{"url":"/sota/sentiment-analysis-on-sst-2-binary","task":{"name":"Sentiment Analysis","url":"/task/sentiment-analysis","note":null},"dataset":{"name":"SST-2 Binary classification","url":"/dataset/sst"},"category":"Computer Vision","categories":["Computer Vision","Natural Language Processing"],"category_note":null,"description":"**Sentiment Analysis** is the task of classifying the polarity of a given text. For instance, a text-based tweet can be categorized into either \"positive\", \"negative\", or \"neutral\". Given the text and accompanying labels, a model can be trained to predict the correct sentiment. \r\n\r\n**Sentiment Analysis** techniques can be categorized into machine learning approaches, lexicon-based approaches, and even hybrid methods. Some subcategories of research in sentiment analysis include: multimodal sentiment analysis, aspect-based sentiment analysis, fine-grained opinion analysis, language specific sentiment analysis.\r\n\r\nMore recently, deep learning techniques, such as RoBERTa and T5, are used to train high-performing sentiment classifiers that are evaluated using metrics like F1, recall, and precision. To evaluate sentiment analysis systems, benchmark datasets like SST, GLUE, and IMDB movie reviews are used.\r\n\r\nFurther readings:\r\n\r\n- [Sentiment Analysis Based on Deep Learning: A Comparative Study](https://paperswithcode.com/paper/sentiment-analysis-based-on-deep-learning-a)","description_from":"task","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","rank":"the archive's row order at snapshot; not re-ranked","rows_end_at":"2025-07-28","rows_withheld_as_spam":0,"metric_values":"the archive's strings, untouched"},"metrics":["Accuracy","Dev Accuracy","Attack Success Rate"],"metric_direction":{"note":"inferred from the metric name only (the archive records no direction); null = not inferred, chart draws points only","by_metric":{"Accuracy":"higher","Dev Accuracy":"higher","Attack Success Rate":"higher"}},"counts":{"rows":87,"rows_with_code":79,"rows_with_paper_page":87,"rows_dated":87,"rows_using_additional_data":2},"rows":[{"rank_in_archive_order":1,"model":"T5-11B","metrics":{"Accuracy":"97.5"},"uses_additional_data":false,"paper_date":"2019-10-23","paper":"/paper/exploring-the-limits-of-transfer-learning","paper_url":"https://arxiv.org/abs/1910.10683v4","paper_title":"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer","code":"https://github.com/huggingface/transformers","n_code_links":57,"syntology":{"n_ran":2,"n_unverified":29,"n_samples":31,"n_pointer_only_licence":0}},{"rank_in_archive_order":2,"model":"MT-DNN-SMART","metrics":{"Accuracy":"97.5"},"uses_additional_data":false,"paper_date":"2019-11-08","paper":"/paper/smart-robust-and-efficient-fine-tuning-for","paper_url":"https://arxiv.org/abs/1911.03437v5","paper_title":"SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization","code":"https://github.com/namisan/mt-dnn","n_code_links":6,"syntology":{"n_ran":6,"n_unverified":2,"n_samples":8,"n_pointer_only_licence":1}},{"rank_in_archive_order":3,"model":"T5-3B","metrics":{"Accuracy":"97.4"},"uses_additional_data":false,"paper_date":"2019-10-23","paper":"/paper/exploring-the-limits-of-transfer-learning","paper_url":"https://arxiv.org/abs/1910.10683v4","paper_title":"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer","code":"https://github.com/huggingface/transformers","n_code_links":57,"syntology":{"n_ran":2,"n_unverified":29,"n_samples":31,"n_pointer_only_licence":0}},{"rank_in_archive_order":4,"model":"MUPPET Roberta Large","metrics":{"Accuracy":"97.4"},"uses_additional_data":false,"paper_date":"2021-01-26","paper":"/paper/muppet-massive-multi-task-representations","paper_url":"https://arxiv.org/abs/2101.11038v1","paper_title":"Muppet: Massive Multi-task Representations with Pre-Finetuning","code":"https://huggingface.co/facebook/muppet-roberta-base","n_code_links":2,"syntology":null},{"rank_in_archive_order":5,"model":"ALBERT","metrics":{"Accuracy":"97.1"},"uses_additional_data":false,"paper_date":"2019-09-26","paper":"/paper/albert-a-lite-bert-for-self-supervised","paper_url":"https://arxiv.org/abs/1909.11942v6","paper_title":"ALBERT: A Lite BERT for Self-supervised Learning of Language Representations","code":"https://github.com/huggingface/transformers","n_code_links":48,"syntology":{"n_ran":46,"n_unverified":80,"n_samples":126,"n_pointer_only_licence":22}},{"rank_in_archive_order":6,"model":"StructBERTRoBERTa ensemble","metrics":{"Accuracy":"97.1"},"uses_additional_data":false,"paper_date":"2019-08-13","paper":"/paper/structbert-incorporating-language-structures","paper_url":"https://arxiv.org/abs/1908.04577v3","paper_title":"StructBERT: Incorporating Language Structures into Pre-training for Deep Language Understanding","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":7,"model":"XLNet (single model)","metrics":{"Accuracy":"97"},"uses_additional_data":false,"paper_date":"2019-06-19","paper":"/paper/xlnet-generalized-autoregressive-pretraining","paper_url":"https://arxiv.org/abs/1906.08237v2","paper_title":"XLNet: Generalized Autoregressive Pretraining for Language