Thursday, June 29, 2017

2017-06-29: Joint Conference on Digital Libraries (JCDL) 2017 Trip Report

The 2017 Joint Conference on Digital Libraries (JCDL) took place at the University of Toronto, Canada. From June 19-23, we (WS-DL) attended workshops, tutorials, panels, and a doctoral consortium. The theme of this year's conference was #TOscale, #TOanalyze, and #TOdiscover. The conference provided researchers from disciplines such as Digital Library research and information science, with the opportunity to communicate the findings of their respective research areas.
Day 1 (June 19)
The first day (pre-conference) of the conference kicked off with a Doctoral Consortium and a Tutorial - Introduction to Digital Libraries. These events took place in parallel with a Workshop - 6th International Workshop on Mining Scientific Publications (WOSP 2017). The final event of the day was a tutorial titled, "Scholarly Data Mining: Making Sense of the Scientific Literature"

Day 2 (June 20)
The conference official started on the second day with opening remarks from Ian Milligan, shortly followed by a keynote from Liz Lyon in which she presented a retrospective on data management, highlighting the successes and achievements of the last decade, as well as assessing the the current state of data, and providing insight into the research, policies and practices needed to sustain progress.
Following Liz Lyon's keynote, Dr. Justin Brunelle opened the Web archives paper session with a presentation for a full paper titled, "Archival Crawlers and JavaScript: Discover More Stuff but Crawl More Slowly." In this presentation, he discussed the challenges Web archives face in crawling pages with deferred representations due to JavaScript, and proposed a method for discovering and archiving deferred representations and their respective descendants which are only visible from the client.
Next, Faryaneh Poursardar presented a short paper - "What is Part of that Resource? User Expectations for Personal Archiving," where she talked about the difficulty users face in deciding the answer to the question: What is part of and what is not part of an Internet resource? She also explored various user perception of this question and its implications on personal archiving.
Next, Dr. Weijia Xu presented a short paper - "A Portable Strategy for Preserving Web Applications and Data Functionality". Dr. Xu proposed a preservation strategy for decoupling web applications and from data and the hosting environment in order to improve reproducibility and portability of the applications across different platforms over time.
Sawood Alam was scheduled to present his short paper titled: "Client-side Reconstruction of Composite Mementos Using ServiceWorker," but his flight was cancelled the previous day, delaying his arrival until after the paper session. 
Dr. Nelson presented the paper on his behalf, and discussed the use of ServiceWorker (SW) web API to help archival replay systems avoid the problem of incorrect URI references due to URL rewriting, by strategically rerouting HTTP requests from embedded resources instead of rewriting URLs.
The conference continued with the second paper session (Semantics and Linking) after a break. This session consisted of a pair of full paper presentations followed by a pair of short paper presentations.
First, Pavlos Fafalios presented - "Building and Querying Semantic Layers for Web Archives," which was also a Vannevar Bush Best Paper Nominee. Pavlos Fafalios proposed a means to improve the use of web archives. He highlighted the lack of efficient and meaningful methods for exploring web archives, and proposed an RDF/S model and distributed framework that describes semantic information about the content of web archives.
Second, Abhik Jana presented "WikiM: Metapaths based Wikification of Scientific Abstracts" - a method of wikifying scientific publication abstracts - in order to effectively help readers decide whether to read the full articles. 
Third, Dr. Jian Wu presented "HESDK: A Hybrid Approach to Extracting Scientific Domain Knowledge Entities." Dr. Jian Wu presented a variant of automatic keyphrase extraction called Scientific Domain Knowledge Entity (SDKE) extraction. Unlike keyphrases (important noun phrases of a document), SDKEs refer to a span of text which represents a concept which can be classified as a process, material, task, dataset etc.
Fourth, Xiao Yang presented "Smart Library: Identifying Books in a Library using Richly Supervised Deep Scene Text" - a library inventory building/retrieval system based on scene text reading methods, which has the potential of reducing the manual labor required to manage book inventories.
