A5102845658- Topic Modeling
- Natural Language Processing Techniques
- Advanced Steganography and Watermarking Techniques
- Data Quality and Management
- Text and Document Classification Technologies
- Multimodal Machine Learning Applications
- Advanced Text Analysis Techniques
- Internet Traffic Analysis and Secure E-voting
- Semantic Web and Ontologies
- Hate Speech and Cyberbullying Detection
- Handwritten Text Recognition Techniques
- Advanced Welding Techniques Analysis
- Web Data Mining and Analysis
- Advanced Image and Video Retrieval Techniques
- Electronic Packaging and Soldering Technologies
- Digital Rights Management and Security
- Adversarial Robustness in Machine Learning
- Recycling and Waste Management Techniques
- Biometric Identification and Security
- Speech and dialogue systems
- Speech Recognition and Synthesis
- Misinformation and Its Impacts
- Vehicle License Plate Recognition
- Biomedical Text Mining and Ontologies
- Extraction and Separation Processes
Tsinghua University
2022-2024
Beijing University of Technology
2022-2024
This paper presents the first comprehensive analysis of ChatGPT's Text-to-SQL ability. Given recent emergence large-scale conversational language model ChatGPT and its impressive capabilities in both abilities code generation, we sought to evaluate performance. We conducted experiments on 12 benchmark datasets with different languages, settings, or scenarios, results demonstrate that has strong text-to-SQL abilities. Although there is still a gap from current state-of-the-art (SOTA)...
Natural Language Inference (NLI) is a growingly essential task in natural language understanding, which requires inferring the relationship between sentence pairs ( <bold xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">premise</b> and xmlns:xlink="http://www.w3.org/1999/xlink">hypothesis</b> ). Recently, low-resource inference has gained increasing attention, due to significant savings manual annotation costs better fit with real-world scenarios....
Cold-start problem is one of the long-standing challenges in recommender systems, focusing on accurately modeling new or interaction-limited users items to provide better recommendations. Due diversification internet platforms and exponential growth items, importance cold-start recommendation (CSR) becoming increasingly evident. At same time, large language models (LLMs) have achieved tremendous success possess strong capabilities user item information, providing potential for However,...
The radioactive nature of Large Language Model (LLM) watermarking enables the detection watermarks inherited by student models when trained on outputs watermarked teacher models, making it a promising tool for preventing unauthorized knowledge distillation. However, robustness watermark radioactivity against adversarial actors remains largely unexplored. In this paper, we investigate whether can acquire capabilities through distillation while avoiding inheritance. We propose two categories...
The generalizability to new databases is of vital importance Text-to-SQL systems which aim parse human utterances into SQL statements. Existing works achieve this goal by leveraging the exact matching method identify lexical between question words and schema items. However, these methods fail in other challenging scenarios, such as synonym substitution surface form differs corresponding In paper, we propose a framework named ISESL-SQL iteratively build semantic enhanced schema-linking graph...
Aspect-based sentiment analysis (ABSA) is a fine-grained classification task. Many recent works have used dependency trees to extract the relation between aspects and contexts achieved significant improvements. However, further improvement limited due potential mismatch tree as syntactic structure semantic To alleviate this gap, we replace with named Abstract Meaning Representation (AMR) propose model called AMR-based Path Aggregation Relational Network (APARN) take full advantage of...
Xuming Hu, Zhijiang Guo, GuanYu Wu, Aiwei Liu, Lijie Wen, Philip Yu. Proceedings of the 2022 Conference North American Chapter Association for Computational Linguistics: Human Language Technologies. 2022.
We propose the first character-level white-box adversarial attack method against transformer models. The intuition of our comes from observation that words are split into subtokens before being fed models and substitution between two close has a similar effect with character modification. Our mainly contains three steps. First, gradient-based is adopted to find most vulnerable in sentence. Then we selected replace origin tokenization result tokenizer. Finally, utilize an loss guide...
Relation extraction (RE) tasks show promising performance in extracting relations from two entities mentioned sentences, given sufficient annotations available during training. Such would be labor-intensive to obtain practice. Existing work adopts data augmentation techniques generate pseudo-annotated sentences beyond limited annotations. These neither preserve the semantic consistency of original when rule-based augmentations are adopted, nor syntax structure expressing using seq2seq...
How can we better extract entities and relations from text? Using multimodal extraction with images text obtains more signals for relations, aligns them through graphs or hierarchical fusion, aiding in extraction. Despite attempts at various fusions, previous works have overlooked many unlabeled image-caption pairs, such as NewsCLIPing. This paper proposes innovative pre-training objectives entity-object relation-image alignment, extracting objects aligning entity relation prompts soft...
Data augmentation techniques have been used to alleviate the problem of scarce labeled data in various NER tasks (flat, nested, and discontinuous tasks). Existing either manipulate words original text that break semantic coherence text, or exploit generative models ignore preserving entities which impedes use on nested tasks. In this work, we propose a novel Entity-to-Text based technique named EnTDA add, delete, replace swap entity list texts, adopt these augmented lists generate...
Named Entity Recognition (NER) is a well and widely studied task in natural language processing. Recently, the nested NER has attracted more attention since its practicality difficulty. Existing works for ignore recognition order boundary position relation of entities. To address these issues, we propose novel seq2seq model named GPRL, which formulates as an entity triplet sequence generation process. GPRL adopts reinforcement learning method to generate triplets de-coupling gold labels...
Augmenting Large Language Models (LLMs) with retrieved external knowledge has proven effective for improving the factual accuracy of generated responses. Despite their success, retrieval-augmented LLMs still face distractibility issue, where responses are negatively influenced by noise from both and internal sources. In this paper, we introduce a novel, training-free decoding method guided entropy considerations to mitigate issue. Our approach utilizes entropy-based document-parallel...
Open relation extraction is the task of extracting open-domain facts from natural language sentences. Existing works either utilize distant-supervised annotations to train a supervised classifier over pre-defined relations, or adopt unsupervised methods with additional dependency on external assumptions. However, these can only obtain information signals limited existing knowledge bases datasets. In this work, we propose self-supervised framework named <monospace...