Predicting Temporal Intention in Resource Sharing

SalahEldeen, Hany M.; Nelson, Michael L.

doi:10.1145/2756406.2756921

Cited by 4 publications

(5 citation statements)

References 26 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We can always undersample major SCs, but this means we have to reduce sample sizes of all SCs down to about 15 (Art; Section 5 ), which is too small for training robust neural network models. The oversampling strategies such as SMOTE ( Chawla et al, 2002 ) works for problems involving continuous numerical quantities, e.g., SalahEldeen and Nelson (2015) . In our case, the synthesized vectors of “abstracts” by SMOTE will not map to any actual words because word representations are very sparsely distributed in the large WE space.…”

Section: Discussionmentioning

confidence: 99%

Large Scale Subject Category Classification of Scholarly Papers With Deep Attentive Neural Networks

Kandimalla

Rohatgi

et al. 2021

Front. Res. Metr. Anal.

View full text Add to dashboard Cite

Subject categories of scholarly papers generally refer to the knowledge domain(s) to which the papers belong, examples being computer science or physics. Subject category classification is a prerequisite for bibliometric studies, organizing scientific publications for domain knowledge extraction, and facilitating faceted searches for digital library search engines. Unfortunately, many academic papers do not have such information as part of their metadata. Most existing methods for solving this task focus on unsupervised learning that often relies on citation networks. However, a complete list of papers citing the current paper may not be readily available. In particular, new papers that have few or no citations cannot be classified using such methods. Here, we propose a deep attentive neural network (DANN) that classifies scholarly papers using only their abstracts. The network is trained using nine million abstracts from Web of Science (WoS). We also use the WoS schema that covers 104 subject categories. The proposed network consists of two bi-directional recurrent neural networks followed by an attention layer. We compare our model against baselines by varying the architecture and text representation. Our best model achieves micro-F1 measure of 0.76 with F1 of individual subject categories ranging from 0.50 to 0.95. The results showed the importance of retraining word embedding models to maximize the vocabulary overlap and the effectiveness of the attention mechanism. The combination of word vectors with TFIDF outperforms character and sentence level embedding models. We discuss imbalanced samples and overlapping categories and suggest possible strategies for mitigation. We also determine the subject category distribution in CiteSeerX by classifying a random sample of one million academic papers.

show abstract

Section: Discussionmentioning

confidence: 99%

Large Scale Subject Category Classification of Scholarly Papers With Deep Attentive Neural Networks

Kandimalla

Rohatgi

et al. 2021

Front. Res. Metr. Anal.

View full text Add to dashboard Cite

show abstract

“…In addition, Twitter contains not only texts but also unique features such as hashtags, followers and followees (i.e., Twitter users who follow or are followed by a particular user), and URLs. Using these features, past studies focused on (among others) automatic hashtag labeling by hashtag-based pooling tweets [43], analyzing factors affecting response [15], readability of crisis communications [56], language diversity [41], language and locations [59], detecting influencers in Twitter [60], classifying user's temporal intention when sharing resources [51], ranking users [58] or meme tracking in blogosphere [40].…”

Section: Twitter Analysismentioning

confidence: 99%

Analyzing history-related posts in twitter

Sumikawa

Jatowt

2020

Int J Digit Libr

View full text Add to dashboard Cite

Microblogging platforms such as Twitter have been increasingly used nowadays to share information between users. They are also convenient means for propagating content related to history. Hence, from the research viewpoint they can offer opportunities to analyze the way in which users refer to the past, and how as well when such references appear and what purposes they serve. Such study could allow to quantify the interest degree and the mechanisms behind content dissemination. We report the results of a large scale exploratory analysis of history-oriented posts in microblogs based on a 28-month-long snapshot of Twitter data. The results can increase our understanding of the characteristics of history-focused content sharing in Twitter. They can also be used for guiding the design of content recommendation systems as well as time-aware search applications.

show abstract

“…For each URI-R, a pair of mementos consisting of m 0 and one of the four categories of m 1 were evaluated by five turkers for a total of 280 evaluations. We follow the precedent of using five turkers to establish turker opinion as established by SalahEldeen and Nelson [35].…”

Section: Users' Perception Of Damagementioning

confidence: 99%

Not all mementos are created equal: measuring the impact of missing resources

et al. 2015

Self Cite

View full text Add to dashboard Cite

Web archives do not always capture every resource on every page that they attempt to archive. This results in archived pages missing a portion of their embedded resources. These embedded resources have varying historic, utility, and importance values. The proportion of missing embedded resources does not provide an accurate measure of their impact on the Web page; some embedded resources are more important to the utility of a page than others. We propose a method to measure the relative value of embedded resources and assign a damage rating to archived pages as a way to evaluate archival success. In this paper, we show that Web users' perceptions of damage are not accurately estimated by the proportion of missing embedded resources. In fact, the proportion of missing embedded resources is a less accurate estimate of resource damage than a random selection. We propose a damage rating algorithm that provides closer alignment to Web user perception, providing an overall improved agreement with users on memento damage by 17 % and an improvement by 51 % if the mementos have a damage rating delta >0.30. We use our algorithm to measure damage in the Internet Archive, showing that it is getting better at mitigating damage over time (going from a damage ). However, we show that a greater number of important embedded resources (2.05 per memento on average) are missing over time. Alternatively, the damage in WebCite is increasing over time (going from 0.375 in 2007 to 0.475 in 2014), while the missing embedded resources remain constant (13 % of the resources are missing on average). Finally, we investigate the impact of JavaScript on the damage of the archives, showing that a crawler that can archive JavaScript-dependent representations will reduce memento damage by 13.5 %.

show abstract

Predicting Temporal Intention in Resource Sharing

Cited by 4 publications

References 26 publications

Large Scale Subject Category Classification of Scholarly Papers With Deep Attentive Neural Networks

Large Scale Subject Category Classification of Scholarly Papers With Deep Attentive Neural Networks

Analyzing history-related posts in twitter

Not all mementos are created equal: measuring the impact of missing resources

Contact Info

Product

Resources

About