The third paper session (Collection Access and Indexing) began with Martin Toepfer's presentation of his full paper (Vannevar Bush Best Paper Nominee) titled: "Descriptor-invariant Fusion Architectures for Automatic Subject Indexing: Analysis and Empirical Results on Short Texts." He discussed the need for digital libraries to automatically index documents accurately especially considering concept drift and amid a rapid increase in content such as scientific publication. Martin Toepfer also discussed the approaches for automatically indexing as a means to help researchers and practitioners in digital libraries decide the appropriate methods for automatic indexing. Next, Guillaume Chiron, presented his short paper titled: "Impact of OCR errors on the use of digital libraries. Towards a better access to information." He discussed his research to estimate the impact of OCR errors on the use of the Gallica Digital Library from the French National Library, and proposed a means for predicting the relative mismatch between queried terms and the target resources due to OCR errors.
Next,  Dr. Kevin Page presented a short paper titled: "Information-Seeking in Large-Scale Digital Libraries: Strategies for Scholarly Workset Creation." He discussed his research which examined the information-seeking models ('worksets') proposed by the HathiTrust Research Center for research into the 15 million volumes of HathiTrust content. This research also involved assessing whether the information-seeking models effectively capture emergent user activities of scholarly investigation.
Next, Dr. Peter Darch presented a short paper titles: "Uncertainty About the Long-Term: Digital Libraries, Astronomy Data, and Open Source Software." Dr. Darch talked about the uncertainty Digital Library developers experience when designing and implementing Digital libraries by presenting the case study of building the Large Synoptic Survey Telescope (LSST) Digital Library.
The third paper session was concluded with a short paper presentation from Jaimie Murdock titled: "Towards Publishing Secure Capsule-based Analysis," in which he discussed recent advancements in providing aid to HTDL (HathiTrust Digital Library) researchers who intend to publish there results from Big Data analysis from HTDL. The advancements include provenance, workflows, worksets, and non-consumptive exports.
After the Day 2 paper sessions, Dr. Nelson conducted the JCDL plenary community meeting in which attendees where given the opportunity to give feedback to improve the conference. The plenary community meeting was followed by Minute Madness - a session in which authors of posters had one minute to convince the audience to visit their poster stands.
The Minute Madness gave way to the poster session and a reception followed. 
Day 3 (June 21)
Day 3 started with a keynote from Dr. Raymond Siemens, in which he discussed the ways social scholarship framing of the production, accumulation, organization, retrieval, and navigation of knowledge, encourages building knowledge to scale in a Humanistic context.
Following the keynote, the fourth paper session (Citation Analysis) began with a prerecorded full paper (Vannevar Bush Best Paper Nominee) presentation from Dr. Saeed-Ul Hassan titled: "Identifying Important Citations using Contextual Information from Full Text," in which he addressed the problem of classifying cited work into important and non-important classes with respect to the developments presented in a research publication, as an important step for algorithms designed to track emerging research topics. Next, Luca Weihs presented a full paper titled: "Learning to Predict Citation-Based Impact Measures." He presented non-linear probabilistic techniques for predicting the future scientific impact of impact of a research paper. This is unlike linear probabilistic methods which focus on understanding the past and present impact of a paper. The final full paper presentation from this session was titled: "Understanding the Impact of Early Citers on Long-Term Scientific Impact" and presented by Mayank Singh. Mayank Singh presented his investigation to see if the set of authors who cite a paper early (within 1-2 years), affect the paper's Long-Term Scientific Impact (LTSI). In his research he discovered that influential early citers negatively affect LTSI probably due to "attention stealing."
The conference continued with fifth paper session (Exploring and Analyzing Collections) consisting of three full paper presentations. The first (Student Paper Award Nominee), titled: "Matrix-based News Aggregation: Exploring Different News Perspectives," was presented by Norman Meuschke. He presented NewsBird, Matrix-based News Analysis system (MNA) which help users see news from various perspectives, as a means to help avoid a biased news consumption.
The second paper (Vannevar Bush Best Paper Nominee), titled: "Quill: A Framework for Constructing Negotiated Texts - with a Case Study on the US Constitutional Convention of 1787," was presented by Dr. Nicholas Cole, who presented the Quill framework. Quill is a new approach to present and study formal negotiation records such as creation of constitutions, treaties, and legislation. Quill currently hosts the records of the Constitutional Convention of 1787 that wrote the Constitution of the United States.
The final presentation for this session was from Dr. Kevin Page, titled: "Realising a Layered Digital Library: Exploration and Analysis of the Live Music Archive through Linked Data," in which he discussed his research which followed a Linked Data approach to build a layered Digital Library, utilizing content form the Internet Archive's Live Music Archive.