Understanding","code":"https://github.com/huggingface/transformers","n_code_links":27,"syntology":{"n_ran":10,"n_unverified":14,"n_samples":24,"n_pointer_only_licence":3}},{"rank_in_archive_order":8,"model":"ELECTRA","metrics":{"Accuracy":"96.9"},"uses_additional_data":false,"paper_date":"2020-03-23","paper":"/paper/electra-pre-training-text-encoders-as-1","paper_url":"https://arxiv.org/abs/2003.10555v1","paper_title":"ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators","code":"https://github.com/huggingface/transformers","n_code_links":19,"syntology":{"n_ran":26,"n_unverified":14,"n_samples":40,"n_pointer_only_licence":10}},{"rank_in_archive_order":9,"model":"RoBERTa-large 355M + Entailment as Few-shot Learner","metrics":{"Accuracy":"96.9"},"uses_additional_data":false,"paper_date":"2021-04-29","paper":"/paper/entailment-as-few-shot-learner","paper_url":"https://arxiv.org/abs/2104.14690v1","paper_title":"Entailment as Few-Shot Learner","code":"https://github.com/PaddlePaddle/PaddleNLP/tree/develop/examples/few_shot/efl","n_code_links":3,"syntology":{"n_ran":1,"n_unverified":2,"n_samples":3,"n_pointer_only_licence":0}},{"rank_in_archive_order":10,"model":"XLNet-Large (ensemble)","metrics":{"Accuracy":"96.8"},"uses_additional_data":false,"paper_date":"2019-06-19","paper":"/paper/xlnet-generalized-autoregressive-pretraining","paper_url":"https://arxiv.org/abs/1906.08237v2","paper_title":"XLNet: Generalized Autoregressive Pretraining for Language Understanding","code":"https://github.com/huggingface/transformers","n_code_links":27,"syntology":{"n_ran":10,"n_unverified":14,"n_samples":24,"n_pointer_only_licence":3}},{"rank_in_archive_order":11,"model":"FLOATER-large","metrics":{"Accuracy":"96.7"},"uses_additional_data":false,"paper_date":"2020-03-13","paper":"/paper/learning-to-encode-position-for-transformer","paper_url":"https://arxiv.org/abs/2003.09229v1","paper_title":"Learning to Encode Position for Transformer with Continuous Dynamical Model","code":"https://github.com/xuanqing94/FLOATER","n_code_links":1,"syntology":{"n_ran":3,"n_unverified":3,"n_samples":6,"n_pointer_only_licence":6}},{"rank_in_archive_order":12,"model":"MUPPET Roberta base","metrics":{"Accuracy":"96.7"},"uses_additional_data":false,"paper_date":"2021-01-26","paper":"/paper/muppet-massive-multi-task-representations","paper_url":"https://arxiv.org/abs/2101.11038v1","paper_title":"Muppet: Massive Multi-task Representations with Pre-Finetuning","code":"https://huggingface.co/facebook/muppet-roberta-base","n_code_links":2,"syntology":null},{"rank_in_archive_order":13,"model":"RoBERTa (ensemble)","metrics":{"Accuracy":"96.7"},"uses_additional_data":false,"paper_date":"2019-07-26","paper":"/paper/roberta-a-robustly-optimized-bert-pretraining","paper_url":"https://arxiv.org/abs/1907.11692v1","paper_title":"RoBERTa: A Robustly Optimized BERT Pretraining Approach","code":"https://github.com/huggingface/transformers","n_code_links":67,"syntology":{"n_ran":22,"n_unverified":26,"n_samples":48,"n_pointer_only_licence":23}},{"rank_in_archive_order":14,"model":"DeBERTa (large)","metrics":{"Accuracy":"96.5"},"uses_additional_data":false,"paper_date":"2020-06-05","paper":"/paper/deberta-decoding-enhanced-bert-with","paper_url":"https://arxiv.org/abs/2006.03654v6","paper_title":"DeBERTa: Decoding-enhanced BERT with Disentangled Attention","code":"https://github.com/huggingface/transformers","n_code_links":14,"syntology":{"n_ran":4,"n_unverified":9,"n_samples":13,"n_pointer_only_licence":3}},{"rank_in_archive_order":15,"model":"MT-DNN-ensemble","metrics":{"Accuracy":"96.5"},"uses_additional_data":false,"paper_date":"2019-04-20","paper":"/paper/improving-multi-task-deep-neural-networks-via","paper_url":"http://arxiv.org/abs/1904.09482v1","paper_title":"Improving Multi-Task Deep Neural Networks via Knowledge Distillation for Natural Language Understanding","code":"https://github.com/namisan/mt-dnn","n_code_links":3,"syntology":null},{"rank_in_archive_order":16,"model":"RoBERTa-large 355M (MLP quantized vector-wise, fine-tuned)","metrics":{"Accuracy":"96.4"},"uses_additional_data":false,"paper_date":"2022-08-15","paper":"/paper/llm-int8-8-bit-matrix-multiplication-for","paper_url":"https://arxiv.org/abs/2208.07339v2","paper_title":"LLM.int8(): 8-bit Matrix Multiplication for Transformers at Scale","code":"https://github.com/timdettmers/bitsandbytes","n_code_links":4,"syntology":{"n_ran":2,"n_unverified":3,"n_samples":5,"n_pointer_only_licence":0}},{"rank_in_archive_order":17,"model":"ASA + RoBERTa","metrics":{"Accuracy":"96.3"},"uses_additional_data":false,"paper_date":"2022-06-25","paper":"/paper/adversarial-self-attention-for-language","paper_url":"https://arxiv.org/abs/2206.12608v3","paper_title":"Adversarial Self-Attention for Language Understanding","code":"https://github.com/gingasan/adversarialsa","n_code_links":1,"syntology":null},{"rank_in_archive_order":18,"model":"T5-Large 770M","metrics":{"Accuracy":"96.3"},"uses_additional_data":false,"paper_date":"2019-10-23","paper":"/paper/exploring-the-limits-of-transfer-learning","paper_url":"https://arxiv.org/abs/1910.10683v4","paper_title":"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer","code":"https://github.com/huggingface/transformers","n_code_links":57,"syntology":{"n_ran":2,"n_unverified":29,"n_samples":31,"n_pointer_only_licence":0}},{"rank_in_archive_order":19,"model":"Snorkel MeTaL(ensemble)","metrics":{"Accuracy":"96.2"},"uses_additional_data":false,"paper_date":"2018-10-05","paper":"/paper/training-complex-models-with-multi-task-weak","paper_url":"http://arxiv.org/abs/1810.02840v2","paper_title":"Training Complex Models with Multi-Task Weak Supervision","code":"https://github.com/HazyResearch/metal","n_code_links":1,"syntology":null},{"rank_in_archive_order":20,"model":"PSQ (Chen et al., 2020)","metrics":{"Accuracy":"96.2"},"uses_additional_data":false,"paper_date":"2020-10-27","paper":"/paper/a-statistical-framework-for-low-bitwidth","paper_url":"https://arxiv.org/abs/2010.14298v1","paper_title":"A Statistical Framework for Low-bitwidth Training of Deep Neural Networks","code":"https://github.com/cjf00000/StatQuant","n_code_links":2,"syntology":{"n_ran":1,"n_unverified":3,"n_samples":4,"n_pointer_only_licence":1}},{"rank_in_archive_order":21,"model":"Heinsen Routing + RoBERTa-large","metrics":{"Accuracy":"96.0"},"uses_additional_data":true,"paper_date":"2022-11-20","paper":"/paper/an-algorithm-for-routing-vectors-in-sequences","paper_url":"https://arxiv.org/abs/2211.11754v3","paper_title":"An Algorithm for Routing Vectors in Sequences","code":"https://github.com/glassroom/heinsen_routing","n_code_links":1,"syntology":null},{"rank_in_archive_order":22,"model":"MT-DNN","metrics":{"Accuracy":"95.6"},"uses_additional_data":false,"paper_date":"2019-01-31","paper":"/paper/multi-task-deep-neural-networks-for-natural","paper_url":"https://arxiv.org/abs/1901.11504v2","paper_title":"Multi-Task Deep Neural Networks for Natural Language Understanding","code":"https://github.com/namisan/mt-dnn","n_code_links":7,"syntology":{"n_ran":5,"n_unverified":8,"n_samples":13,"n_pointer_only_licence":2}},{"rank_in_archive_order":23,"model":"Heinsen Routing + GPT-2","metrics":{"Accuracy":"95.6"},"uses_additional_data":true,"paper_date":"2019-11-02","paper":"/paper/an-algorithm-for-routing-capsules-in-all","paper_url":"https://arxiv.org/abs/1911.00792v6","paper_title":"An Algorithm for Routing Capsules in All Domains","code":"https://github.com/glassroom/heinsen_routing","n_code_links":1,"syntology":null},{"rank_in_archive_order":24,"model":"T5-Base","metrics":{"Accuracy":"95.2"},"uses_additional_data":false,"paper_date":"2019-10-23","paper":"/paper/exploring-the-limits-of-transfer-learning","paper_url":"https://arxiv.org/abs/1910.10683v4","paper_title":"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer","code":"https://github.com/huggingface/transformers","n_code_links":57,"syntology":{"n_ran":2,"n_unverified":29,"n_samples":31,"n_pointer_only_licence":0}},{"rank_in_archive_order":25,"model":"ERNIE 2.0 Base","metrics":{"Accuracy":"95"},"uses_additional_data":false,"paper_date":"2019-07-29","paper":"/paper/ernie-20-a-continual-pre-training-framework","paper_url":"https://arxiv.org/abs/1907.12412v2","paper_title":"ERNIE 2.0: A Continual Pre-training Framework for Language Understanding","code":"https://github.com/PaddlePaddle/PaddleNLP/tree/develop/model_zoo/ernie-1.0","n_code_links":3,"syntology":{"n_ran":0,"n_unverified":1,"n_samples":1,"n_pointer_only_licence":1}},{"rank_in_archive_order":26,"model":"RoBERTa+DualCL","metrics":{"Accuracy":"94.91"},"uses_additional_data":false,"paper_date":"2022-01-21","paper":"/paper/dual-contrastive-learning-text-classification","paper_url":"https://arxiv.org/abs/2201.08702v1","paper_title":"Dual Contrastive Learning: Text Classification via Label-Aware Data Augmentation","code":"https://github.com/hiyouga/dual-contrastive-learning","n_code_links":2,"syntology":null},{"rank_in_archive_order":27,"model":"BERT-LARGE","metrics":{"Accuracy":"94.9"},"uses_additional_data":false,"paper_date":"2018-10-11","paper":"/paper/bert-pre-training-of-deep-bidirectional","paper_url":"https://arxiv.org/abs/1810.04805v2","paper_title":"BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding","code":"https://github.com/huggingface/transformers","n_code_links":534,"syntology":{"n_ran":204,"n_unverified":455,"n_samples":659,"n_pointer_only_licence":149}},{"rank_in_archive_order":28,"model":"RoBERTa + SubRegWeigh (K-means)","metrics":{"Accuracy":"94.84"},"uses_additional_data":false,"paper_date":"2024-09-10","paper":"/paper/subregweigh-effective-and-efficient","paper_url":"https://arxiv.org/abs/2409.06216v2","paper_title":"SubRegWeigh: Effective and Efficient Annotation Weighing with Subword Regularization","code":"https://github.com/4ldk/SubRegWeigh","n_code_links":1,"syntology":null},{"rank_in_archive_order":29,"model":"SpanBERT","metrics":{"Accuracy":"94.8"},"uses_additional_data":false,"paper_date":"2019-07-24","paper":"/paper/spanbert-improving-pre-training-by","paper_url":"https://arxiv.org/abs/1907.10529v3","paper_title":"SpanBERT: Improving Pre-training by Representing and Predicting Spans","code":"https://github.com/facebookresearch/SpanBERT","n_code_links":6,"syntology":{"n_ran":3,"n_unverified":12,"n_samples":15,"n_pointer_only_licence":4}},{"rank_in_archive_order":30,"model":"gMLP-large","metrics":{"Accuracy":"94.8"},"uses_additional_data":false,"paper_date":"2021-05-17","paper":"/paper/pay-attention-to-mlps","paper_url":"https://arxiv.org/abs/2105.08050v2","paper_title":"Pay Attention to MLPs","code":"https://github.com/labmlai/annotated_deep_learning_paper_implementations","n_code_links":20,"syntology":{"n_ran":34,"n_unverified":10,"n_samples":44,"n_pointer_only_licence":11}},{"rank_in_archive_order":31,"model":"Q-BERT (Shen et al., 2020)","metrics":{"Accuracy":"94.8"},"uses_additional_data":false,"paper_date":"2019-09-12","paper":"/paper/q-bert-hessian-based-ultra-low-precision","paper_url":"https://arxiv.org/abs/1909.05840v2","paper_title":"Q-BERT: Hessian Based Ultra Low Precision Quantization of BERT","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":32,"model":"Q8BERT (Zafrir et al., 2019)","metrics":{"Accuracy":"94.7"},"uses_additional_data":false,"paper_date":"2019-10-14","paper":"/paper/q8bert-quantized-8bit-bert","paper_url":"https://arxiv.org/abs/1910.06188v2","paper_title":"Q8BERT: Quantized 8Bit BERT","code":"https://github.com/NervanaSystems/nlp-architect/blob/master/nlp_architect/models/transformers/quantized_bert.py","n_code_links":5,"syntology":{"n_ran":3,"n_unverified":8,"n_samples":11,"n_pointer_only_licence":3}},{"rank_in_archive_order":33,"model":"CNN Large","metrics":{"Accuracy":"94.6"},"uses_additional_data":false,"paper_date":"2019-03-19","paper":"/paper/cloze-driven-pretraining-of-self-attention","paper_url":"http://arxiv.org/abs/1903.07785v1","paper_title":"Cloze-driven Pretraining of Self-attention Networks","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":34,"model":"BigBird","metrics":{"Accuracy":"94.6"},"uses_additional_data":false,"paper_date":"2020-07-28","paper":"/paper/big-bird-transformers-for-longer-sequences","paper_url":"https://arxiv.org/abs/2007.14062v2","paper_title":"Big Bird: Transformers for Longer Sequences","code":"https://github.com/huggingface/transformers","n_code_links":14,"syntology":{"n_ran":10,"n_unverified":5,"n_samples":15,"n_pointer_only_licence":11}},{"rank_in_archive_order":35,"model":"MLM+ del-word+ reorder","metrics":{"Accuracy":"94.5"},"uses_additional_data":false,"paper_date":"2020-12-31","paper":"/paper/clear-contrastive-learning-for-sentence","paper_url":"https://arxiv.org/abs/2012.15466v1","paper_title":"CLEAR: Contrastive Learning for Sentence Representation","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":36,"model":"ASA + BERT-base","metrics":{"Accuracy":"94.1"},"uses_additional_data":false,"paper_date":"2022-06-25","paper":"/paper/adversarial-self-attention-for-language","paper_url":"https://arxiv.org/abs/2206.12608v3","paper_title":"Adversarial Self-Attention for Language Understanding","code":"https://github.com/gingasan/adversarialsa","n_code_links":1,"syntology":null},{"rank_in_archive_order":37,"model":"RealFormer","metrics":{"Accuracy":"94.04"},"uses_additional_data":false,"paper_date":"2020-12-21","paper":"/paper/informer-transformer-likes-informed-attention","paper_url":"https://arxiv.org/abs/2012.11747v3","paper_title":"RealFormer: Transformer Likes Residual Attention","code":"https://github.com/google-research/google-research","n_code_links":5,"syntology":null},{"rank_in_archive_order":38,"model":"FNet-Large","metrics":{"Accuracy":"94"},"uses_additional_data":false,"paper_date":"2021-05-09","paper":"/paper/fnet-mixing-tokens-with-fourier-transforms","paper_url":"https://arxiv.org/abs/2105.03824v4","paper_title":"FNet: Mixing Tokens with Fourier Transforms","code":"https://github.com/labmlai/annotated_deep_learning_paper_implementations","n_code_links":12,"syntology":{"n_ran":2,"n_unverified":0,"n_samples":2,"n_pointer_only_licence":1}},{"rank_in_archive_order":39,"model":"MT-DNN","metrics":{"Accuracy":"93.6"},"uses_additional_data":false,"paper_date":"2019-11-08","paper":"/paper/smart-robust-and-efficient-fine-tuning-for","paper_url":"https://arxiv.org/abs/1911.03437v5","paper_title":"SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization","code":"https://github.com/namisan/mt-dnn","n_code_links":6,"syntology":{"n_ran":6,"n_unverified":2,"n_samples":8,"n_pointer_only_licence":1}},{"rank_in_archive_order":40,"model":"ERNIE","metrics":{"Accuracy":"93.5"},"uses_additional_data":false,"paper_date":"2019-05-17","paper":"/paper/ernie-enhanced-language-representation-with","paper_url":"https://arxiv.org/abs/1905.07129v3","paper_title":"ERNIE: Enhanced Language Representation with Informative Entities","code":"https://github.com/thunlp/ERNIE","n_code_links":2,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":1}},{"rank_in_archive_order":41,"model":"Block-sparse LSTM","metrics":{"Accuracy":"93.2"},"uses_additional_data":false,"paper_date":"2017-12-01","paper":"/paper/gpu-kernels-for-block-sparse-weights","paper_url":"https://blog.openai.com/block-sparse-gpu-kernels/","paper_title":"GPU Kernels for Block-Sparse Weights","code":"https://github.com/openai/blocksparse","n_code_links":1,"syntology":null},{"rank_in_archive_order":42,"model":"LM-CPPF RoBERTa-base","metrics":{"Accuracy":"93.2"},"uses_additional_data":false,"paper_date":"2023-05-29","paper":"/paper/lm-cppf-paraphrasing-guided-data-augmentation","paper_url":"https://arxiv.org/abs/2305.18169v3","paper_title":"LM-CPPF: Paraphrasing-Guided Data Augmentation for Contrastive Prompt-Based Few-Shot Fine-Tuning","code":"https://github.com/amirabaskohi/lm-cppf","n_code_links":1,"syntology":null},{"rank_in_archive_order":43,"model":"TinyBERT-6 67M","metrics":{"Accuracy":"93.1"},"uses_additional_data":false,"paper_date":"2019-09-23","paper":"/paper/190910351","paper_url":"https://arxiv.org/abs/1909.10351v5","paper_title":"TinyBERT: Distilling BERT for Natural Language Understanding","code":"https://github.com/PaddlePaddle/PaddleNLP/tree/develop/paddlenlp/transformers/tinybert","n_code_links":10,"syntology":{"n_ran":0,"n_unverified":4,"n_samples":4,"n_pointer_only_licence":4}},{"rank_in_archive_order":44,"model":"24hBERT","metrics":{"Accuracy":"93.0"},"uses_additional_data":false,"paper_date":"2021-04-15","paper":"/paper/how-to-train-bert-with-an-academic-budget","paper_url":"https://arxiv.org/abs/2104.07705v2","paper_title":"How to Train BERT