Existing multi-label topic models face several challenges when interpreting texts annotated with multiple labels: (1) they often associate irrelevant text segments with incorrect labels, which negatively impacts both segment and label interpretation; (2) they fail to effectively capture the semantic relationships between tokens and labels within the segment; (3) they do not integrate contextualized knowledge that could improve interpretability. To overcome these issues, we introduce the Contextualized Prompting Topic Model (CPTM). CPTM utilizes Chain-of-Thought (CoT) prompting to better align text segments with their semantically relevant labels. Furthermore, it integrates label-specific token visualization and topic mining procedure to facilitate the interpretation of tokens and labels. Experimental evaluations conducted on three multi-label text datasets show that CPTM significantly outperforms existing models in both segment and label interpretation. Human assessments also verify CPTM's effectiveness in accurately identifying label-relevant tokens within segments and providing insightful token-level interpretation.

Mining topics relevant to the advanced AI dialogue system, such as ChatGPT, from short-length posts on social media poses several challenges for existing topic-mining approaches. Firstly, Bag-Of-Words approaches, including probabilistic topic models and their embedding-based variants, may struggle to extract interpretable topics due to insufficient word co-occurrence. Secondly, contextualized based approaches, built on the autoencoding framework, often yield entangled topic spaces, resulting in the mixing of irrelevant words into topics. To address these limitations, we propose a novel Dis entangled Contextualized-neural Topic Model (DisCTM) based on textual representation learning. DisCTM leverages a pre-trained transformer language model to incorporate word sequence information and deal with the sparsity in short text. Additionally, it employs a topic disentangling mechanism to decorrelate dimensions of the latent topic space, effectively separating semantically irrelevant words into different topics. Extensive experiments have been conducted on three publicly available text corpora, and the results demonstrate the effectiveness of DisCTM in extracting high-quality topics, as measured by topic coherence and diversity metrics.

Customer-generated reviews on e-commerce websites often contain valuable insights into users' interests in product genres and provide a rich source for mining user preferences. However, most existing neural topic models tend to generate meaningless topics that share low correlations with product genres. Furthermore, they often fail to mine user preferences and discover personalized topic profiles due to the absence of explicit user modeling. To address these limitations, we propose a novel Genre-aware Personalized neural Topic Model (GPTM), which incorporates product genre information into the topic modeling process to ensure the relevance between mined topics and product genres. Moreover, it could produce a personalized topic profile for each user by performing user preference modeling. Extensive experimental results on three publicly available Amazon review corpora validate the effectiveness of the proposed GPTM in genre-aware topic modeling. Furthermore, GPTM surpasses state-of-the-art baselines in user preference mining and generates high-quality personalized topic profiles.

Neural networks have demonstrated exceptional performance in supervised learning, benefiting from abundant high-quality annotated data. However, obtaining such data in real-world scenarios is costly and labor-intensive. Semi-supervised learning (SSL) offers a solution to this problem by utilizing a small amount of labeled data along with a large volume of unlabeled data . Recent studies, such as Semi-ViT and Noisy Student, which employ consistency regularization or pseudo-labeling, have demonstrated significant achievements. However, they still face challenges, particularly in accurately selecting sufficient high-quality pseudo-labels due to their reliance on fixed thresholds. Recent methods such as FlexMatch and FreeMatch have introduced flexible or self-adaptive thresholding techniques, greatly advancing SSL research. Nonetheless, their process of updating thresholds at each iteration is deemed time-consuming, computationally intensive, and potentially unnecessary. To address these issues, we propose Self-training with Self-adaptive Thresholding (SST), a novel, effective, and efficient SSL framework. SST integrates with both supervised (Super-SST) and semi-supervised (Semi-SST) learning. SST introduces an innovative Self-Adaptive Thresholding (SAT) mechanism that adaptively adjusts class-specific thresholds based on the model’s learning progress. SAT ensures the selection of high-quality pseudo-labeled data, mitigating the risks of inaccurate pseudo-labels and confirmation bias (where models reinforce their own mistakes during training). Specifically, SAT prevents the model from prematurely incorporating low-confidence pseudo-labels, reducing error reinforcement and enhancing model performance. Extensive experiments demonstrate that SST achieves state-of-the-art performance with remarkable efficiency, generalization, and scalability across various architectures and datasets. Notably, Semi-SST-ViT-Huge achieves the best results on competitive ImageNet-1K SSL benchmarks (no external data), with 80.7%/84.9% Top-1 accuracy using only 1%/10% labeled data. Compared to the fully-supervised DeiT-III-ViT-Huge, which achieves 84.8% Top-1 accuracy using 100% labeled data, our method demonstrates superior performance using only 10% labeled data. This indicates a tenfold reduction in human annotation costs, significantly narrowing the performance disparity between semi-supervised and fully-supervised methods. These advancements pave the way for further innovations in SSL and practical applications where obtaining labeled data is either challenging or costly.

Topic models are often used to discover latent semantic patterns from document collections. However, existing unsupervised approaches have the following drawbacks: (1) The mined topics may not match user interests; (2) They are prone to extract semantically similar topics and sacrifice diversity; (3) The mined topics often have low interpretability, which does not meet common sense knowledge. To address these limitations, we propose the Distributed Keyword-guided Topic Model (DiskTM) that incorporates Gaussian-distributed keyword prior knowledge into the modeling process to mine user-interested topics. Furthermore, to inject common-sense knowledge and improve the topic’s interpretability, we extend DiskTM and propose the Distributed Keyword-guided Topic Model with Lexical Knowledge (DiskTM-LK). Experimental results on three publicly available text corpora show that our proposed approaches could extract topics that match user interests (keywords). Moreover, DiskTM and DiskTM-LK could also obtain more coherent and diverse topics, outperforming the state-of-the-art baseline approaches.