{"noiseShare":0.4513,"topics":[{"claims":[259,414,463,501,562,543,599,570,502],"deltaPp":-1.8018,"id":0,"keywords":["dialogue","conversational","task-oriented","response","user"],"label":"Task-Oriented Dialogue","labelFull":"Modeling and Evaluation of Task-Oriented Dialogue Systems","papers":2027,"papersByYear":[114,175,194,213,233,238,276,292,292],"rel":-0.2986,"reviewed":true,"share":[6.0349,8.4337,8.0066,8.105,7.6519,8.6451,5.6006,4.8336,4.2331],"size":4413,"slope":-0.003713},{"claims":[92,118,207,270,326,336,864,1171,746],"deltaPp":4.4598,"id":1,"keywords":["languages","multilingual","cross-lingual","transfer","fine-tuning"],"label":"Multilingual LLM Benchmarking","labelFull":"Multilingual and Low-Resource Benchmarking of LLMs","papers":2895,"papersByYear":[65,90,147,188,232,240,611,777,545],"rel":1.2961,"reviewed":true,"share":[3.441,4.3373,6.0669,7.1537,7.619,8.7178,12.3985,12.8621,7.9008],"size":4130,"slope":0.009607},{"claims":[246,297,358,401,494,410,568,697,535],"deltaPp":-0.3868,"id":2,"keywords":["question","answering","knowledge","generation","pairs"],"label":"Question Answering and Generation","labelFull":"Question Answering and Generation with Large Language Models","papers":2180,"papersByYear":[110,129,172,199,233,211,325,426,375],"rel":-0.0664,"reviewed":true,"share":[5.8232,6.2169,7.0986,7.5723,7.6519,7.6644,6.595,7.0518,5.4364],"size":4006,"slope":7e-06},{"claims":[403,390,525,528,494,352,531,424,353],"deltaPp":-7.1665,"id":3,"keywords":["translation","machine","neural","quality","english"],"label":"Low-Resource Neural Machine Translation","labelFull":"Low-Resource Neural Machine Translation Quality Gains","papers":1742,"papersByYear":[177,172,222,232,211,159,223,194,152],"rel":-0.7648,"reviewed":true,"share":[9.37,8.2892,9.1622,8.828,6.9294,5.7755,4.5252,3.2114,2.2035],"size":4000,"slope":-0.009371},{"claims":[75,109,141,168,230,274,694,727,788],"deltaPp":1.9606,"id":4,"keywords":["speech","audio","acoustic","spoken","asr"],"label":"Speech Language Models","labelFull":"Speech Language Models: Architectures, Training, Evaluation","papers":1201,"papersByYear":[38,52,65,71,90,114,254,243,274],"rel":0.9746,"reviewed":true,"share":[2.0116,2.506,2.6826,2.7017,2.9557,4.1409,5.1542,4.0225,3.9722],"size":3206,"slope":0.003129},{"claims":[24,36,45,66,102,145,284,697,1390],"deltaPp":11.6103,"id":5,"keywords":["reasoning","mathematical","math","chain-of-thought","benchmarks"],"label":"Chain-of-Thought Reasoning","labelFull":"Chain-of-Thought Reasoning: Benchmarks and Efficient Models","papers":1677,"papersByYear":[14,26,24,32,58,90,183,398,852],"rel":15.6663,"reviewed":true,"share":[0.7411,1.253,0.9905,1.2177,1.9048,3.2692,3.7135,6.5883,12.3514],"size":2789,"slope":0.011657},{"claims":[198,235,254,327,440,293,394,291,157],"deltaPp":-2.81,"id":6,"keywords":["summarization","abstractive","extractive","document","generated"],"label":"Summarization Evaluation","labelFull":"Metrics, Benchmarks, and Human Evaluation for Summarization","papers":1081,"papersByYear":[78,90,98,132,169,115,168,140,91],"rel":-0.6805,"reviewed":true,"share":[4.1292,4.3373,4.0446,5.0228,5.5501,4.1773,3.4091,2.3175,1.3192],"size":2589,"slope":-0.003236},{"claims":[33,47,46,107,168,146,303,448,513],"deltaPp":3.7397,"id":7,"keywords":["retrieval","query","dense","ranking","relevance"],"label":"Dense Retrieval","labelFull":"Neural Dense Retrieval for Information Retrieval","papers":1092,"papersByYear":[20,29,33,67,91,89,169,263,331],"rel":3.532,"reviewed":true,"share":[1.0588,1.3976,1.3619,2.5495,2.9885,3.2328,3.4294,4.3536,4.7985],"size":1811,"slope":0.004774},{"claims":[213,199,213,243,262,150,252,148,92],"deltaPp":-4.9106,"id":8,"keywords":["entity","named","recognition","linking","mentions"],"label":"Named Entity Recognition","labelFull":"Named Entity Recognition and Entity Linking","papers":917,"papersByYear":[107,109,103,117,143,80,127,79,52],"rel":-0.8669,"reviewed":true,"share":[5.6644,5.253,4.2509,4.4521,4.6962,2.9059,2.5771,1.3077,0.7538],"size":1772,"slope":-0.006062},{"claims":[41,76,100,119,132,142,238,369,478],"deltaPp":2.0152,"id":9,"keywords":["clinical","medical","biomedical","radiology","report"],"label":"Medical LLM Evaluation","labelFull":"Benchmarks and Evaluation for Medical Large Language Models","papers":779,"papersByYear":[23,40,51,59,61,61,106,155,223],"rel":1.6551,"reviewed":true,"share":[1.2176,1.9277,2.1048,2.2451,2.0033,2.2158,2.151,2.5658,3.2328],"size":1695,"slope":0.001673},{"claims":[282,265,247,220,226,133,179,72,52],"deltaPp":-6.5819,"id":10,"keywords":["parsing","dependency","parser","syntactic","constituency"],"label":"Dependency Parsing","labelFull":"Neural Dependency Parsing (Syntactic and Semantic)","papers":783,"papersByYear":[132,112,109,104,110,63,84,41,28],"rel":-0.9419,"reviewed":true,"share":[6.9878,5.3976,4.4986,3.9574,3.6125,2.2884,1.7045,0.6787,0.4059],"size":1676,"slope":-0.007957},{"claims":[21,33,32,53,103,119,244,425,599],"deltaPp":4.208,"id":11,"keywords":["vision-language","visual","image","navigation","reasoning"],"label":"Vision-Language Model Evaluation","labelFull":"Evaluation and Benchmarking of Vision-Language Models","papers":955,"papersByYear":[18,24,26,32,64,69,134,232,356],"rel":4.416,"reviewed":true,"share":[0.9529,1.1566,1.073,1.2177,2.1018,2.5064,2.7192,3.8404,5.1609],"size":1629,"slope":0.004911},{"claims":[210,192,219,203,196,108,196,161,70],"deltaPp":-4.6182,"id":12,"keywords":["sentiment","aspect","analysis","opinion","classification"],"label":"Aspect-Based