with an Academic Budget","code":"https://github.com/peteriz/academic-budget-bert","n_code_links":4,"syntology":null},{"rank_in_archive_order":45,"model":"SMART+BERT-BASE","metrics":{"Accuracy":"93"},"uses_additional_data":false,"paper_date":"2019-11-08","paper":"/paper/smart-robust-and-efficient-fine-tuning-for","paper_url":"https://arxiv.org/abs/1911.03437v5","paper_title":"SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization","code":"https://github.com/namisan/mt-dnn","n_code_links":6,"syntology":{"n_ran":6,"n_unverified":2,"n_samples":8,"n_pointer_only_licence":1}},{"rank_in_archive_order":46,"model":"TinyBERT-4 14.5M","metrics":{"Accuracy":"92.6"},"uses_additional_data":false,"paper_date":"2019-09-23","paper":"/paper/190910351","paper_url":"https://arxiv.org/abs/1909.10351v5","paper_title":"TinyBERT: Distilling BERT for Natural Language Understanding","code":"https://github.com/PaddlePaddle/PaddleNLP/tree/develop/paddlenlp/transformers/tinybert","n_code_links":10,"syntology":{"n_ran":0,"n_unverified":4,"n_samples":4,"n_pointer_only_licence":4}},{"rank_in_archive_order":47,"model":"bmLSTM","metrics":{"Accuracy":"91.8"},"uses_additional_data":false,"paper_date":"2017-04-05","paper":"/paper/learning-to-generate-reviews-and-discovering","paper_url":"http://arxiv.org/abs/1704.01444v2","paper_title":"Learning to Generate Reviews and Discovering Sentiment","code":"https://github.com/openai/generating-reviews-discovering-sentiment","n_code_links":2,"syntology":{"n_ran":0,"n_unverified":1,"n_samples":1,"n_pointer_only_licence":0}},{"rank_in_archive_order":48,"model":"T5-Small","metrics":{"Accuracy":"91.8"},"uses_additional_data":false,"paper_date":"2019-10-23","paper":"/paper/exploring-the-limits-of-transfer-learning","paper_url":"https://arxiv.org/abs/1910.10683v4","paper_title":"Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer","code":"https://github.com/huggingface/transformers","n_code_links":57,"syntology":{"n_ran":2,"n_unverified":29,"n_samples":31,"n_pointer_only_licence":0}},{"rank_in_archive_order":49,"model":"byte mLSTM7","metrics":{"Accuracy":"91.7"},"uses_additional_data":false,"paper_date":"2018-05-14","paper":"/paper/a-la-carte-embedding-cheap-but-effective","paper_url":"http://arxiv.org/abs/1805.05388v1","paper_title":"A La Carte Embedding: Cheap but Effective Induction of Semantic Feature Vectors","code":"https://github.com/NLPrinceton/ALaCarte","n_code_links":1,"syntology":{"n_ran":2,"n_unverified":1,"n_samples":3,"n_pointer_only_licence":0}},{"rank_in_archive_order":50,"model":"PAR BERT Base","metrics":{"Accuracy":"91.6"},"uses_additional_data":false,"paper_date":"2020-09-09","paper":"/paper/pay-attention-when-required","paper_url":"https://arxiv.org/abs/2009.04534v3","paper_title":"Pay Attention when Required","code":"https://github.com/NVIDIA/DeepLearningExamples/tree/master/PyTorch/LanguageModeling","n_code_links":2,"syntology":null},{"rank_in_archive_order":51,"model":"Charformer-Base","metrics":{"Accuracy":"91.6"},"uses_additional_data":false,"paper_date":"2021-06-23","paper":"/paper/charformer-fast-character-transformers-via","paper_url":"https://arxiv.org/abs/2106.12672v3","paper_title":"Charformer: Fast Character Transformers via Gradient-based Subword Tokenization","code":"https://github.com/google-research/google-research","n_code_links":2,"syntology":{"n_ran":7,"n_unverified":3,"n_samples":10,"n_pointer_only_licence":0}},{"rank_in_archive_order":52,"model":"SqueezeBERT","metrics":{"Accuracy":"91.4"},"uses_additional_data":false,"paper_date":"2020-06-19","paper":"/paper/squeezebert-what-can-computer-vision-teach","paper_url":"https://arxiv.org/abs/2006.11316v1","paper_title":"SqueezeBERT: What can computer vision teach NLP about efficient neural networks?","code":"https://github.com/huggingface/transformers/blob/main/src/transformers/models/squeezebert/modeling_squeezebert.py","n_code_links":6,"syntology":{"n_ran":0,"n_unverified":1,"n_samples":1,"n_pointer_only_licence":0}},{"rank_in_archive_order":53,"model":"Nyströmformer","metrics":{"Accuracy":"91.4"},"uses_additional_data":false,"paper_date":"2021-02-07","paper":"/paper/nystromformer-a-nystrom-based-algorithm-for","paper_url":"https://arxiv.org/abs/2102.03902v3","paper_title":"Nyströmformer: A Nyström-Based Algorithm for Approximating Self-Attention","code":"https://github.com/facebookresearch/xformers","n_code_links":10,"syntology":{"n_ran":1,"n_unverified":1,"n_samples":2,"n_pointer_only_licence":1}},{"rank_in_archive_order":54,"model":"Bi-CAS-LSTM","metrics":{"Accuracy":"91.3"},"uses_additional_data":false,"paper_date":"2018-09-07","paper":"/paper/cell-aware-stacked-lstms-for-modeling","paper_url":"https://arxiv.org/abs/1809.02279v2","paper_title":"Cell-aware Stacked LSTMs for Modeling Sentences","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":55,"model":"DistilBERT 66M","metrics":{"Accuracy":"91.3"},"uses_additional_data":false,"paper_date":"2019-10-02","paper":"/paper/distilbert-a-distilled-version-of-bert","paper_url":"https://arxiv.org/abs/1910.01108v4","paper_title":"DistilBERT, a distilled version of BERT: smaller, faster, cheaper and lighter","code":"https://github.com/huggingface/transformers","n_code_links":37,"syntology":{"n_ran":19,"n_unverified":8,"n_samples":27,"n_pointer_only_licence":0}},{"rank_in_archive_order":56,"model":"CNN","metrics":{"Accuracy":"91.2"},"uses_additional_data":false,"paper_date":"2017-07-06","paper":"/paper/on-the-role-of-text-preprocessing-in-neural","paper_url":"http://arxiv.org/abs/1707.01780v3","paper_title":"On