传统编目分类和规则匹配方法存在工作效能低、过度依赖专家知识、缺乏对古籍文本自身语义的深层次挖掘、编目主题边界模糊、较难实现对古籍文本领域主题的精准推荐等问题。为此,本文结合古籍语料特征探究如何实现精准推荐符合研究者需...传统编目分类和规则匹配方法存在工作效能低、过度依赖专家知识、缺乏对古籍文本自身语义的深层次挖掘、编目主题边界模糊、较难实现对古籍文本领域主题的精准推荐等问题。为此,本文结合古籍语料特征探究如何实现精准推荐符合研究者需求的文本主题内容的方法,以推动数字人文研究的进一步发展。首先,选取本课题组前期标注的古籍语料数据进行主题类别标注和视图分类;其次,构建融合BERT(bidirectional encoder representation from transformers)预训练模型、改进卷积神经网络、循环神经网络和多头注意力机制的语义挖掘模型;最后,融入“主体-关系-客体”多视图的语义增强模型,构建DJ-TextRCNN(DianJi-recurrent convolutional neural networks for text classification)模型实现对典籍文本更细粒度、更深层次、更多维度的语义挖掘。研究结果发现,DJ-TextRCNN模型在不同视图下的古籍主题推荐任务的准确率均为最优。在“主体-关系-客体”视图下,精确率达到88.54%,初步实现了对古籍文本的精准主题推荐,对中华文化深层次、细粒度的语义挖掘具有一定的指导意义。展开更多
以编目分类和规则匹配为主的古籍文本主题分类方法存在工作效能低、专家知识依赖性强、分类依据单一化、古籍文本主题自动分类难等问题。对此,本文结合古籍文本内容和文字特征,尝试从古籍内容分类得到符合研究者需求的主题,推动数字人...以编目分类和规则匹配为主的古籍文本主题分类方法存在工作效能低、专家知识依赖性强、分类依据单一化、古籍文本主题自动分类难等问题。对此,本文结合古籍文本内容和文字特征,尝试从古籍内容分类得到符合研究者需求的主题,推动数字人文研究范式的转型。首先,参照东汉古籍《说文解字》对文字的分析方式,以前期标注的古籍语料数据集为基础,构建全新的“字音(说)-原文(文)-结构(解)-字形(字)”四维特征数据集。其次,设计四维特征向量提取模型(speaking,word,pattern,and font to vector,SWPF2vec),并结合预训练模型实现对古籍文本细粒度的特征表示。再其次,构建融合卷积神经网络、循环神经网络和多头注意力机制的古籍文本主题分类模型(dianji-recurrent convolutional neural networks for text classification,DJ-TextRCNN)。最后,融入四维语义特征,实现对古籍文本多维度、深层次、细粒度的语义挖掘。在古籍文本主题分类任务上,DJ-TextRCNN模型在不同维度特征下的主题分类准确率均为最优,在“说文解字”四维特征下达到76.23%的准确率,初步实现了对古籍文本的精准主题分类。展开更多
To promote behavioral change among adolescents in Zambia, the National HIV/AIDS/STI/TB Council, in collaboration with UNICEF, developed the Zambia U-Report platform. This platform provides young people with improved a...To promote behavioral change among adolescents in Zambia, the National HIV/AIDS/STI/TB Council, in collaboration with UNICEF, developed the Zambia U-Report platform. This platform provides young people with improved access to information on various Sexual Reproductive Health topics through Short Messaging Service (SMS) messages. Over the years, the platform has accumulated millions of incoming and outgoing messages, which need to be categorized into key thematic areas for better tracking of sexual reproductive health knowledge gaps among young people. The current manual categorization process of these text messages is inefficient and time-consuming and this study aims to automate the process for improved analysis using text-mining techniques. Firstly, the study investigates the current text message categorization process and identifies a list of categories adopted by counselors over time which are then used to build and train a categorization model. Secondly, the study presents a proof of concept tool that automates the categorization of U-report messages into key thematic areas using the developed categorization model. Finally, it compares the performance and effectiveness of the developed proof of concept tool against the manual system. The study used a dataset comprising 206,625 text messages. The current process would take roughly 2.82 years to categorise this dataset whereas the trained SVM model would require only 6.4 minutes while achieving an accuracy of 70.4% demonstrating that the automated method is significantly faster, more scalable, and consistent when compared to the current manual categorization. These advantages make the SVM model a more efficient and effective tool for categorizing large unstructured text datasets. These results and the proof-of-concept tool developed demonstrate the potential for enhancing the efficiency and accuracy of message categorization on the Zambia U-report platform and other similar text messages-based platforms.展开更多
