A5053537544- Topic Modeling
- Advanced Text Analysis Techniques
- Natural Language Processing Techniques
- Web Data Mining and Analysis
- Wikis in Education and Collaboration
- Text and Document Classification Technologies
- Recommender Systems and Techniques
- Machine Learning and Algorithms
- Expert finding and Q&A systems
- Information Retrieval and Search Behavior
- Service-Oriented Architecture and Web Services
- Mobile Crowdsensing and Crowdsourcing
- Bayesian Modeling and Causal Inference
- AI-based Problem Solving and Planning
- Control and Dynamics of Mobile Robots
- Robotic Path Planning Algorithms
- Image Enhancement Techniques
- Advanced Graph Neural Networks
- Software System Performance and Reliability
- Algorithms and Data Compression
- Text Readability and Simplification
- Data Mining Algorithms and Applications
- Data Quality and Management
- Video Surveillance and Tracking Methods
- Innovative Educational Techniques
Meta (United States)
2023-2024
Carnegie Mellon University
2023
Chinese University of Hong Kong
2023
Google (United States)
2023
De Anza College
2023
Jimei University
2022
Cambridge University Press
2022
New York University Press
2022
University of Science and Technology of China
2022
Zhejiang University
2021
Mapping out the challenges and strategies for widespread adoption of service computing.
Automatic citation recommendation can be very useful for authoring a paper and is an AI-complete problem due to the challenge of bridging semantic gap between context cited paper. It not always easy knowledgeable researchers give accurate or find right cite given context. To help with this problem, we propose novel neural probabilistic model that jointly learns representations contexts papers. The probability citing estimated by training multi-layer network. We implement evaluate our on...
A prerequisite relation describes a basic among concepts in cognition, education and other areas.However, as semantic relation, it has not been well studied computational linguistics.We investigate the problem of measuring relations propose simple link-based metric, namely reference distance (RefD), that effectively models by how differently two refer to each other.Evaluations on datasets include seven domains show our single metric based method outperforms existing supervised learning methods.
Proposal of large-scale datasets has facilitated research on deep neural models for news summarization.Deep learning can also be potentially useful spoken dialogue summarization, which benefit a range reallife scenarios including customer service management and medication tracking.To this end, we propose DIALOGSUM, labeled summarization dataset.We conduct empirical analysis DIALOGSUM using state-of-the-art summarizers.Experimental results show unique challenges in such as terms, special...
We present a framework for constructing specific type of knowledge graph, concept map from textbooks. Using Wikipedia, we derive prerequisite relations among these concepts. A traditional approach extraction consists two sub-problems: key and relationship identification. Previous work the most part had considered sub-problems independently. propose that jointly optimizes investigates methods identify relationships. Experiments on maps are manually extracted in six educational areas (computer...
Scene text detection has attracted great attention these years. Text potentially exist in a wide variety of images or videos and play an important role understanding the scene. In this paper, we present novel algorithm which is composed two cascaded steps: (1) multi-scale fully convolutional neural network (FCN) proposed to extract block regions, (2) instance (word line) aware segmentation designed further remove false positives obtain word instances. The can accurately localize line...
Prerequisite relations among concepts play an important role in many educational applications such as intelligent tutoring system and curriculum planning. With the increasing amount of data available, automatic discovery concept prerequisite has become both emerging research opportunity open challenge. Here, we investigate how to recover from course dependencies propose optimization based framework address problem. We create first real dataset for empirically studying this problem, which...
Concept prerequisite learning focuses on machine methods for measuring the relation among concepts. With importance of prerequisites education, it has recently become a promising research direction. A major obstacle to extracting at scale is lack large-scale labels which will enable effective data-driven solutions. We investigate applicability active concept learning.We propose novel set features tailored classification and compare effectiveness four widely used query strategies....
Distributed representations of words have been shown to capture lexical semantics, based on their effectiveness in word similarity and analogical relation tasks. But, these tasks only evaluate semantics indirectly. In this paper, we study whether it is possible utilize distributed generate dictionary definitions words, as a more direct transparent representation the embeddings' semantics. We introduce definition modeling, task generating for given its embedding. present different model...