The sixth paper session (Text Extraction and Analysis) consisted of three full paper presentation. The first, titled: "A Benchmark and Evaluation for Text Extraction," was presented by Dr. Hannah Bast. Dr. Bast highlighted the difficulty of extracting text from PDF documents due to the fact that PDF is a layout-based format which specifies position information of characters rather than semantic information (e.g., body text or footnote). She also presented her evaluation result of 13 state of the art tools for extracting text from PDF. She showed that her method Icecite outperformed other tools, but is not perfect, and outlined the steps necessary to make text extraction from PDF a solved problem. Next, Kresimir Duretec presented "A text extraction software benchmark based on a synthesized dataset." To help text data processing workflows in digital libraries, he described a dataset generation method based on model driven engineering principles and use it to synthesize a dataset and its ground truth directly from a model. He also presented a benchmark for text extraction tools. This paper session was concluded with a presentation by Tokinori Suzuki titled: "Mathematical Document Categorization with Structure of Mathematical Expressions." He presented his research in Mathematical Document Categorization (MDC) - a task of classifying mathematical documents into mathematical categories such as Probability theory and Set theory. He proposed a classification method that uses text and structures of mathematical expressions. 
The seventh paper session (Collection Building) consisted of three full paper presentation, and began with Dr. Federico Nanni's presentation (Best Student Paper Award Nominee) titled: "Building Entity-Centric Event Collections." Federico Nanni presented an approach that utilizes large web archives to build event-centric sub-collections consisting of core documents related to the events as well as documents associated with the premise and consequences of events.
Next, Jan R. Benetka, presented a paper titled: "Towards Building a Knowledge Base of Monetary Transactions from a News Collection," where he  addressed the problem of extracting structured representations of economic events (e.g., large company buyouts) from a large corpus of news articles. He presented a method which combines natural language processing and machine learning techniques to address this task.
I concluded the seventh paper session with a presentation titled: "Local Memory Project: providing tools to build collections of stories for local events from local sources". In this presentation, I discussed the need to expose local media sources, and introduced two tools under the umbrella of the Local Memory Project. The first tool - Geo, helps users discover nearby local news media sources such as newspapers, TV, and radio stations. The next - a Collection building tool, helps users build, save, share, and archive collections of local events from local sources for US and non-US media sources.
Here are the slides I presented:
The eighth paper session (Classification and Clustering) occurred in parallel with the sixth paper session. It consisted of a pair of full papers and a pair of short papers. The first paper titled: "Classifying Short Unstructured Data using the Apache Spark Platform," was presented by Saurabh Chakravarty. Saurabh Chakravarty highlighted the difficulty traditional classifiers have in classifying tweets. This difficulty is partly due to the shortness of tweets, and the presence of abbreviations, hashtags, emojis, and non-standard usage of written language. Consequently, he proposed the used of the Spark platform to implement two shot text classification strategies. He also showed these strategies are able to effectively classify millions of text composed of thousands of distinct features and classes. Next, Abel Elekes presented his full paper (Best Student Paper Award Nominee) titled: "On the Various Semantics of Similarity in Word Embedding Models," in which he discussed results running two experiments to determine when exactly similarity scores of word embedding model is meaningful. He proposed that his method could provide a better understanding of the notion of similarity in embedding models and improve the the evaluation of such models. Next, Mirco Kocher presented his short paper titled: "Author Clustering Using Spatium." Mirco Kocher proposed a model for clustering authors after presenting the author clustering problem as it relates to authorship attribution questions. The model he proposed uses a distance measure called Spatium which was derived from weighted version of L1 norm (Canberra measure). He showed that this model evaluation produced high precision and F1 values when tested with a 20 test collection. Finally Shaobin Xu presented a short paper titled: "Retrieving and Combining Repeated Passages to Improve OCR." He presented a new method to improve the output of Optical Character Recognition (OCR) systems. The method begins with detecting duplicate passages, then it performs a consensus decoding which is combined with a language model.
The ninth paper session (Content Provenance and Reuse), began with Dr. David Bamman full paper presentation titled: "Estimating the Date of First Publication in a Large-Scale Digital Library." Dr. David Bamman discussed his finding from evaluating methods for approximating date of first publication. The methods considered (and used in practice) include: using the date of publication from available metadata, multiple deduplication methods, and automatically predicting the date of composition from text of the book. He found that using a simple heuristic of metadata-based deduplication performs best in practice.