Modeling topics in short texts presents significant challenges due to feature sparsity, particularly when analyzing content generated by large-scale online users. This sparsity can substantially impair semantic captur...Modeling topics in short texts presents significant challenges due to feature sparsity, particularly when analyzing content generated by large-scale online users. This sparsity can substantially impair semantic capture accuracy. We propose a novel approach that incorporates pre-clustered knowledge into the BERTopic model while reducing the l2 norm for low-frequency words. Our method effectively mitigates feature sparsity during cluster mapping. Empirical evaluation on the StackOverflow dataset demonstrates that our approach outperforms baseline models, achieving superior Macro-F1 scores. These results validate the effectiveness of our proposed feature sparsity reduction technique for short-text topic modeling.展开更多
Video-text retrieval (VTR) is an essential task in multimodal learning, aiming to bridge the semantic gap between visual and textual data. Effective video frame sampling plays a crucial role in improving retrieval per...Video-text retrieval (VTR) is an essential task in multimodal learning, aiming to bridge the semantic gap between visual and textual data. Effective video frame sampling plays a crucial role in improving retrieval performance, as it determines the quality of the visual content representation. Traditional sampling methods, such as uniform sampling and optical flow-based techniques, often fail to capture the full semantic range of videos, leading to redundancy and inefficiencies. In this work, we propose CLIP4Video-Sampling: Global Semantics-Guided Multi-Granularity Frame Sampling for Video-Text Retrieval, a global semantics-guided multi-granularity frame sampling strategy designed to optimize both computational efficiency and retrieval accuracy. By integrating multi-scale global and local temporal sampling and leveraging the CLIP (Contrastive Language-Image Pre-training) model’s powerful feature extraction capabilities, our method significantly outperforms existing approaches in both zero-shot and fine-tuned video-text retrieval tasks on popular datasets. CLIP4Video-Sampling reduces redundancy, ensures keyframe coverage, and serves as an adaptable pre-processing module for multimodal models.展开更多
文摘传统编目分类和规则匹配方法存在工作效能低、过度依赖专家知识、缺乏对古籍文本自身语义的深层次挖掘、编目主题边界模糊、较难实现对古籍文本领域主题的精准推荐等问题。为此,本文结合古籍语料特征探究如何实现精准推荐符合研究者需求的文本主题内容的方法,以推动数字人文研究的进一步发展。首先,选取本课题组前期标注的古籍语料数据进行主题类别标注和视图分类;其次,构建融合BERT(bidirectional encoder representation from transformers)预训练模型、改进卷积神经网络、循环神经网络和多头注意力机制的语义挖掘模型;最后,融入“主体-关系-客体”多视图的语义增强模型,构建DJ-TextRCNN(DianJi-recurrent convolutional neural networks for text classification)模型实现对典籍文本更细粒度、更深层次、更多维度的语义挖掘。研究结果发现,DJ-TextRCNN模型在不同视图下的古籍主题推荐任务的准确率均为最优。在“主体-关系-客体”视图下,精确率达到88.54%,初步实现了对古籍文本的精准主题推荐,对中华文化深层次、细粒度的语义挖掘具有一定的指导意义。
文摘以编目分类和规则匹配为主的古籍文本主题分类方法存在工作效能低、专家知识依赖性强、分类依据单一化、古籍文本主题自动分类难等问题。对此,本文结合古籍文本内容和文字特征,尝试从古籍内容分类得到符合研究者需求的主题,推动数字人文研究范式的转型。首先,参照东汉古籍《说文解字》对文字的分析方式,以前期标注的古籍语料数据集为基础,构建全新的“字音(说)-原文(文)-结构(解)-字形(字)”四维特征数据集。其次,设计四维特征向量提取模型(speaking,word,pattern,and font to vector,SWPF2vec),并结合预训练模型实现对古籍文本细粒度的特征表示。再其次,构建融合卷积神经网络、循环神经网络和多头注意力机制的古籍文本主题分类模型(dianji-recurrent convolutional neural networks for text classification,DJ-TextRCNN)。最后,融入四维语义特征,实现对古籍文本多维度、深层次、细粒度的语义挖掘。在古籍文本主题分类任务上,DJ-TextRCNN模型在不同维度特征下的主题分类准确率均为最优,在“说文解字”四维特征下达到76.23%的准确率,初步实现了对古籍文本的精准主题分类。
文摘To promote behavioral change among adolescents in Zambia, the National HIV/AIDS/STI/TB Council, in collaboration with UNICEF, developed the Zambia U-Report platform. This platform provides young people with improved access to information on various Sexual Reproductive Health topics through Short Messaging Service (SMS) messages. Over the years, the platform has accumulated millions of incoming and outgoing messages, which need to be categorized into key thematic areas for better tracking of sexual reproductive health knowledge gaps among young people. The current manual categorization process of these text messages is inefficient and time-consuming and this study aims to automate the process for improved analysis using text-mining techniques. Firstly, the study investigates the current text message categorization process and identifies a list of categories adopted by counselors over time which are then used to build and train a categorization model. Secondly, the study presents a proof of concept tool that automates the categorization of U-report messages into key thematic areas using the developed categorization model. Finally, it compares the performance and effectiveness of the developed proof of concept tool against the manual system. The study used a dataset comprising 206,625 text messages. The current process would take roughly 2.82 years to categorise this dataset whereas the trained SVM model would require only 6.4 minutes while achieving an accuracy of 70.4% demonstrating that the automated method is significantly faster, more scalable, and consistent when compared to the current manual categorization. These advantages make the SVM model a more efficient and effective tool for categorizing large unstructured text datasets. These results and the proof-of-concept tool developed demonstrate the potential for enhancing the efficiency and accuracy of message categorization on the Zambia U-report platform and other similar text messages-based platforms.
文摘Modeling topics in short texts presents significant challenges due to feature sparsity, particularly when analyzing content generated by large-scale online users. This sparsity can substantially impair semantic capture accuracy. We propose a novel approach that incorporates pre-clustered knowledge into the BERTopic model while reducing the l2 norm for low-frequency words. Our method effectively mitigates feature sparsity during cluster mapping. Empirical evaluation on the StackOverflow dataset demonstrates that our approach outperforms baseline models, achieving superior Macro-F1 scores. These results validate the effectiveness of our proposed feature sparsity reduction technique for short-text topic modeling.
文摘Video-text retrieval (VTR) is an essential task in multimodal learning, aiming to bridge the semantic gap between visual and textual data. Effective video frame sampling plays a crucial role in improving retrieval performance, as it determines the quality of the visual content representation. Traditional sampling methods, such as uniform sampling and optical flow-based techniques, often fail to capture the full semantic range of videos, leading to redundancy and inefficiencies. In this work, we propose CLIP4Video-Sampling: Global Semantics-Guided Multi-Granularity Frame Sampling for Video-Text Retrieval, a global semantics-guided multi-granularity frame sampling strategy designed to optimize both computational efficiency and retrieval accuracy. By integrating multi-scale global and local temporal sampling and leveraging the CLIP (Contrastive Language-Image Pre-training) model’s powerful feature extraction capabilities, our method significantly outperforms existing approaches in both zero-shot and fine-tuned video-text retrieval tasks on popular datasets. CLIP4Video-Sampling reduces redundancy, ensures keyframe coverage, and serves as an adaptable pre-processing module for multimodal models.
