Article
Computer Science, Artificial Intelligence
Hao Peng, Ruitong Zhang, Shaoning Li, Yuwei Cao, Shirui Pan, Philip S. Yu
Summary: This paper presents a novel reinforced, incremental, and cross-lingual social event detection architecture, FinEvent, which models social messages into heterogeneous graphs and uses reinforcement learning algorithm to select optimal aggregation thresholds. It addresses the challenges of ambiguous event features, dispersive text contents, and multiple languages in existing event detection methods for streaming social messages, thereby improving accuracy and generalization ability.
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE
(2023)
Article
Computer Science, Artificial Intelligence
Xinchen Shi, Bin Li, Ling Chen, Chao Yang
Summary: The objective of this paper is to identify equivalent entities in knowledge graphs across different languages. The authors propose integrating entities and attributes into a unified graph and design a novel Bi-Neighborhood Graph Neural Network (BNGNN) to model the graph. Experimental results demonstrate that the BNGNN model outperforms existing EA methods on multiple datasets.
KNOWLEDGE-BASED SYSTEMS
(2023)
Article
Computer Science, Information Systems
Rizka W. Sholikah, Agus Z. Arifin, Chastine Fatichah, Ayu Purwarianti
Summary: Semantic relation detection is crucial in natural language processing. This paper proposes a framework for cross-lingual semantic relation identification using multi-task learning and a general vector space, which overcomes the limitation of cross-lingual data and improves relation identification accuracy.
JOURNAL OF KING SAUD UNIVERSITY-COMPUTER AND INFORMATION SCIENCES
(2022)
Article
Computer Science, Information Systems
Hanqian Wu, Zhike Wang, Feng Qing, Shoushan Li
Summary: This paper proposes a method for conducting Cross-Lingual Aspect Sentiment Classification (CLASC) task by leveraging rich resources in a source language for aspect sentiment classification in a target language, using bilingual lexicon and machine translation tools for sentence translation. The proposed approach combines Reinforced Transformer with Cross-Lingual Distillation to mitigate the undesirable effects of translation ambiguities in different target languages, outperforming existing methods in experimental results.
Article
Computer Science, Information Systems
Wenjian Dong, Mayu Otani, Noa Garcia, Yuta Nakashima, Chenhui Chu
Summary: This paper introduces the first work on cross-lingual visual grounding, expanding the task to different languages by constructing a dataset for French and proposing a method to transfer knowledge from an English model to a French model. Despite the smaller size of the French dataset compared to the English dataset, experiments show that the model achieves an accuracy of 65.17%, comparable to the English model's 69.04%.
Article
Acoustics
Yingwen Fu, Nankai Lin, Boyu Chen, Ziyu Yang, Shengyi Jiang
Summary: Previous works on cross-lingual Named Entity Recognition (NER) have achieved great success, but few of them consider the effect of language families on the performance. This study finds that the NER performance of a target language decreases when its source language belongs to a different language family. To address this issue, a novel cross-lingual NER framework, SD-BBN, is proposed. SD-BBN learns source-language NER knowledge from supervised datasets and target-language knowledge from weakly supervised datasets, and achieves better performance by fusing these two kinds of knowledge using self-distillation mechanism. Evaluation on 9 language datasets shows that SD-BBN outperforms baseline methods, especially when the source and target languages are from different language families. This finding suggests that obtaining language-specific knowledge from the target language is crucial for improving cross-lingual NER.
IEEE-ACM TRANSACTIONS ON AUDIO SPEECH AND LANGUAGE PROCESSING
(2023)
Article
Automation & Control Systems
Deqing Wang, Junjie Wu, Jingyuan Yang, Baoyu Jing, Wenjie Zhang, Xiaonan He, Hui Zhang
Summary: This article introduces a semisupervised transfer learning approach called ssSCL-ST, which uses structural correspondence learning and space transfer to improve cross-lingual sentiment analysis, aiming to explore intrinsic sentiment knowledge in the target language domain and enhance classification accuracy.
IEEE TRANSACTIONS ON CYBERNETICS
(2022)
Article
Chemistry, Multidisciplinary
Luis Espinosa-Anke, Geraint Palmer, Padraig Corcoran, Maxim Filimonov, Irena Spasic, Dawn Knight
Summary: Cross-lingual embeddings are vector representations where word translations are closely located together, enabling transfer learning across languages. By combining Welsh and English corpora and using bilingual dictionaries and alignment strategies, effective bilingual dictionary induction and cross-lingual sentiment analysis are achieved in this paper. The best results are obtained using monolingual fastText embeddings and the CSLS metric, and including automatically translated training documents can significantly improve the performance of cross-lingual text classifiers for Welsh.