Sentiment Analysis","labelFull":"Aspect-Based Sentiment Analysis: Models and Benchmarks","papers":692,"papersByYear":[96,93,91,90,81,54,86,69,32],"rel":-0.9087,"reviewed":true,"share":[5.0821,4.4819,3.7557,3.4247,2.6601,1.9615,1.7451,1.1422,0.4639],"size":1555,"slope":-0.005663},{"claims":[6,6,3,5,7,17,98,385,956],"deltaPp":7.6702,"id":13,"keywords":["agents","llm","systems","framework","benchmark"],"label":"LLM Agents","labelFull":"LLM-Based Autonomous Agents: Architectures, Benchmarks, and Safety","papers":839,"papersByYear":[6,6,2,4,6,13,53,198,551],"rel":24.1505,"reviewed":true,"share":[0.3176,0.2892,0.0825,0.1522,0.197,0.4722,1.0755,3.2776,7.9878],"size":1483,"slope":0.006992},{"claims":[230,230,204,140,153,92,124,115,54],"deltaPp":-6.6291,"id":14,"keywords":["embeddings","word","similarity","sentence","semantic"],"label":"Word and Sentence Embeddings","labelFull":"Learning, Aligning, and Evaluating Word/Sentence Embeddings","papers":816,"papersByYear":[137,130,122,90,92,59,78,65,43],"rel":-0.914,"reviewed":true,"share":[7.2525,6.2651,5.0351,3.4247,3.0213,2.1431,1.5828,1.076,0.6234],"size":1342,"slope":-0.008378},{"claims":[105,97,131,169,163,173,213,119,79],"deltaPp":-1.7922,"id":15,"keywords":["event","extraction","temporal","argument","detection"],"label":"Event Extraction","labelFull":"Event Extraction and Temporal Relation Extraction","papers":529,"papersByYear":[47,41,53,66,60,74,85,55,48],"rel":-0.7203,"reviewed":true,"share":[2.4881,1.9759,2.1874,2.5114,1.9704,2.688,1.7248,0.9104,0.6959],"size":1249,"slope":-0.001852},{"claims":[45,96,105,135,144,138,217,218,127],"deltaPp":-0.216,"id":16,"keywords":["text generation","natural generation","detection","evaluation","quality"],"label":"Neural Text Generation","labelFull":"Neural Text Generation: Models, Datasets, and Evaluation","papers":803,"papersByYear":[29,61,61,87,94,95,146,139,91],"rel":-0.1407,"reviewed":true,"share":[1.5352,2.9398,2.5175,3.3105,3.087,3.4508,2.9627,2.3009,1.3192],"size":1225,"slope":-0.000292},{"claims":[37,60,60,77,97,92,154,276,345],"deltaPp":1.9983,"id":17,"keywords":["low-rank","fine-tuning","gradient","parameters","layers"],"label":"Parameter-Efficient Fine-Tuning","labelFull":"Parameter-Efficient Fine-Tuning for Large Language Models","papers":927,"papersByYear":[34,54,52,63,78,69,118,197,262],"rel":1.1102,"reviewed":true,"share":[1.7999,2.6024,2.1461,2.3973,2.5616,2.5064,2.3945,3.261,3.7982],"size":1198,"slope":0.001762},{"claims":[79,105,96,112,132,117,139,159,180],"deltaPp":-1.4914,"id":18,"keywords":["state-of-the-art","outperforms","baselines","benchmark","superior"],"label":"State-of-the-Art Benchmarking","labelFull":"Achieving State-of-the-Art Performance on Benchmark Datasets","papers":1063,"papersByYear":[75,95,92,106,127,114,135,148,171],"rel":-0.3756,"reviewed":true,"share":[3.9704,4.5783,3.7969,4.0335,4.1708,4.1409,2.7394,2.4499,2.479],"size":1119,"slope":-0.002393},{"claims":[15,14,47,51,63,95,102,222,410],"deltaPp":3.5688,"id":19,"keywords":["token","reduces"],"label":"Token Efficiency","labelFull":"Token Efficiency and Reduction in Language Models","papers":731,"papersByYear":[12,12,34,36,44,70,78,155,290],"rel":5.6175,"reviewed":true,"share":[0.6353,0.5783,1.4032,1.3699,1.445,2.5427,1.5828,2.5658,4.2041],"size":1019,"slope":0.003628},{"claims":[94,101,100,103,95,65,78,183,198],"deltaPp":-1.3736,"id":20,"keywords":["attention","heads","recurrent","layers","transformer"],"label":"Efficient Attention","labelFull":"Efficient Attention Mechanisms for Transformer Computation","papers":641,"papersByYear":[61,61,65,63,58,42,53,110,128],"rel":-0.4254,"reviewed":true,"share":[3.2292,2.9398,2.6826,2.3973,1.9048,1.5256,1.0755,1.8209,1.8556],"size":1017,"slope":-0.002156},{"claims":[22,27,23,14,15,16,77,158,640],"deltaPp":5.5707,"id":21,"keywords":["reward","reinforcement","policy","learning","optimization"],"label":"Reward Modeling for LLMs","labelFull":"Reward Modeling and Policy Optimization for LLM Alignment","papers":718,"papersByYear":[18,23,20,13,13,14,63,104,450],"rel":5.846,"reviewed":true,"share":[0.9529,1.1084,0.8254,0.4947,0.4269,0.5085,1.2784,1.7216,6.5236],"size":992,"slope":0.004174},{"claims":[96,71,130,90,93,58,165,130,135],"deltaPp":-1.2143,"id":22,"keywords":["emotion","recognition","affective","cause","multimodal"],"label":"Emotion Detection","labelFull":"Emotion Detection and Classification in Text","papers":396,"papersByYear":[38,31,47,41,42,23,58,61,55],"rel":-0.6036,"reviewed":true,"share":[2.0116,1.494,1.9397,1.5601,1.3793,0.8355,1.1769,1.0098,0.7973],"size":968,"slope":-0.001427},{"claims":[70,94,79,62,77,81,203,150,84],"deltaPp":-1.2541,"id":23,"keywords":["argument","essay","mining","scoring","automated"],"label":"Argument Mining","labelFull":"Argument Mining, Quality Assessment, Counterargument Gen","papers":365,"papersByYear":[33,35,36,28,32,32,82,53,34],"rel":-0.7179,"reviewed":true,"share":[1.747,1.6867,1.4858,1.0654,1.0509,1.1624,1.664,0.8773,0.4929],"size":900,"slope":-0.001165},{"claims":[50,49,111,116,120,138,133,89,65],"deltaPp":-0.8198,"id":24,"keywords":["nlp","nli","nlg","nlu","research"],"label":"NLP Model Evaluation","labelFull":"Evaluation