We investigate how machine learning models, specifically ranking can be used to select useful distractors for multiple choice questions. Our proposed models learn that resemble those in actual exam questions, which is different from most existing unsupervised ontology-based and similarity-based methods. empirically study feature-based neural net (NN) based with experiments on the recently released SciQ dataset our MCQL dataset. Experimental results show ensemble methods (random forest...
Concept hierarchies have been useful tools for presenting and organizing knowledge. With the rapid growth in number of online knowledge resources, automatic concept hierarchy extraction is increasingly attractive. Here, we focus on from textbooks based Wikipedia. Given a book, extract important concepts each book chapter using Wikipedia as resource this construct that book. We define local global features capture both relatedness coherence embedded textbook. In order to evaluate proposed...
Distractor generation is a crucial step for fill-in-the-blank question generation. We propose generative model learned from training adversarial nets (GANs) to create useful distractors. Our method utilizes only context information and does not use the correct answer, which completely different previous Ontology-based or similarity-based approaches. Trained on Wikipedia corpus, proposed able predict Wiki entities as evaluated two biology datasets collected actual college-level exams....
In this work, we present a novel approach for Chinese Ink-and-Wash style transfer using GAN structure. The proposed method incorporates specially designed smooth loss tailored task, and an end-to-end framework that seamlessly integrates various components efficient effective image transferring. To demonstrate the superiority of our approach, comparative results against other popular methods such as CycleGAN is presented. experimentation showcased notable improvements achieved with in terms...
Large language models (LLMs) outperform information retrieval techniques for downstream knowledge-intensive tasks when being prompted to generate world knowledge. However, community concerns abound regarding the factuality and potential implications of using this uncensored In light this, we introduce CONNER, a COmpreheNsive kNowledge Evaluation fRamework, designed systematically automatically evaluate generated knowledge from six important perspectives – Factuality, Relevance, Coherence,...
Fast and efficient learning over large bodies of commonsense knowledge is a key requirement for cognitive systems. Semantic web bases provide an important new resource ground facts from which plausible inferences can be learned. This paper applies structured logistic regression with analogical generalization (SLogAn) to make use structural as well statistical information achieve rapid robust learning. SLogAn achieves state-of-the-art performance in standard triplet classification task on two...
Strict partial order is a mathematical structure commonly seen in relational data. One obstacle to extracting such type of relations at scale the lack large-scale labels for building effective data-driven solutions. We develop an active learning framework mining subject strict order. Our approach incorporates reasoning not only finding new unlabeled pairs whose can be deduced from existing label set, but also devising query strategies that consider labels. experiments on concept prerequisite...
As more educational resources become available online, it is possible to acquire up-to-date knowledge and information. We propose BBookX, a novel computer facilitated system that automatically collaboratively builds free open online books using publicly such as Wikipedia. BBookX has two separate components: one creates an version of existing by linking different book chapters Wikipedia articles, while another with interactive user interface supports real-time creation where users are allowed...
Effective user representations are pivotal in personalized advertising. However, stringent constraints on training throughput, serving latency, and memory, often limit the complexity input feature set of online ads ranking models. This challenge is magnified extensive systems like Meta's, which encompass hundreds models with diverse specifications, rendering tailoring representation learning for each model impractical. To address these challenges, we present Scaling User Modeling (SUM), a...
This paper proposes a conceptual hybrid cognitive architecture for robots to learn behaviors from demonstrations in robotic aid situations. Unlike the current architectures, this puts concentration on requirements of safety, interaction, and non-centralized processing Imitation learning technologies have been integrated into rapidly transferring knowledge skills between human teachers robots.
To better organize and understand online news information, we propose Storybase 1 , a knowledge base for events that builds upon Wikipedia current daily Web news.It first constructs stories their timelines based on then detects links to enrich those with more comprehensive events.We encode develop efficient event clustering chaining techniques in an space.We demonstrate search engine helps find historical ongoing inspect dynamic timelines.
We investigate a variant of the problem automatic keyphrase extraction from scientific documents which we define as Scientific Domain Knowledge Entity (SDKE) extraction. Keyphrases are noun phrases important to themselves. In contrasxt, an SDKE is text that refers concept and can be classified process, material, task, dataset etc. A represents domain knowledge, but not necessarily document it in. Supervised algorithms using non-sequential classifiers global measures informativeness (PMI,...