APPLIED SCIENCES-BASEL
(2021)
Article
Business
Chia-Hsuan Chang, Christopher C. Yang
Summary: Information search is a profitable activity in electronic commerce. Consumer concerns about search quality arise as search engines become the primary source for health information. The vocabulary difference between general consumers and professionals leads to unsatisfactory retrieval performance and misleading information, which becomes more complex when searching health information across languages. To address this issue, a cross-lingual retrieval framework is proposed, incorporating semantic space construction and information retrieval modules. This framework utilizes a weakly-supervised approach to determine a cross-lingual word space (CLWS) and suggests strategies that utilize the translations of CLWS for improved retrieval performance.
ELECTRONIC COMMERCE RESEARCH AND APPLICATIONS
(2023)
Article
Computer Science, Artificial Intelligence
Dong Zhou, Xiaoya Peng, Lin Li, Jun-mei Han
Summary: This work focuses on generating effective Cross-Lingual Embeddings (CLEs) using auxiliary Topic Models, utilizing both monolingual and bilingual topic models to generate monolingual embedding spaces and seed dictionaries for projection. Through comprehensive evaluation using bilingual lexicon extraction, cross-lingual semantic word similarity, and cross-lingual document classification tasks, the proposed model outperforms existing supervised and unsupervised CLE models while surpassing CLE models generated from representative monolingual topical word embeddings.
EXPERT SYSTEMS WITH APPLICATIONS
(2022)
Article
Computer Science, Artificial Intelligence
Mirko Agarla, Simone Bianco, Luigi Celona, Paolo Napoletano, Alexey Petrovsky, Flavio Piccoli, Raimondo Schettini, Ivan Shanin
Summary: Thanks to deep learning techniques, performance in Speech Emotion Recognition(SER) on a single language has greatly increased in recent years. However, cross-lingual SER remains a challenge in real world applications due to the big gap among source and target domain distributions and the availability of unlabeled utterances. To address this, a Semi-Supervised Learning (SSL) method based on Transformer is proposed, which adapts to the new domain by exploiting pseudo-labeling strategy on the unlabeled utterances, achieving robustness across five languages and significant improvement in unweighted accuracy compared to state-of-the-art methods.
EXPERT SYSTEMS WITH APPLICATIONS
(2024)
Article
Computer Science, Artificial Intelligence
Liyang Xie, Zhongcheng Wu, Xian Zhang, Yong Li
Summary: This paper proposes a novel Federated Bert Network (FBN) that combines Bidirectional Encoder Representations from Transformers (Bert) with a Federated Learning (FL) framework. The FBN model employs a Length Alignment Algorithm to handle sequence length variations and allows independent learning of local models on different clients. The server uses an improved Federated Average Algorithm with Reward-Punishment Mechanism (FedAvgRP) to aggregate the local models and generate a global model. The FBN model outperforms state-of-the-art methods in both random and skilled forgery scenarios, and maintains high performance in the face of data attacks.
PATTERN RECOGNITION
(2023)
Article
Computer Science, Artificial Intelligence
Han Zhang, Suyi Yang, Hongqing Zhu
Summary: This paper proposes a cross-lingual pre-training method to adapt low-resource languages and enable the transfer of text-to-image generation ability between different languages.
KNOWLEDGE-BASED SYSTEMS
(2022)
Article
Computer Science, Information Systems
Jianliang Gao, Xiangyue Liu, Yibo Chen, Fan Xiong
Summary: This paper proposes a new embedding-based framework called MHGCN, which considers entity alignment from multiple views and utilizes a highway graph convolutional network for entity embedding. The framework weights and fuses the multiple views to obtain a better entity embedding. The experimental results demonstrate that MHGCN outperforms state-of-the-art alignment methods.
TSINGHUA SCIENCE AND TECHNOLOGY
(2022)
Article
Computer Science, Information Systems
Qingran Zhan, Xiang Xie, Chenguang Hu, Juan Zuluaga-Gomez, Jing Wang, Haobo Cheng
Summary: This paper investigates the extraction of reliable AFs using a DANN for cross-lingual speech recognition. By training AFs detectors in source languages and transferring phonological knowledge to the target language, along with the fusion of acoustic features and cross-lingual AFs using multi-stream techniques, improved performance is achieved. The experiments show that using CNN with domain-adversarial learning and the MHA-based multi-stream approach yield significant improvements in performance compared to other methods, especially when considering low-resource languages.