Benchmarks, Toolkits, and Best Practices for NLP","papers":587,"papersByYear":[30,36,68,72,79,93,94,62,53],"rel":-0.5162,"reviewed":true,"share":[1.5881,1.7349,2.8064,2.7397,2.5944,3.3781,1.9075,1.0263,0.7683],"size":871,"slope":-0.001094},{"claims":[68,93,98,107,143,91,116,70,25],"deltaPp":-2.2272,"id":25,"keywords":["text classification","multi-label text","hierarchical","few-shot","document"],"label":"Text Classification","labelFull":"Text Classification: State-of-the-Art Models and Benchmarks","papers":505,"papersByYear":[47,54,70,67,85,57,65,42,18],"rel":-0.8951,"reviewed":true,"share":[2.4881,2.6024,2.889,2.5495,2.7915,2.0705,1.319,0.6952,0.2609],"size":811,"slope":-0.003042},{"claims":[53,29,81,46,76,64,154,112,183],"deltaPp":-0.0144,"id":26,"keywords":["story","narrative","generation","visual","human"],"label":"Story Generation","labelFull":"Story Generation, Datasets, and Evaluation","papers":361,"papersByYear":[23,11,35,25,36,31,57,60,83],"rel":-0.0118,"reviewed":true,"share":[1.2176,0.5301,1.4445,0.9513,1.1823,1.126,1.1567,0.9932,1.2032],"size":798,"slope":0.000155},{"claims":[0,0,5,14,93,102,192,218,167],"deltaPp":1.8266,"id":27,"keywords":["prompt","tuning","soft","optimization","multiple"],"label":"Prompt Optimization","labelFull":"Automatic Prompt Optimization for Large Language Models","papers":550,"papersByYear":[0,0,3,6,51,67,141,156,126],"rel":null,"reviewed":true,"share":[0.0,0.0,0.1238,0.2283,1.6749,2.4337,2.8612,2.5824,1.8266],"size":791,"slope":0.003789},{"claims":[54,74,86,66,118,76,114,102,73],"deltaPp":-0.8967,"id":28,"keywords":["knowledge graph","graph completion","link prediction","embedding","entities"],"label":"Knowledge Graph Completion","labelFull":"Knowledge Graph Completion and Embedding Approaches","papers":451,"papersByYear":[32,48,47,43,60,42,63,61,55],"rel":-0.5293,"reviewed":true,"share":[1.694,2.3133,1.9397,1.6362,1.9704,1.5256,1.2784,1.0098,0.7973],"size":763,"slope":-0.001488},{"claims":[54,78,81,119,132,75,90,72,50],"deltaPp":-1.6094,"id":29,"keywords":["label","domain","supervised","unlabeled","contrastive"],"label":"Label-Efficient Text Classification","labelFull":"Label-Efficient Text Classification via Contrastive Learning","papers":628,"papersByYear":[43,67,66,100,104,65,78,59,46],"rel":-0.707,"reviewed":true,"share":[2.2763,3.2289,2.7239,3.8052,3.4154,2.3611,1.5828,0.9767,0.6669],"size":751,"slope":-0.00282},{"claims":[7,13,37,66,124,104,77,121,196],"deltaPp":1.4655,"id":30,"keywords":["distillation","student","teacher","knowledge","on-policy"],"label":"Language Model Knowledge Distillation","labelFull":"Knowledge Distillation Methods for Efficient Language Model Compression","papers":447,"papersByYear":[6,7,26,35,69,55,50,76,123],"rel":4.6143,"reviewed":true,"share":[0.3176,0.3373,1.073,1.3318,2.266,1.9978,1.0146,1.2581,1.7831],"size":745,"slope":0.001529},{"claims":[0,0,0,9,14,17,122,286,278],"deltaPp":2.6529,"id":31,"keywords":["retrieval-augmented generation","generation systems","framework","multimodal","knowledge"],"label":"Retrieval-Augmented Generation","labelFull":"Retrieval-Augmented Generation for Knowledge-Intensive Tasks","papers":469,"papersByYear":[0,0,0,3,11,8,78,186,183],"rel":null,"reviewed":true,"share":[0.0,0.0,0.0,0.1142,0.3612,0.2906,1.5828,3.079,2.6529],"size":726,"slope":0.003865},{"claims":[37,50,74,51,69,64,142,111,99],"deltaPp":-0.4958,"id":32,"keywords":["annotation","agreement","crowdsourcing","human","quality"],"label":"Inter-Annotator Agreement","labelFull":"Inter-Annotator Agreement and Annotation Quality Assurance","papers":524,"papersByYear":[31,39,57,40,46,48,97,87,79],"rel":-0.3021,"reviewed":true,"share":[1.6411,1.8795,2.3525,1.5221,1.5107,1.7436,1.9683,1.4402,1.1453],"size":697,"slope":-0.000641},{"claims":[8,6,12,25,41,51,114,167,263],"deltaPp":2.3353,"id":33,"keywords":["code","program","test","execution","generation"],"label":"Code Generation Benchmarks","labelFull":"Benchmarks For Evaluating LLM Code Generation","papers":437,"papersByYear":[6,3,9,13,24,26,64,109,183],"rel":7.353,"reviewed":true,"share":[0.3176,0.1446,0.3714,0.4947,0.7882,0.9444,1.2987,1.8043,2.6529],"size":687,"slope":0.002771},{"claims":[24,23,48,85,84,85,101,126,93],"deltaPp":0.3172,"id":34,"keywords":["explanations","rationales","causal","faithfulness","human"],"label":"Faithful Natural Language Explanations","labelFull":"Faithful Natural Language Explanations for Model Predictions","papers":389,"papersByYear":[14,16,26,41,43,48,57,71,73],"rel":0.428,"reviewed":true,"share":[0.7411,0.7711,1.073,1.5601,1.4122,1.7436,1.1567,1.1753,1.0583],"size":669,"slope":0.000472},{"claims":[54,95,76,80,117,81,65,73,21],"deltaPp":-1.6694,"id":35,"keywords":["relation","extraction","document-level","distant","supervision"],"label":"Relation Extraction","labelFull":"Relation Extraction (Document-Level & Distant Supervision)","papers":347,"papersByYear":[34,52,44,42,57,40,35,34,9],"rel":-0.9275,"reviewed":true,"share":[1.7999,2.506,1.8159,1.5982,1.8719,1.453,0.7102,0.5628,0.1305],"size":662,"slope":-0.002477},{"claims":[71,66,76,61,76,59,107,79,45],"deltaPp":-1.3505,"id":36,"keywords":["political","tweets","misinformation","media","twitter"],"label":"Political Tweet Classification","labelFull":"Political Tweet Classification: Stance, Ideology, Propaganda","papers":327,"papersByYear":[34,34,38,26,36,27,60,41,31],"rel":-0.7503,"reviewed":true,"share":[1.7999,1.6386,1.5683,0.9893,1.1823,0.9807,1.2175,0.6787,0.4494],"size":640,"slope":-0.001499},{"claims":[2,0,3,13,12,40,132,186,238],"deltaPp":1.9332,"id":37,"keywords":["hallucination","detection","vision-language","reduces","mitigates"],"label":"LLM Hallucination Detection","labelFull":"Hallucination Detection and Mitigation in LLMs","papers":347,"papersByYear":[1,0,3,6,8,20,69,103,137],"rel":36.5444,"reviewed":true,"share":[0.0529,0.0,0.1238,0.2283,0.2627,0.7265,1.4002,1.705,1.9861],"size":626,"slope":0.00265},{"claims":[21,21,32,28,32,26,95,161,208],"deltaPp":1.3275,"id":38,"keywords":["judgments","human","evaluation","metrics","llm-as-a-judge"],"label":"LLM-as-a-Judge Evaluation","labelFull":"LLM-as-a-Judge Evaluation and Human Alignment Metrics","papers":475,"papersByYear":[16,19,26,25,25,22,79,113,150],"rel":1.5673,"reviewed":true,"share":[0.847,0.9157,1.073,0.9513,0.821,0.7991,1.6031,1.8706,2.1745],"size":624,"slope":0.001514},{"claims":[1,3,3,7,10,44,139,195,203],"deltaPp":2.2231,"id":39,"keywords":["reasoning","logical","capabilities","inference","natural"],"label":"LLM Reasoning","labelFull":"Reasoning Capabilities and Methods for Large Language Models","papers":452,"papersByYear":[1,3,3,6,10,33,88,151,157],"rel":42.0246,"reviewed":true,"share":[0.0529,0.1446,0.1238,0.2283,0.3284,1.1987,1.7857,2.4996,2.276],"size":605,"slope":0.003375},{"claims":[39,48,43,43,42,45,83,125,129],"deltaPp":-0.0099,"id":40,"keywords":["citation","review","scientific","peer","literature"],"label":"Citation Generation","labelFull":"Citation Generation for Scientific Writing","papers":326,"papersByYear":[21,26,23,24,28,22,42,64,76],"rel":-0.0089,"reviewed":true,"share":[1.1117,1.253,0.9492,0.9132,0.9195,0.7991,0.8523,1.0594,1.1018],"size":597,"slope":-0.000155},{"claims":[10,10,21,20,25,34,56,127,287],"deltaPp":3.0413,"id":41,"keywords":["gpu","routing","latency","memory","reduces"],"label":"Efficient LLM Inference","labelFull":"Efficient Large Language Model Inference and Deployment","papers":499,"papersByYear":[8,9,18,19,22,29,44,111,239],"rel":7.1813,"reviewed":true,"share":[0.4235,0.4337,0.7429,0.723,0.7225,1.0534,0.8929,1.8374,3.4648],"size":590,"slope":0.002834},{"claims":[74,68,57,76,58,47,93,73,40],"deltaPp":-1.6064,"id":42,"keywords":["topic","modeling","coherence","neural","document"],"label":"Neural Topic Modeling","labelFull":"Neural Topic Modeling with Coherence and Diversity Enhancements","papers":260,"papersByYear":[35,29,26,33,28,22,38,32,17],"rel":-0.867,"reviewed":true,"share":[1.8528,1.3976,1.073,1.2557,0.9195,0.7991,0.7711,0.5297,0.2464],"size":586,"slope":-0.001682},{"claims":[28,27,32,51,39,51,78,135,138],"deltaPp":0.5996,"id":43,"keywords":["calibration","uncertainty","released","analysis","provides"],"label":"Uncertainty Calibration","labelFull":"Calibration and Quantification of Uncertainty in NLP Models","papers":522,"papersByYear":[24,26,26,47,36,47,67,120,129],"rel":0.4719,"reviewed":true,"share":[1.2705,1.253,1.073,1.7884,1.1823,1.7072,1.3596,1.9864,1.8701],"size":579,"slope":0.000848},{"claims":[60,64,62,82,97,49,62,57,40],"deltaPp":-1.3556,"id":44,"keywords":["paraphrase","generation","syntactic","identification","unsupervised"],"label":"Neural Paraphrase Generation","labelFull":"Paraphrase Generation: Supervised, Unsupervised, Controlled","papers":292,"papersByYear":[33,34,31,35,44,20,37,31,27],"rel":-0.776,"reviewed":true,"share":[1.747,1.6386,1.2794,1.3318,1.445,0.7265,0.7508,0.5132,0.3914],"size":573,"slope":-0.001744},{"claims":[31,39,58,63,83,56,58,74,89],"deltaPp":-0.1397,"id":45,"keywords":["qa","state-of-the-art","benchmarks","open-domain","outperforms"],"label":"Question Answering Benchmarks","labelFull":"State-of-the-Art Performance on Question Answering Benchmarks","papers":439,"papersByYear":[24,33,45,49,60,41,47,62,78],"rel":-0.11,"reviewed":true,"share":[1.2705,1.5904,1.8572,1.8645,1.9704,1.4893,0.9537,1.0263,1.1308],"size":551,"slope":-0.000739},{"claims":[17,68,43,30,40,53,96,113,91],"deltaPp":0.4608,"id":46,"keywords":["gender","bias","cultural","stereotypes","demographic"],"label":"Gender Bias in LLMs","labelFull":"Gender Bias Evaluation & Mitigation in Large Language Models","papers":312,"papersByYear":[8,31,17,17,25,30,55,68,61],"rel":1.0881,"reviewed":true,"share":[0.4235,1.494,0.7016,0.6469,0.821,1.0897,1.1161,1.1256,0.8843],"size":551,"slope":0.000335},{"claims":[3,2,8,30,10,20,66,158,248],"deltaPp":2.1411,"id":47,"keywords":["attack","success","jailbreak","backdoor","defense"],"label":"Jailbreak Attacks on LLMs","labelFull":"Jailbreak Attacks and Defenses for Large Language Models","papers":346,"papersByYear":[2,2,6,15,9,14,41,102,155],"rel":20.2181,"reviewed":true,"share":[0.1059,0.0964,0.2476,0.5708,0.2956,0.5085,0.832,1.6885,2.247],"size":545,"slope":0.002408},{"claims":[8,12,41,61,87,89,71,96,76],"deltaPp":0.4898,"id":48,"keywords":["few-shot","zero-shot","segmentation","settings","learning"],"label":"Few-Shot & Zero-Shot Classification","labelFull":"Few-Shot and Zero-Shot Text Classification","papers":443,"papersByYear":[8,11,31,42,72,74,65,77,63],"rel":1.1566,"reviewed":true,"share":[0.4235,0.5301,1.2794,1.5982,2.3645,2.688,1.319,1.2746,0.9133],"size":541,"slope":0.000894},{"claims":[19,39,44,77,66,29,103,91,69],"deltaPp":0.0032,"id":49,"keywords":["adversarial","attacks","examples","text","backdoor"],"label":"Adversarial