the Role of Text Preprocessing in Neural Network Architectures: An Evaluation Study on Text Categorization and Sentiment Analysis","code":"https://github.com/pedrada88/preproc-textclassification","n_code_links":3,"syntology":null},{"rank_in_archive_order":57,"model":"Suffix BiLSTM","metrics":{"Accuracy":"91.2"},"uses_additional_data":false,"paper_date":"2018-05-18","paper":"/paper/improved-sentence-modeling-using-suffix","paper_url":"http://arxiv.org/abs/1805.07340v2","paper_title":"Improved Sentence Modeling using Suffix Bidirectional LSTM","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":58,"model":"BERT Base","metrics":{"Accuracy":"91.2"},"uses_additional_data":false,"paper_date":"2019-10-04","paper":"/paper/fine-grained-sentiment-classification-using","paper_url":"https://arxiv.org/abs/1910.03474v1","paper_title":"Fine-grained Sentiment Classification using BERT","code":"https://github.com/munikarmanish/bert-sentiment","n_code_links":1,"syntology":{"n_ran":0,"n_unverified":2,"n_samples":2,"n_pointer_only_licence":0}},{"rank_in_archive_order":59,"model":"Transformer (finetune)","metrics":{"Accuracy":"90.9"},"uses_additional_data":false,"paper_date":"2018-12-04","paper":"/paper/practical-text-classification-with-large-pre","paper_url":"http://arxiv.org/abs/1812.01207v1","paper_title":"Practical Text Classification With Large Pre-Trained Language Models","code":"https://github.com/NVIDIA/sentiment-discovery","n_code_links":1,"syntology":null},{"rank_in_archive_order":60,"model":"Single layer bilstm distilled from BERT","metrics":{"Accuracy":"90.7"},"uses_additional_data":false,"paper_date":"2019-03-28","paper":"/paper/distilling-task-specific-knowledge-from-bert","paper_url":"http://arxiv.org/abs/1903.12136v1","paper_title":"Distilling Task-Specific Knowledge from BERT into Simple Neural Networks","code":"https://github.com/pvgladkov/knowledge-distillation","n_code_links":4,"syntology":{"n_ran":9,"n_unverified":0,"n_samples":9,"n_pointer_only_licence":0}},{"rank_in_archive_order":61,"model":"BCN+Char+CoVe","metrics":{"Accuracy":"90.3"},"uses_additional_data":false,"paper_date":"2017-08-01","paper":"/paper/learned-in-translation-contextualized-word","paper_url":"http://arxiv.org/abs/1708.00107v2","paper_title":"Learned in Translation: Contextualized Word Vectors","code":"https://github.com/salesforce/cove","n_code_links":5,"syntology":{"n_ran":3,"n_unverified":0,"n_samples":3,"n_pointer_only_licence":2}},{"rank_in_archive_order":62,"model":"CNN-RNF-LSTM","metrics":{"Accuracy":"90.0"},"uses_additional_data":false,"paper_date":"2018-08-28","paper":"/paper/convolutional-neural-networks-with-recurrent","paper_url":"http://arxiv.org/abs/1808.09315v1","paper_title":"Convolutional Neural Networks with Recurrent Neural Filters","code":"https://github.com/bloomberg/cnn-rnf","n_code_links":2,"syntology":null},{"rank_in_archive_order":63,"model":"Neural Semantic Encoder","metrics":{"Accuracy":"89.7"},"uses_additional_data":false,"paper_date":"2016-07-14","paper":"/paper/neural-semantic-encoders","paper_url":"http://arxiv.org/abs/1607.04315v3","paper_title":"Neural Semantic Encoders","code":"https://github.com/Smerity/keras_snli","n_code_links":3,"syntology":null},{"rank_in_archive_order":64,"model":"BLSTM-2DCNN","metrics":{"Accuracy":"89.5"},"uses_additional_data":false,"paper_date":"2016-11-21","paper":"/paper/text-classification-improved-by-integrating","paper_url":"http://arxiv.org/abs/1611.06639v1","paper_title":"Text Classification Improved by Integrating Bidirectional LSTM with Two-dimensional Max Pooling","code":"https://github.com/ManuelVs/NNForTextClassification","n_code_links":3,"syntology":null},{"rank_in_archive_order":65,"model":"CNN + Logic rules","metrics":{"Accuracy":"89.3"},"uses_additional_data":false,"paper_date":"2016-03-21","paper":"/paper/harnessing-deep-neural-networks-with-logic","paper_url":"https://arxiv.org/abs/1603.06318v6","paper_title":"Harnessing Deep Neural Networks with Logic Rules","code":"https://github.com/ZhitingHu/logicnn","n_code_links":2,"syntology":null},{"rank_in_archive_order":66,"model":"DMN [ankit16]","metrics":{"Accuracy":"88.6"},"uses_additional_data":false,"paper_date":"2015-06-24","paper":"/paper/ask-me-anything-dynamic-memory-networks-for","paper_url":"http://arxiv.org/abs/1506.07285v5","paper_title":"Ask Me Anything: Dynamic Memory Networks for Natural Language Processing","code":"https://github.com/DongjunLee/dmn-tensorflow","n_code_links":10,"syntology":{"n_ran":2,"n_unverified":3,"n_samples":5,"n_pointer_only_licence":0}},{"rank_in_archive_order":67,"model":"CNN-multichannel [kim2013]","metrics":{"Accuracy":"88.1"},"uses_additional_data":false,"paper_date":"2014-08-25","paper":"/paper/convolutional-neural-networks-for-sentence","paper_url":"http://arxiv.org/abs/1408.5882v2","paper_title":"Convolutional Neural Networks for Sentence Classification","code":"https://github.com/PaddlePaddle/PaddleNLP/tree/develop/examples/sentiment_analysis/textcnn","n_code_links":118,"syntology":{"n_ran":19,"n_unverified":58,"n_samples":77,"n_pointer_only_licence":15}},{"rank_in_archive_order":68,"model":"Consistency Tree LSTM with tuned Glove vectors [tai2015improved]","metrics":{"Accuracy":"88.0"},"uses_additional_data":false,"paper_date":"2015-02-28","paper":"/paper/improved-semantic-representations-from-tree","paper_url":"http://arxiv.org/abs/1503.00075v3","paper_title":"Improved