A5023856576- Topic Modeling
- Multimodal Machine Learning Applications
- Advanced Image and Video Retrieval Techniques
- Sentiment Analysis and Opinion Mining
- Data-Driven Disease Surveillance
- Video Analysis and Summarization
- Natural Language Processing Techniques
- Advanced Wireless Communication Technologies
- Advanced Text Analysis Techniques
- UAV Applications and Optimization
- Speech and dialogue systems
- Misinformation and Its Impacts
- Human Pose and Action Recognition
- Satellite Communication Systems
Xinjiang University
2022-2023
The amount of training data is small in the field Weibo rumor detection, and online news changes constantly, but existing models do not have ability continuous learning, which means they cannot achieve knowledge accumulation update. Hence, will require many examples to improve their detection ability. In contrast, Lifelong Machine Learning (LML) paradigm has capability retains learned past uses it help future learning tasks. After some updated. With growth update knowledge, performance each...
Image-text matching is a crucial aspect of multi-modal intelligence. The main challenge in this area accurately measuring the relevance between image and text, using evidence obtained through matching. Previous studies either concentrated on obtaining well-represented global feature to measure similarity directly or investigating complex patterns at local level before aggregating them, with little attention paid combining them. We propose Globally Guided Confidence Enhancement Network that...
神经网络在方面级情感分类任务上已经取得了良好的性能. 然而, 由于复杂且耗时的数据标注流程, 方面级情感分类在很多领域上是低资源甚至是零资源的, 这限制了该任务在实际场景中的应用. 为了解决这个挑战性的问题, 本文关注跨领域的方面级情感分类, 并提出一种基于语法和语义分割的跨领域方面情感分类方法. 具体而言, 针对不同领域用词差异造成的领域漂移和注意力泛化问题, 本文首次提出利用单纯的语法信息来获取可在领域之间迁移的语法注意力, 并引入与目标领域相近的文档情感分类任务来增强神经网络模型对目标领域的情感识别能力, 最终从语法和语义两个层面分别提升模型的注意力机制和文本上下文表示. 实验在6个跨领域方面级情感分类任务上进行, 结果表明, 本文的方法在6个任务上都取得了最先进的性能, 在平均准确率和平均macro F1-score两个指标上比之前最好的模型DIFD分别提升7.14%和7.6%. 此外, 即使以大规模预训练模型BERT、BERT-ADA、RoBERTa等作为骨干网络, 本文的方法仍能实现3.5%以上的平均准确率提升和平均macro F1-score提升.
Intent detection is an important component in dialog systems. Existing studies mainly focus on intent with sufficient labeled data. However, these methods are unable to detect intents that do not exist training To tackle this problem, we propose implicit-network and explicit-network model for zero-shot detection, which capable of learning similarities between utterances description from word level sentence level. enhance the representation intent, introduce slot types as description. We...
Paraphrase generation is an important natural language-processing task widely used in downstream research, such as machine translation, question answering, and text classification.Recently, some researchers have explored the advantages of paraphrasing based on syntax sampling, generating multiple paraphrases by sampling various syntactic representations from probabilistic latent space.However, sampled posterior distribution always show high similarity generate low-diversity...