Text Attacks","labelFull":"Adversarial Attacks and Defense Strategies for NLP Models","papers":368,"papersByYear":[15,23,28,49,45,20,66,67,55],"rel":0.004,"reviewed":true,"share":[0.7941,1.1084,1.1556,1.8645,1.4778,0.7265,1.3393,1.1091,0.7973],"size":537,"slope":-0.000126},{"claims":[1,1,10,11,22,23,112,210,136],"deltaPp":1.3388,"id":50,"keywords":["bias","cultural","social","exhibit","alignment"],"label":"Cultural Bias in LLMs","labelFull":"Cultural and Social Bias Evaluation and Mitigation in LLMs","papers":342,"papersByYear":[1,1,7,8,16,17,72,124,96],"rel":25.3081,"reviewed":true,"share":[0.0529,0.0482,0.2889,0.3044,0.5255,0.6175,1.461,2.0526,1.3917],"size":526,"slope":0.002338},{"claims":[53,35,64,52,59,51,81,95,19],"deltaPp":-0.9377,"id":51,"keywords":["correction","grammatical error","spelling","chinese","check"],"label":"Grammatical Error Correction","labelFull":"Neural Grammatical Error Correction for English and Chinese","papers":213,"papersByYear":[21,17,28,21,21,20,35,38,12],"rel":-0.8435,"reviewed":true,"share":[1.1117,0.8193,1.1556,0.7991,0.6897,0.7265,0.7102,0.629,0.174],"size":509,"slope":-0.000881},{"claims":[22,22,30,35,33,36,59,99,157],"deltaPp":0.5844,"id":52,"keywords":["decoding","speculative","minimum","encoder","risk"],"label":"Speculative Decoding","labelFull":"Speculative Decoding Techniques for Efficient LLM Inference","papers":375,"papersByYear":[21,19,23,29,28,31,42,65,117],"rel":0.5257,"reviewed":true,"share":[1.1117,0.9157,0.9492,1.1035,0.9195,1.126,0.8523,1.076,1.6961],"size":493,"slope":0.000441},{"claims":[57,48,68,92,54,46,60,34,29],"deltaPp":-1.2843,"id":53,"keywords":["coreference","resolution","cross-document","event","anaphora"],"label":"Coreference Resolution","labelFull":"Coreference Resolution: Models, Datasets, and Benchmarks","papers":222,"papersByYear":[27,24,30,44,24,19,26,18,10],"rel":-0.8986,"reviewed":true,"share":[1.4293,1.1566,1.2381,1.6743,0.7882,0.6902,0.5276,0.298,0.145],"size":488,"slope":-0.001686},{"claims":[49,81,65,60,40,37,73,42,31],"deltaPp":-0.9422,"id":54,"keywords":["discourse","relation","implicit","treebank","parsing"],"label":"Discourse Parsing","labelFull":"Discourse Parsing and Implicit Relation Classification","papers":239,"papersByYear":[23,40,32,33,21,19,31,21,19],"rel":-0.7738,"reviewed":true,"share":[1.2176,1.9277,1.3207,1.2557,0.6897,0.6902,0.6291,0.3476,0.2754],"size":478,"slope":-0.001743},{"claims":[84,83,61,44,46,24,58,49,14],"deltaPp":-2.2133,"id":55,"keywords":["morphological","inflection","tagging","lemmatization","part-of-speech"],"label":"Morphological Analysis","labelFull":"Neural Morphological Analysis and Generation","papers":246,"papersByYear":[44,41,33,23,26,15,31,25,8],"rel":-0.9502,"reviewed":true,"share":[2.3293,1.9759,1.3619,0.8752,0.8539,0.5449,0.6291,0.4138,0.116],"size":463,"slope":-0.002556},{"claims":[26,33,22,41,38,36,78,81,107],"deltaPp":0.307,"id":56,"keywords":["claim","evidence","verification","fact-checking","scientific"],"label":"Automated Fact-Checking","labelFull":"Fact-Checking: Detection, Evidence Retrieval, Verification","papers":240,"papersByYear":[12,13,13,20,14,19,40,44,65],"rel":0.4832,"reviewed":true,"share":[0.6353,0.6265,0.5365,0.761,0.4598,0.6902,0.8117,0.7284,0.9423],"size":462,"slope":0.000336},{"claims":[57,66,71,58,47,19,67,40,28],"deltaPp":-1.6978,"id":57,"keywords":["reading","comprehension","readability","machine","assessment"],"label":"Machine Reading Comprehension","labelFull":"Machine Reading Comprehension Models, Datasets & Benchmarks","papers":283,"papersByYear":[37,42,47,31,32,14,33,29,18],"rel":-0.8668,"reviewed":true,"share":[1.9587,2.0241,1.9397,1.1796,1.0509,0.5085,0.6696,0.4801,0.2609],"size":453,"slope":-0.002439},{"claims":[4,16,13,25,16,31,89,121,128],"deltaPp":1.3394,"id":58,"keywords":["reinforcement learning","game","agents","text-based","reward"],"label":"Multi-Agent LLM Reinforcement Learning","labelFull":"Multi-Agent RL for Large Language Model Agents","papers":336,"papersByYear":[4,11,9,14,10,23,68,90,107],"rel":6.3239,"reviewed":true,"share":[0.2118,0.5301,0.3714,0.5327,0.3284,0.8355,1.3799,1.4898,1.5512],"size":443,"slope":0.001759},{"claims":[17,17,27,52,35,28,73,97,92],"deltaPp":0.4325,"id":59,"keywords":["visual question","question answering","video","image","medical"],"label":"Knowledge-Based VQA","labelFull":"Knowledge-Based Visual Question Answering","papers":281,"papersByYear":[11,10,13,23,23,17,43,71,70],"rel":0.7427,"reviewed":true,"share":[0.5823,0.4819,0.5365,0.8752,0.7553,0.6175,0.8726,1.1753,1.0148],"size":438,"slope":0.000704},{"claims":[44,46,58,77,49,36,68,48,8],"deltaPp":-1.4482,"id":60,"keywords":["sense","disambiguation","word","semantic change","lexical"],"label":"Word Sense Disambiguation","labelFull":"Neural Word Sense Disambiguation and Sense Embedding Methods","papers":240,"papersByYear":[29,26,32,38,27,19,38,25,6],"rel":-0.9433,"reviewed":true,"share":[1.5352,1.253,1.3207,1.446,0.8867,0.6902,0.7711,0.4138,0.087],"size":434,"slope":-0.001694},{"claims":[9,22,48,34,52,30,46,98,81],"deltaPp":0.1854,"id":61,"keywords":["graph","node","networks","text-attributed","convolutional"],"label":"NLP Graph Neural Networks","labelFull":"Graph Neural Networks for Natural Language