Semantic Representations From Tree-Structured Long Short-Term Memory Networks","code":"https://github.com/dmlc/dgl/tree/master/examples/pytorch/tree_lstm","n_code_links":16,"syntology":{"n_ran":6,"n_unverified":9,"n_samples":15,"n_pointer_only_licence":6}},{"rank_in_archive_order":69,"model":"C-LSTM","metrics":{"Accuracy":"87.8"},"uses_additional_data":false,"paper_date":"2015-11-27","paper":"/paper/a-c-lstm-neural-network-for-text","paper_url":"http://arxiv.org/abs/1511.08630v2","paper_title":"A C-LSTM Neural Network for Text Classification","code":"https://github.com/zackhy/TextClassification","n_code_links":10,"syntology":null},{"rank_in_archive_order":70,"model":"MPAD-path","metrics":{"Accuracy":"87.75"},"uses_additional_data":false,"paper_date":"2019-08-17","paper":"/paper/message-passing-attention-networks-for","paper_url":"https://arxiv.org/abs/1908.06267v2","paper_title":"Message Passing Attention Networks for Document Understanding","code":"https://github.com/giannisnik/mpad","n_code_links":2,"syntology":null},{"rank_in_archive_order":71,"model":"Standard DR-AGG","metrics":{"Accuracy":"87.6"},"uses_additional_data":false,"paper_date":"2018-06-05","paper":"/paper/information-aggregation-via-dynamic-routing","paper_url":"http://arxiv.org/abs/1806.01501v1","paper_title":"Information Aggregation via Dynamic Routing for Sequence Encoding","code":"https://github.com/FudanNLP/Capsule4TextClassification","n_code_links":2,"syntology":null},{"rank_in_archive_order":72,"model":"USE_T+CNN (lrn w.e.)","metrics":{"Accuracy":"87.21"},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/universal-sentence-encoder","paper_url":"http://arxiv.org/abs/1803.11175v2","paper_title":"Universal Sentence Encoder","code":"https://github.com/facebookresearch/InferSent","n_code_links":24,"syntology":{"n_ran":1,"n_unverified":21,"n_samples":22,"n_pointer_only_licence":1}},{"rank_in_archive_order":73,"model":"Reverse DR-AGG","metrics":{"Accuracy":"87.2"},"uses_additional_data":false,"paper_date":"2018-06-05","paper":"/paper/information-aggregation-via-dynamic-routing","paper_url":"http://arxiv.org/abs/1806.01501v1","paper_title":"Information Aggregation via Dynamic Routing for Sequence Encoding","code":"https://github.com/FudanNLP/Capsule4TextClassification","n_code_links":2,"syntology":null},{"rank_in_archive_order":74,"model":"DC-MCNN","metrics":{"Accuracy":"86.99"},"uses_additional_data":false,"paper_date":"2018-07-01","paper":"/paper/a-helping-hand-transfer-learning-for-deep","paper_url":"https://aclanthology.org/P18-1235","paper_title":"A Helping Hand: Transfer Learning for Deep Sentiment Analysis","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":75,"model":"STM+TSED+PT+2L","metrics":{"Accuracy":"86.95"},"uses_additional_data":false,"paper_date":"2019-05-31","paper":"/paper/190600095","paper_url":"https://arxiv.org/abs/1906.00095v1","paper_title":"The Pupil Has Become the Master: Teacher-Student Model-Based Word Embedding Distillation with Ensemble Learning","code":"https://github.com/bgshin/distill_demo","n_code_links":1,"syntology":null},{"rank_in_archive_order":76,"model":"Capsule-B","metrics":{"Accuracy":"86.8"},"uses_additional_data":false,"paper_date":"2018-03-29","paper":"/paper/investigating-capsule-networks-with-dynamic","paper_url":"http://arxiv.org/abs/1804.00538v4","paper_title":"Investigating Capsule Networks with Dynamic Routing for Text Classification","code":"https://github.com/andyweizhao/capsule_text_classification","n_code_links":4,"syntology":null},{"rank_in_archive_order":77,"model":"2-layer LSTM [tai2015improved]","metrics":{"Accuracy":"86.3"},"uses_additional_data":false,"paper_date":"2015-02-28","paper":"/paper/improved-semantic-representations-from-tree","paper_url":"http://arxiv.org/abs/1503.00075v3","paper_title":"Improved Semantic Representations From Tree-Structured Long Short-Term Memory Networks","code":"https://github.com/dmlc/dgl/tree/master/examples/pytorch/tree_lstm","n_code_links":16,"syntology":{"n_ran":6,"n_unverified":9,"n_samples":15,"n_pointer_only_licence":6}},{"rank_in_archive_order":78,"model":"SWEM-concat","metrics":{"Accuracy":"84.3"},"uses_additional_data":false,"paper_date":"2018-05-24","paper":"/paper/baseline-needs-more-love-on-simple-word","paper_url":"http://arxiv.org/abs/1805.09843v1","paper_title":"Baseline Needs More Love: On Simple Word-Embedding-Based Models and Associated Pooling Mechanisms","code":"https://github.com/dinghanshen/SWEM","n_code_links":2,"syntology":null},{"rank_in_archive_order":79,"model":"MV-RNN","metrics":{"Accuracy":"82.9"},"uses_additional_data":false,"paper_date":"2013-10-01","paper":"/paper/recursive-deep-models-for-semantic","paper_url":"https://aclanthology.org/D13-1170","paper_title":"Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank","code":"https://github.com/stanfordnlp/CoreNLP","n_code_links":2,"syntology":null},{"rank_in_archive_order":80,"model":"GloVe+Emo2Vec","metrics":{"Accuracy":"82.3"},"uses_additional_data":false,"paper_date":"2018-09-12","paper":"/paper/emo2vec-learning-generalized-emotion","paper_url":"http://arxiv.org/abs/1809.04505v1","paper_title":"Emo2Vec: Learning Generalized Emotion Representation by Multi-task