested web pages.

Parsing the semantic structure of a web page is a key component of web information extraction. Successful extraction algorithms usually require large-scale training and evaluation datasets, which are difficult to acquire. Recently, crowdsourcing has proven to be an effective method of collecting large-scale training data in domains that do not require much domain knowledge. For more complex domains, researchers have proposed sophisticated quality control mechanisms to replicate tasks in parallel or sequential ways and then aggregate responses from multiple workers. Conventional annotation integration methods often put more trust in the workers with high historical performance; thus, they are called performance-based methods. Recently, Rzeszotarski and Kittur have demonstrated that behavioral features are also highly correlated with annotation quality in several crowdsourcing applications. In this article, we present a new crowdsourcing system, called Wernicke, to provide annotations for web information extraction. Wernicke collects a wide set of behavioral features and, based on these features, predicts annotation quality for a challenging task domain: annotating web page structure. We evaluate the effectiveness of quality control using behavioral features through a case study where 32 workers annotate 200 Q&A web pages from five popular websites. In doing so, we discover several things: (1) Many behavioral features are significant predictors for crowdsourcing quality. (2) The behavioral-feature-based method outperforms performance-based methods in recall prediction, while performing equally with precision prediction. In addition, using behavioral features is less vulnerable to the cold-start problem, and the corresponding prediction model is more generalizable for predicting recall than precision for cross-website quality analysis. (3) One can effectively combine workers’ behavioral information and historical performance information to further reduce prediction errors.