Processing","papers":275,"papersByYear":[8,16,36,25,38,22,33,55,42],"rel":0.4378,"reviewed":true,"share":[0.4235,0.7711,1.4858,0.9513,1.2479,0.7991,0.6696,0.9104,0.6089],"size":420,"slope":-0.000104},{"claims":[18,17,35,43,48,33,50,78,95],"deltaPp":0.3121,"id":62,"keywords":["unlearning","forgetting","multi-task","catastrophic","learning"],"label":"Multi-Task Continual Learning","labelFull":"Multi-Task Continual Learning with Forgetting Mitigation","papers":319,"papersByYear":[13,17,29,36,32,26,42,55,69],"rel":0.4535,"reviewed":true,"share":[0.6882,0.8193,1.1969,1.3699,1.0509,0.9444,0.8523,0.9104,1.0003],"size":417,"slope":6.8e-05},{"claims":[32,12,20,24,35,29,73,93,98],"deltaPp":0.1199,"id":63,"keywords":["multimodal","modality","fusion","unimodal","representations"],"label":"Multimodal Model Evaluation","labelFull":"Evaluation and Benchmarking of Multimodal Models","papers":308,"papersByYear":[18,6,14,17,28,24,53,74,74],"rel":0.1258,"reviewed":true,"share":[0.9529,0.2892,0.5778,0.6469,0.9195,0.8718,1.0755,1.225,1.0728],"size":416,"slope":0.000751},{"claims":[30,33,39,47,56,44,71,64,29],"deltaPp":-0.6056,"id":64,"keywords":["stance","rumor","detection","zero-shot","social"],"label":"Social Media Stance Detection","labelFull":"Social Media Stance Detection for Rumor Verification","papers":168,"papersByYear":[15,11,17,19,20,15,28,30,13],"rel":-0.7626,"reviewed":true,"share":[0.7941,0.5301,0.7016,0.723,0.6568,0.5449,0.5682,0.4966,0.1885],"size":413,"slope":-0.000495},{"claims":[24,33,29,26,32,35,49,67,108],"deltaPp":0.1054,"id":65,"keywords":["user","response","persona","personalization","feedback"],"label":"Persona-Based Dialogue Generation","labelFull":"Persona-Based Dialogue Response Generation","papers":309,"papersByYear":[18,26,23,25,21,28,41,54,73],"rel":0.1106,"reviewed":true,"share":[0.9529,1.253,0.9492,0.9513,0.6897,1.0171,0.832,0.8939,1.0583],"size":403,"slope":-0.000137},{"claims":[25,39,38,52,86,52,53,33,22],"deltaPp":-0.6346,"id":66,"keywords":["ner","nested","few-shot","chinese","state-of-the-art"],"label":"Low-Resource NER","labelFull":"Low-Resource NER Across Domains and Languages","papers":205,"papersByYear":[15,19,20,25,40,27,31,17,11],"rel":-0.7991,"reviewed":true,"share":[0.7941,0.9157,0.8254,0.9513,1.3136,0.9807,0.6291,0.2814,0.1595],"size":400,"slope":-0.000801},{"claims":[40,31,36,55,38,45,59,50,40],"deltaPp":-0.7783,"id":67,"keywords":["captioning","image","video","metrics","visual"],"label":"Image Captioning","labelFull":"Image Captioning: Models, Datasets, and Evaluation Metrics","papers":197,"papersByYear":[21,15,17,25,17,21,28,30,23],"rel":-0.7001,"reviewed":true,"share":[1.1117,0.7229,0.7016,0.9513,0.5583,0.7628,0.5682,0.4966,0.3334],"size":394,"slope":-0.000708},{"claims":[5,14,20,25,62,24,94,87,58],"deltaPp":0.2906,"id":68,"keywords":["hate","speech detection","meme","implicit","classification"],"label":"Implicit Hate Speech Detection","labelFull":"Implicit Hate Speech Detection and Cross‑Dataset Evaluation","papers":188,"papersByYear":[3,8,12,13,28,11,45,37,31],"rel":1.83,"reviewed":true,"share":[0.1588,0.3855,0.4953,0.4947,0.9195,0.3996,0.9131,0.6125,0.4494],"size":389,"slope":0.000431},{"claims":[19,23,35,34,25,38,30,99,83],"deltaPp":0.1999,"id":69,"keywords":["text-to-sql","queries","schema","database","spider"],"label":"Text-to-SQL","labelFull":"Text-to-SQL Parsing and Benchmarking","papers":185,"papersByYear":[8,14,15,14,11,18,13,49,43],"rel":0.472,"reviewed":true,"share":[0.4235,0.6747,0.6191,0.5327,0.3612,0.6538,0.2638,0.8111,0.6234],"size":386,"slope":0.000103},{"claims":[33,31,43,36,26,29,119,41,22],"deltaPp":-0.9132,"id":70,"keywords":["corpus","parallel","corpora","cognate","sentence"],"label":"Parallel Corpus Release","labelFull":"Public Release of Parallel Corpora and Associated Tools","papers":276,"papersByYear":[23,27,29,28,22,20,76,30,21],"rel":-0.75,"reviewed":true,"share":[1.2176,1.3012,1.1969,1.0654,0.7225,0.7265,1.5422,0.4966,0.3044],"size":380,"slope":-0.000952},{"claims":[2,7,17,19,45,35,66,70,100],"deltaPp":0.9959,"id":71,"keywords":["curriculum","in-context","learning","difficulty","student"],"label":"In-Context Curriculum Learning","labelFull":"Curriculum Learning for In-Context Learning in LLMs","papers":271,"papersByYear":[2,7,15,15,32,24,46,54,76],"rel":9.4042,"reviewed":true,"share":[0.1059,0.3373,0.6191,0.5708,1.0509,0.8718,0.9334,0.8939,1.1018],"size":361,"slope":0.001097},{"claims":[9,18,18,11,30,36,59,97,79],"deltaPp":0.5371,"id":72,"keywords":["counseling","empathy","mental","health","medical"],"label":"Therapeutic Dialogue Systems","labelFull":"Empathetic Interaction In Therapeutic Dialogue Systems","papers":173,"papersByYear":[3,6,8,5,13,15,30,45,48],"rel":3.3822,"reviewed":true,"share":[0.1588,0.2892,0.3302,0.1903,0.4269,0.5449,0.6088,0.7449,0.6959],"size":357,"slope":0.000738},{"claims":[2,17,18,11,13,29,85,101,78],"deltaPp":0.561,"id":73,"keywords":["editing","knowledge","recipe","outperforms","text"],"label":"Knowledge Editing","labelFull":"Knowledge Editing for Large Language Models","papers":182,"papersByYear":[2,12,9,7,9,15,33,49,46],"rel":5.2975,"reviewed":true,"share":[0.1059,0.5783,0.3714,0.2664,0.2956,0.5449,0.6696,0.8111,0.6669],"size":354,"slope":0.000636},{"claims":[15,48,42,68,46,20,45,36,30],"deltaPp":-0.0662,"id":74,"keywords":["style","transfer","text","content","formality"],"label":"Text