tf-idf vectorization

Text classification (TC) faces challenges stemming from semantic complexity and cross-lingual generalization. To address these, this work proposes a component-level deep evaluation framework that systematically dissects the entire TC pipeline—encompassing data, preprocessing, representation learning, model architecture, and evaluation—thereby transcending conventional survey paradigms. Leveraging multilingual empirical studies and reproducible case analyses, we integrate statistical methods, word embeddings, pretrained language models (PLMs), and lightweight classifiers to establish a multidimensional evaluation suite measuring accuracy, F1-score, robustness, and inference efficiency. Experimental results reveal that preprocessing strategies and representation choices exert decisive influence on cross-lingual generalization. We further construct a highly comparable benchmark matrix and identify synergistic optimization pathways for low-resource language adaptation and computational efficiency.

This study investigates the applicability of quantum machine learning to natural language processing, with a focus on sentiment analysis. The authors propose a hybrid quantum-classical architecture that integrates parameterized quantum circuits with a classical feedforward neural network, processing text data represented as TF-IDF vectors. The model is evaluated on a real-world dataset of COVID-19-related tweets, achieving sentiment classification accuracy comparable to purely classical approaches. Furthermore, through transfer learning, the framework is successfully applied to SMS spam detection, where it improves spam-class identification accuracy by 15 percentage points—from 66% to 81%—demonstrating enhanced generalization capability and distinctive training dynamics relative to conventional models.

This work addresses the three-class sentiment analysis task for financial texts by proposing QDisCoCirc, a hybrid architecture integrating quantum semantics with classical structural modeling. Methodologically, sentences are segmented and mapped onto shallow quantum circuits, where Bloch vector encoding yields interpretable quantum semantic representations; Combinatory Categorial Grammar (CCG) type embeddings explicitly encode syntactic structure, while a lightweight Transformer encoder captures word order and long-range dependencies. The key contribution is the first application of the QDisCoCirc framework to financial NLP—enabling synergistic syntax–semantics modeling without sacrificing quantum representation interpretability. Experiments demonstrate statistically significant improvements in macro-F1 on the test set. Ablation and attribution analyses confirm that predictions are predominantly driven by critical semantic blocks, and CCG type embeddings exhibit higher consistency in correctly classified instances, validating their role in enhancing structural coherence.

Classical Transformer and MLP architectures face fundamental limitations in text generation—particularly in repetition suppression and lexical diversity—despite strong overall perplexity performance. Method: This work systematically evaluates quantum-enhanced text generation models against classical baselines across NLP tasks, proposing three hybrid quantum-classical architectures: Quantum Kernel Self-Attention Network (QKSAN), Quantized RWKV (QRWKV), and Quantum Adaptive Sparse Attention (QASA). These integrate quantum kernel methods with co-training strategies and are evaluated on BLEU, perplexity, and Distinct metrics. Results: While classical Transformers achieve the lowest overall perplexity (1.21), QKSAN attains superior BLEU-1 (0.2800) and zero repetition rate; QRWKV achieves 100% lexical diversity. This study provides the first empirical validation that quantum mechanisms can overcome repetition and diversity bottlenecks without compromising generation quality—establishing a new paradigm and foundational evidence for scalable, practical quantum NLP.

To address the high computational cost and hyperparameter sensitivity of BERT fine-tuning in text classification, this paper proposes a classical–quantum hybrid model: an *n*-qubit variational quantum circuit is embedded within the pre-trained BERT architecture, enabling end-to-end co-training of feature encoding and classification layers. This work constitutes the first instance of differentiable quantum circuits being jointly fine-tuned with BERT without modifying its backbone—quantum enhancement is achieved solely through lightweight, trainable quantum modules that augment semantic representation capacity. The model demonstrates cross-dataset adaptability and achieves accuracy comparable to or exceeding state-of-the-art classical models on standard benchmarks (e.g., AG News, SST-2), while maintaining controlled increases in parameter count and inference latency. Empirical results validate the feasibility, effectiveness, and generalization potential of quantum-enhanced approaches in NLP tasks.

This study investigates the representational capacity of quantum-inspired high-dimensional document embeddings in information retrieval and their potential to complement conventional dense embeddings. The authors propose a 1024-dimensional embedding framework incorporating EigAngle semantic projection, circuit-inspired feature mapping, and optional teacher–student distillation. To evaluate its efficacy, they develop a diagnostic toolkit comprising dynamic and static BM25–embedding interpolation, candidate fusion strategies, and alpha-oracle upper-bound analysis. Experimental results indicate that, when used in isolation, the proposed embedding yields weak and unstable ranking signals and suffers from distance compression issues. However, when integrated with BM25 in a hybrid retrieval setting, it significantly enhances retrieval performance, thereby demonstrating its value as a complementary signal rather than a standalone representation.

This study addresses the challenge of automatically identifying speculative language in biomedical texts by proposing and systematically evaluating a range of deep learning and traditional approaches. It introduces, for the first time, the application of Recursive Neural Tensor Networks (RNTN) to this task and compares its performance against Paragraph Vector models, Support Vector Machines (SVM), Naive Bayes classifiers, and pattern-matching techniques. Experimental results demonstrate that RNTN achieves the best performance with an F1-score of 0.885, marginally outperforming linear bigram SVM (F1 = 0.881), while the Paragraph Vector model yields substantially lower accuracy (F1 = 0.368). These findings offer an effective technical pathway for enhancing the precision of biomedical information retrieval, multi-document summarization, and novel knowledge discovery.

This work proposes a semantic similarity computation method that integrates Word Mover’s Distance (WMD) with pretrained word embeddings such as GloVe to better model the semantic relationship between queries and documents in information retrieval. Traditional centroid-based word embedding approaches often fail to capture fine-grained semantic matches, particularly when handling synonymy and polysemy. By minimizing the transportation cost of aligning query and document terms in the embedding space, the proposed method achieves a more precise representation of semantic correspondence. Experimental results demonstrate that this approach significantly outperforms baseline models—including Doc2Vec and Latent Semantic Analysis (LSA)—on similarity ranking tasks, while maintaining domain independence and high retrieval accuracy, thereby confirming its effectiveness and generalizability in practical information retrieval scenarios.

This study systematically investigates the intrinsic semantic and syntactic properties of mainstream word embedding methods—such as Word2Vec and GloVe—and their performance disparities across diverse natural language processing tasks. By establishing a unified evaluation framework that integrates publicly available pretrained embeddings with standard benchmark datasets, the work conducts empirical comparisons on canonical tasks including semantic similarity and analogical reasoning. The findings delineate the performance boundaries and optimal application scenarios for each embedding model, offering practitioners reliable guidance for model selection in real-world settings. Furthermore, the analysis deepens the understanding of the inherent limitations of static word representations, highlighting critical constraints in capturing contextual and compositional linguistic phenomena.

This work addresses the limitation of traditional HAL models in text classification, where mean pooling treats all tokens uniformly and consequently dilutes critical semantic information. To overcome this, we propose a novel approach that integrates a learnable temperature-scaled additive attention mechanism into the pooling stage of HAL, coupled with truncated singular value decomposition (SVD) for dimensionality reduction. This design enables the model to adaptively emphasize sentiment-relevant keywords while suppressing stop words, yielding more discriminative sentence-level representations. Evaluated on the IMDB dataset, our method achieves a test accuracy of 82.38%, representing a 6.74 percentage point improvement over the mean-pooling baseline. Attention visualization further confirms the model’s capacity to effectively attend to salient sentiment-bearing words, enhancing both performance and interpretability.