Dr. George Buchanan presented his full paper titled: "The Lowest form of Flattery: Characterising Text Re-use and Plagiarism Patterns in a Digital Library Corpus," in which he discussed a first assessment of text re-use (plagiarism) for the digital libraries domain, and suggested measures for more rigorous plagiarism detection and management.
Next, Corinna Breitinger presented her short paper titled: "CryptSubmit: Introducing Securely Timestamped Manuscript Submission and Peer Review Feedback using the Blockchain." She introduced CryptSubmit as a means to address the fear researchers have that their work may be leaked or plagiarized by a program committee or anonymous peer reviewers. CryptSubmit utilizes the decentralized Bitcoin blockchain to establish trust and verifiability by creating a publicly verifiable and tamper-proof timestamp for manuscript.
Next, Mayank Singh a short paper titled: "Citation sentence reuse behavior of scientists: A case study on massive bibliographic text dataset of computer science." He proposed a new model of conceptualizing plagiarism in scholarly research based on reuse of explicit citation sentences in scientific research articles, which is unlike traditional plagiarism detection which uses text similarity. He provided examples of plagiarism and revealed that this practice is widespread even for well known researchers.
A conference banquet at Sassafraz Restaurant followed the last paper session of the day.
During the banquet, awards for best poster, best student paper, and the Vannevar Bush best paper award, were given.  Sawood Alam received the most votes for his poster - Impact of URI Canonicalization on Memento Count - thus, received the award for best poster. Felix Hamborg, Norman Meuschke, and Dr. Bella Gipp, received the best student paper award for: "Matrix-based News Aggregation: Exploring Different News Perspectives." Finally, Dr. Nicholas Cole, Alfie Abdul-Rahman, and Grace Mallon received the Vannevar Bush best paper award for "Quill: A Framework for Constructing Negotiated Texts - with a Case Study on the US Constitutional Convention of 1787."
Day 4 (June 22)
Day four of the conference began with a panel session titled: "Can We Really Show This?: Ethics, Representation and Social Justice in Sensitive Digital Space," in which ethical issues experienced by curators who work with sensitive and contentious content from marginalized populations was addressed. The panel consisted of Deborah Maron (Moderator), and the following speakers: Dorothy Berry, Raegan Swanson, and Erin White.
The tenth and last paper session (Scientific Collections and Libraries) followed and consisted of three full paper presentations. First, Dr. Abdussalam Alawini, presented a paper titled: "Automating data citation: the eagle-i experience," in which he highlighted the growing concern of giving credit to contributors and curators of datasets. He presented his research in automating citation generation for an RDF dataset called eagle-i, and discussed a means to generalize this citation framework across a variety of different types of databases. Next, Sandipan Sikdar presented "Influence of Reviewer Interaction Network on Long-term Citations: A Case Study of the Scientific Peer-Review System of the Journal of High Energy Physics" (Best Student Paper Award Nominee). He presented his research which sought to answer the question: "Could the peer review system be improved?" amid a consensus from the research community that it is indispensable but flawed. His research attempted to answer this question by introducing a new reviewer-reviewer interaction network, showing that structural properties of this network surprisingly serve as strong predictors of the long-term citations of a submitted paper. Finally Dr. Martin Klein, presented: "Discovering Scholarly Orphans Using ORCID". Dr. Martin Klein proposed a new paradigm for archiving scholarly orphans - web-native scholarly objects that are largely neglected by current archival practices. He presented his research which investigated the feasibility of using Open Researcher and Contributor ID (ORCID) as a means for discovering the web identities and scholarly orphans for active researchers.
Here are the slides he presented:
Dr. Salvatore Mele gave the keynote of the day. He discussed the significant impact Preprints have had on research such has the High-Energy Physics domain which has benefited from a rich Preprint culture for more than half a century. He also reported on the results of two studies that aimed to assess the coexistence and complementarity between Preprints and academic journals that are less open.
The 2017 JCDL conference officially concluded with Dr. Ed Fox's announcement of the 2018 JCDL conference to be held at the University of North Texas. 

No comments:

Post a Comment