Style Transfer","labelFull":"Neural Text Style Transfer: Models, Datasets, and Evaluation","papers":164,"papersByYear":[7,19,19,28,22,10,21,17,21],"rel":-0.1786,"reviewed":true,"share":[0.3706,0.9157,0.7842,1.0654,0.7225,0.3632,0.4261,0.2814,0.3044],"size":350,"slope":-0.000598},{"claims":[1,6,9,12,12,31,48,93,134],"deltaPp":1.2228,"id":75,"keywords":["quantization","compression","weight","post-training","accuracy"],"label":"LLM Quantization","labelFull":"Quantization and Compression Techniques for Large Language Models","papers":222,"papersByYear":[1,5,7,10,9,15,29,58,88],"rel":23.1153,"reviewed":true,"share":[0.0529,0.241,0.2889,0.3805,0.2956,0.5449,0.5885,0.9601,1.2757],"size":346,"slope":0.001302},{"claims":[6,9,5,21,32,22,59,92,96],"deltaPp":0.4261,"id":76,"keywords":["privacy","private","differential","leakage","anonymization"],"label":"Privacy-Preserving Language Models","labelFull":"Privacy-Preserving Language Models: Evaluation, Fine-Tuning","papers":169,"papersByYear":[4,4,3,11,18,11,29,45,44],"rel":2.0118,"reviewed":true,"share":[0.2118,0.1928,0.1238,0.4186,0.5911,0.3996,0.5885,0.7449,0.6379],"size":342,"slope":0.000712},{"claims":[22,35,32,60,45,31,50,33,34],"deltaPp":-0.6384,"id":77,"keywords":["adversarial","examples","robustness","attacks","samples"],"label":"Adversarial Training","labelFull":"Adversarial Training and Example Generation for NLP Robustness","papers":268,"papersByYear":[20,25,25,47,35,21,36,30,29],"rel":-0.6029,"reviewed":true,"share":[1.0588,1.2048,1.0318,1.7884,1.1494,0.7628,0.7305,0.4966,0.4204],"size":342,"slope":-0.001051},{"claims":[16,31,21,28,15,18,67,94,50],"deltaPp":-0.2917,"id":78,"keywords":["recommendation","user","news","item","llm-based"],"label":"LLM-based Generative Recommendation","labelFull":"LLM-based Generative Recommendation: Methods, Applications","papers":169,"papersByYear":[14,12,12,14,9,9,25,43,31],"rel":-0.3936,"reviewed":true,"share":[0.7411,0.5783,0.4953,0.5327,0.2956,0.3269,0.5073,0.7118,0.4494],"size":340,"slope":-0.000158},{"claims":[1,2,2,0,3,9,42,127,152],"deltaPp":1.5708,"id":79,"keywords":["multimodal","evaluation","visual","reveals","understanding"],"label":"Multimodal LLM Benchmarks","labelFull":"Evaluation Benchmarks for Multimodal Large Language Models","papers":235,"papersByYear":[1,2,2,0,3,4,26,85,112],"rel":29.6938,"reviewed":true,"share":[0.0529,0.0964,0.0825,0.0,0.0985,0.1453,0.5276,1.4071,1.6237],"size":338,"slope":0.001875},{"claims":[2,8,13,8,28,28,61,86,94],"deltaPp":0.6285,"id":80,"keywords":["legal","court","judgment prediction","case","law"],"label":"Legal Judgment Prediction","labelFull":"Legal Judgment Prediction Using Large Language Models","papers":150,"papersByYear":[1,5,5,5,12,14,26,35,47],"rel":11.8809,"reviewed":true,"share":[0.0529,0.241,0.2064,0.1903,0.3941,0.5085,0.5276,0.5794,0.6814],"size":328,"slope":0.000748},{"claims":[16,8,23,18,19,23,63,70,83],"deltaPp":-0.0894,"id":81,"keywords":["arabic","modern standard","dialects","english","released"],"label":"Arabic NLP Benchmarks","labelFull":"Benchmarks, Datasets, Evaluation for Arabic Language Models","papers":144,"papersByYear":[11,5,10,9,10,9,26,30,34],"rel":-0.1535,"reviewed":true,"share":[0.5823,0.241,0.4127,0.3425,0.3284,0.3269,0.5276,0.4966,0.4929],"size":323,"slope":0.000104},{"claims":[14,21,29,41,62,38,60,38,13],"deltaPp":-0.2401,"id":82,"keywords":["intent","slot","detection","filling","discovery"],"label":"Intent Detection","labelFull":"Open, OOD, Multi-Intent Detection with Joint Slot Filling","papers":168,"papersByYear":[7,12,16,23,32,15,31,23,9],"rel":-0.6479,"reviewed":true,"share":[0.3706,0.5783,0.6603,0.8752,1.0509,0.5449,0.6291,0.3807,0.1305],"size":316,"slope":-0.000324},{"claims":[1,12,48,68,37,27,54,37,23],"deltaPp":0.2515,"id":83,"keywords":["transformer","layers","architecture","inference","pretrained"],"label":"Efficient Transformers","labelFull":"Efficient Transformers: Architecture, Training, and Compression","papers":227,"papersByYear":[1,9,29,48,29,20,41,29,21],"rel":4.7543,"reviewed":true,"share":[0.0529,0.4337,1.1969,1.8265,0.9524,0.7265,0.832,0.4801,0.3044],"size":307,"slope":-0.000114},{"claims":[24,20,31,26,48,40,36,34,46],"deltaPp":-0.5999,"id":84,"keywords":["information extraction","open information","document","systems","state-of-the-art"],"label":"Unified Information Extraction","labelFull":"Unified Information Extraction for Diverse IE Tasks","papers":193,"papersByYear":[19,15,19,17,29,23,20,23,28],"rel":-0.5964,"reviewed":true,"share":[1.0058,0.7229,0.7842,0.6469,0.9524,0.8355,0.4058,0.3807,0.4059],"size":305,"slope":-0.000666},{"claims":[27,38,32,17,27,29,64,30,38],"deltaPp":-0.235,"id":85,"keywords":["sarcasm","humor","detection","multimodal","recognition"],"label":"Multimodal Sarcasm Detection","labelFull":"Multimodal Sarcasm Detection and Benchmarking","papers":104,"papersByYear":[8,13,11,5,10,10,23,11,13],"rel":-0.5549,"reviewed":true,"share":[0.4235,0.6265,0.454,0.1903,0.3284,0.3632,0.4667,0.1821,0.1885],"size":302,"slope":-0.000346},{"claims":[18,22,22,38,21,11,34,56,44],"deltaPp":-0.2671,"id":86,"keywords":["embedding","space","semantic"],"label":"Semantic Embedding Learning","labelFull":"Techniques for Improving Semantic Embedding