Leveraging the meta-embedding for text classification in a resource-constrained language

Article Economics

Optimal model averaging based on forward-validation

Xiaomeng Zhang et al.

Summary: In this paper, we propose a frequentist model averaging method based on forward-validation, considering the uncertainty of sample size and establishing its asymptotic optimality in achieving the lowest possible prediction risk.

JOURNAL OF ECONOMETRICS (2023)

添加到收藏夹

Article Computer Science, Artificial Intelligence

CovTiNet: Covid text identification network using attention-based positional embedding feature fusion

Md. Rajib Hossain et al.

Summary: Covid text identification is crucial for controlling societal distrust and panic caused by the infodemic of uninformative texts containing misinformation, disinformation, and malinformation. While little research has been reported in high-resource languages, this paper proposes a deep learning-based network called CovTiNet to identify Covid text in the low-resource language of Bengali, achieving a high accuracy rate of 96.61 +/- .001%.

NEURAL COMPUTING & APPLICATIONS (2023)

添加到收藏夹

Article Automation & Control Systems

Transfer-based taxonomy induction over concept labels

Mayank Kejriwal et al.

Summary: This paper presents a principled approach for taxonomy induction in the e-commerce domain by utilizing a pre-trained language representation learning model and examples of other taxonomies. The proposed method outperforms seven different baselines, including the transformer-based RoBERTa model, on three widely used e-commerce concept-sets.

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE (2022)

添加到收藏夹

Article Automation & Control Systems

Microblog sentiment analysis via embedding social contexts into an attentive LSTM

Yang Jing et al.

Summary: This paper proposes a deep learning method to fully capture the features of microblog relations and enhance sentiment analysis by considering both implicit and explicit features. By constructing a graph to model microblog relations and embedding it, a novel neural network is designed to integrate social context knowledge with text information. The attention mechanism is introduced to handle different contributions of words to the classification results. Experimental results demonstrate that the proposed model consistently and significantly outperforms existing state-of-the-art methods.

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE (2021)

添加到收藏夹

Article Computer Science, Artificial Intelligence

Bengali text document categorization based on very deep convolution neural network

Md. Rajib Hossain et al.

Summary: The amount of digital text contents in Bengali language has increased on online platforms, leading to the proposal of an intelligent text classification system to effectively manage this data. Challenges like resource constraints and morphological variants in Bengali text classification were addressed by using GloVe embedding and VDCNN classifier. The experimental results showed that the proposed model outperformed other models and achieved the highest accuracy in Bengali text classification.

EXPERT SYSTEMS WITH APPLICATIONS (2021)

添加到收藏夹

Article Automation & Control Systems

A multimodal deep fusion graph framework to detect social distancing violations and FCGs in pandemic surveillance

Elizabeth B. Varghese et al.

Summary: This paper proposes a deep learning framework that integrates data from multiple sensor modalities to detect social distance violations and Free-standing Conversation Groups (FCGs). The research shows that the proposed approach excels at analyzing the potential risk of pandemic spread and effectively calculates violation scores and rates.

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE (2021)

添加到收藏夹

Article Automation & Control Systems

Transformer based network for Open Information Extraction

Jiabao Han et al.

Summary: Open Information Extraction is an important area in Natural Language Processing, but existing strategies face challenges such as high human resource requirements and potential error accumulation. This paper proposes an Open IE approach based on the Transformer architecture, showing better performance in experiments compared to existing baselines.

ENGINEERING APPLICATIONS OF ARTIFICIAL INTELLIGENCE (2021)

添加到收藏夹

Article Computer Science, Information Systems

Authorship Classification in a Resource Constraint Language Using Convolutional Neural Networks

Md. Rajib Hossain et al.

Summary: Authorship classification has significantly progressed in high-resource languages, but remains at a primitive stage in resource-constraint languages like Bengali. This study presents an approach using Convolution Neural Networks for Bengali authorship classification, developing new embedding and classification corpora to improve performance. Multiple embedding models and classification techniques were evaluated, with the optimized CNN model achieving high accuracies in classification tasks.

IEEE ACCESS (2021)

添加到收藏夹

Article Computer Science, Information Systems