Training","code":"https://github.com/pxuab/emo2vec_wassa_paper","n_code_links":1,"syntology":null},{"rank_in_archive_order":81,"model":"Emo2Vec","metrics":{"Accuracy":"81.2"},"uses_additional_data":false,"paper_date":"2018-09-12","paper":"/paper/emo2vec-learning-generalized-emotion","paper_url":"http://arxiv.org/abs/1809.04505v1","paper_title":"Emo2Vec: Learning Generalized Emotion Representation by Multi-task Training","code":"https://github.com/pxuab/emo2vec_wassa_paper","n_code_links":1,"syntology":null},{"rank_in_archive_order":82,"model":"ToWE-CBOW","metrics":{"Accuracy":"78.8"},"uses_additional_data":false,"paper_date":"2018-08-01","paper":"/paper/task-oriented-word-embedding-for-text","paper_url":"https://aclanthology.org/C18-1172","paper_title":"Task-oriented Word Embedding for Text Classification","code":"https://github.com/qianliu0708/ToWE","n_code_links":1,"syntology":null},{"rank_in_archive_order":83,"model":"Joined Model Multi-tasking","metrics":{"Accuracy":" 54.72 "},"uses_additional_data":false,"paper_date":"2017-08-01","paper":"/paper/exploring-joint-neural-model-for-sentence","paper_url":"https://aclanthology.org/W17-5535","paper_title":"Exploring Joint Neural Model for Sentence Level Discourse Parsing and Sentiment Analysis","code":null,"n_code_links":0,"syntology":null},{"rank_in_archive_order":84,"model":"SMARTRoBERTa","metrics":{"Dev Accuracy":"96.9"},"uses_additional_data":false,"paper_date":"2019-11-08","paper":"/paper/smart-robust-and-efficient-fine-tuning-for","paper_url":"https://arxiv.org/abs/1911.03437v5","paper_title":"SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization","code":"https://github.com/namisan/mt-dnn","n_code_links":6,"syntology":{"n_ran":6,"n_unverified":2,"n_samples":8,"n_pointer_only_licence":1}},{"rank_in_archive_order":85,"model":"SMART-MT-DNN","metrics":{"Dev Accuracy":"96.1"},"uses_additional_data":false,"paper_date":"2019-11-08","paper":"/paper/smart-robust-and-efficient-fine-tuning-for","paper_url":"https://arxiv.org/abs/1911.03437v5","paper_title":"SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization","code":"https://github.com/namisan/mt-dnn","n_code_links":6,"syntology":{"n_ran":6,"n_unverified":2,"n_samples":8,"n_pointer_only_licence":1}},{"rank_in_archive_order":86,"model":"SMART-BERT","metrics":{"Dev Accuracy":"93.0"},"uses_additional_data":false,"paper_date":"2019-11-08","paper":"/paper/smart-robust-and-efficient-fine-tuning-for","paper_url":"https://arxiv.org/abs/1911.03437v5","paper_title":"SMART: Robust and Efficient Fine-Tuning for Pre-trained Natural Language Models through Principled Regularized Optimization","code":"https://github.com/namisan/mt-dnn","n_code_links":6,"syntology":{"n_ran":6,"n_unverified":2,"n_samples":8,"n_pointer_only_licence":1}},{"rank_in_archive_order":87,"model":"Word+ES (Scratch)","metrics":{"Attack Success Rate":"100"},"uses_additional_data":false,"paper_date":"2022-10-18","paper":"/paper/fine-mixing-mitigating-backdoors-in-fine","paper_url":"https://arxiv.org/abs/2210.09545v1","paper_title":"Fine-mixing: Mitigating Backdoors in Fine-tuned Language Models","code":"https://github.com/huggingface/pytorch-transformers","n_code_links":1,"syntology":{"n_ran":0,"n_unverified":19,"n_samples":19,"n_pointer_only_licence":19}}],"since_archive":{"claim":"Results that newer papers report for their own method, placed here by Syntology. A model pointed at the cell in the paper's own table; the number was read from that cell and checked against this leaderboard's metric, dataset, split and scale; an independent check that saw this leaderboard's other rows and every other leaderboard on the same dataset accepted it. Not reviewed by the paper's authors or by the archive's editors, and not ranked against the archive rows.","extraction_file_present":true,"measurement":{"test_papers":883,"papers_with_output":881,"judged_true":108,"judged":110,"wilson95_lower":0.9361,"measured_on":"2026-09-24","frozen_commit":"0e3de0df94"},"measurement_note":"blind adjudication of accepted entries on a held-out split of archive papers, rules frozen before the test","coverage":{"sentence":"Syntology has checked 6,264 of the 9,581 papers on this site that are newer than the archive; results from the others appear after they are checked.","complete":false,"papers_newer_than_archive":9581,"papers_checked":6264,"papers_extracted_not_yet_verified":0,"boards_without_verdict":2,"papers_not_yet_extracted":3316},"order":"newest first by month (arXiv date, else the arXiv-id month), then arXiv id descending","columns":[],"entries":[]},"syntology":{"read_at":"2026-09-24T18:15:14+00:00","claim":"Per row: N of M harvested code samples from that row's paper executed on a synthesized fixture; the other M-N are unverified. Not a reproduction of the row's number; not a correctness claim. n_pointer_only_licence counts samples the site points at rather than redistributes (a licence axis, independent of ran/unverified).","rows_with_graph_line":47,"rows_with_any_sample_ran":40,"distinct_papers_with_graph_line":35,"distinct_papers_with_any_sample_ran":29,"samples_over_distinct_papers":{"n_ran":456,"n_unverified":815,"n_samples":1271,"n_pointer_only_licence":300,"note":"each paper (arXiv id) counted once, however many rows it is behind; this is the page-level figure"},"samples_row_weighted":{"n_ran":510,"n_unverified":968,"n_samples":1478,"n_pointer_only_licence":318,"note":"row-weighted: a paper behind several rows is counted once per row; inflated relative to samples_over_distinct_papers by design, kept for readers summing the per-row syntology blocks"}}}