Spaces","papers":214,"papersByYear":[16,20,18,26,19,10,27,38,40],"rel":-0.3153,"reviewed":true,"share":[0.847,0.9639,0.7429,0.9893,0.624,0.3632,0.5479,0.629,0.5799],"size":266,"slope":-0.000515},{"claims":[6,7,15,14,25,21,32,52,85],"deltaPp":0.8712,"id":87,"keywords":["inference","speed","faster","latency","speedup"],"label":"Efficient Inference","labelFull":"Efficient Inference: Speedup, Latency, and Memory Optimization","papers":239,"papersByYear":[6,6,13,12,23,19,30,48,82],"rel":2.7431,"reviewed":true,"share":[0.3176,0.2892,0.5365,0.4566,0.7553,0.6902,0.6088,0.7946,1.1888],"size":257,"slope":0.000897},{"claims":[7,6,9,9,18,24,34,60,79],"deltaPp":0.4601,"id":88,"keywords":["video","temporal","understanding","grounding","frames"],"label":"Long Video Understanding","labelFull":"Long Video Understanding and Temporal Grounding","papers":153,"papersByYear":[5,4,6,6,13,12,22,35,50],"rel":1.7382,"reviewed":true,"share":[0.2647,0.1928,0.2476,0.2283,0.4269,0.4359,0.4464,0.5794,0.7248],"size":246,"slope":0.000601},{"claims":[7,8,22,27,30,28,33,49,40],"deltaPp":0.1992,"id":89,"keywords":["fact","verification","factuality","checking","evidence"],"label":"Automated Fact-Checking","labelFull":"Fact Verification and Evidence Retrieval in Fact-Checking","papers":179,"papersByYear":[5,5,14,19,23,22,26,33,32],"rel":0.7526,"reviewed":true,"share":[0.2647,0.241,0.5778,0.723,0.7553,0.7991,0.5276,0.5463,0.4639],"size":244,"slope":0.000281},{"claims":[30,32,28,17,18,37,55,22,4],"deltaPp":-0.7506,"id":90,"keywords":["simplification","sentence","text","lexical","automatic"],"label":"Text Simplification","labelFull":"Neural Text Simplification: Models, Datasets, and Evaluation","papers":105,"papersByYear":[15,11,12,6,14,14,19,11,3],"rel":-0.9452,"reviewed":true,"share":[0.7941,0.5301,0.4953,0.2283,0.4598,0.5085,0.3856,0.1821,0.0435],"size":243,"slope":-0.000664},{"claims":[1,0,0,4,5,25,64,83,60],"deltaPp":0.6719,"id":91,"keywords":["instruction","tuning","benchmarks","code","llms"],"label":"Instruction Tuning","labelFull":"Instruction Tuning for Large Language Models","papers":185,"papersByYear":[1,0,0,2,4,15,49,64,50],"rel":12.7013,"reviewed":true,"share":[0.0529,0.0,0.0,0.0761,0.1314,0.5449,0.9943,1.0594,0.7248],"size":242,"slope":0.001387},{"claims":[7,10,15,12,21,22,25,45,84],"deltaPp":0.6152,"id":92,"keywords":["alignment","seq2seq","sequence","accuracy","outperforms"],"label":"LLM Alignment","labelFull":"Alignment Methods for Large Language Models","papers":203,"papersByYear":[7,10,11,11,16,19,22,39,68],"rel":1.66,"reviewed":true,"share":[0.3706,0.4819,0.454,0.4186,0.5255,0.6902,0.4464,0.6456,0.9858],"size":241,"slope":0.000535},{"claims":[10,23,32,39,47,20,40,17,8],"deltaPp":-0.3749,"id":93,"keywords":["natural","understanding","inference","processing","benchmark"],"label":"NLU Benchmarks","labelFull":"Benchmarks for Multi-Task Natural Language Understanding","papers":198,"papersByYear":[9,20,27,33,34,18,34,16,7],"rel":-0.7869,"reviewed":true,"share":[0.4764,0.9639,1.1143,1.2557,1.1166,0.6538,0.6899,0.2649,0.1015],"size":236,"slope":-0.000841},{"claims":[8,6,7,20,18,27,43,54,51],"deltaPp":0.2768,"id":94,"keywords":["document","layout","understanding","long","information"],"label":"Document Layout Understanding","labelFull":"Multimodal Document Layout Understanding for Long Documents","papers":159,"papersByYear":[6,4,5,14,14,17,22,36,41],"rel":0.8715,"reviewed":true,"share":[0.3176,0.1928,0.2064,0.5327,0.4598,0.6175,0.4464,0.5959,0.5944],"size":234,"slope":0.00048},{"claims":[13,24,20,28,28,22,41,39,19],"deltaPp":-0.1726,"id":95,"keywords":["offensive","abusive","detection","deception","cyberbullying"],"label":"Abusive Language Detection","labelFull":"Abusive Language Detection Using Large Language Models","papers":112,"papersByYear":[6,11,11,13,12,10,19,20,10],"rel":-0.5435,"reviewed":true,"share":[0.3176,0.5301,0.454,0.4947,0.3941,0.3632,0.3856,0.3311,0.145],"size":234,"slope":-0.000259},{"claims":[0,2,4,1,2,5,36,92,90],"deltaPp":0.9713,"id":96,"keywords":["preference","optimization","direct","supervised fine-tuning","alignment"],"label":"Direct Preference Optimization","labelFull":"Direct Preference Optimization and Variants for LLM Alignment","papers":175,"papersByYear":[0,2,4,1,2,3,24,72,67],"rel":null,"reviewed":true,"share":[0.0,0.0964,0.1651,0.0381,0.0657,0.109,0.487,1.1919,0.9713],"size":232,"slope":0.001314},{"claims":[0,5,2,4,2,10,62,62,80],"deltaPp":0.4349,"id":97,"keywords":["watermarking","steganography","linguistic","robustness","detection"],"label":"LLM Watermarking","labelFull":"Robust Watermarking and Detection of LLM-Generated Text","papers":75,"papersByYear":[0,2,1,1,1,2,19,19,30],"rel":null,"reviewed":true,"share":[0.0,0.0964,0.0413,0.0381,0.0328,0.0726,0.3856,0.3145,0.4349],"size":227,"slope":0.000519},{"claims":[4,7,14,16,26,26,33,44,55],"deltaPp":0.595,"id":98,"keywords":["generation","quality","baselines","higher","outperforms"],"label":"Quality-Aware Text Generation","labelFull":"Improving Text Generation Quality Via Efficient Fine-Tuning","papers":202,"papersByYear":[3,7,14,14,22,21,28,41,52],"rel":3.7469,"reviewed":true,"share":[0.1588,0.3373,0.5778,0.5327,0.7225,0.7628,0.5682,0.6787,0.7538],"size":225,"slope":0.000603}],"totals":{"claims":[7769,8863,10502,11604,13416,12415,22305,28437,34566],"papers":[1889,2075,2423,2628,3045,2753,4928,6041,6898]},"years":[2018,2019,2020,2021,2022,2023,2024,2025,2026]}