Review of Research on Chart Question Answering

doi:10.11871/jfdc.issn.2096-742X.2025.01.002

Abstract

Abstract:

[Objective] The purpose of this paper is to comprehensively review the research progress of Chart Question Answering (CQA) technology, analyze existing models and methods, and explore future development directions. [Methods] Firstly, CQA models are divided into two categories: deep learning-based and multi-modal large models. Deep learning-based methods are further subdivided into end-to-end models and two-stage models in this paper. Subsequently, the three core processes taken by the deep learning-based CQA are deeply analyzed, and a detailed classification along with an in-depth analysis of the existing processing methods for each process is provided. CQA models based on multi-modal large models are also explored in this paper, with their advantages, limitations, and future development directions being analyzed. [Results] The current research status of CQA technology is comprehensively summarized, and an in-depth analysis of existing models and methods is conducted. It is found that deep learning-based CQA models perform well in handling standard chart types and simple tasks, but fall short when facing complex, non-standardized charts or tasks requiring deep reasoning. In contrast, CQA models based on multi-modal large models show great potential, but the improvement in model performance often comes with an increase in model size and computational complexity. Future research should focus on developing more lightweight question answering models and enhancing model interpretability.

Key words: chart question answering, visual question answering, deep learning, multi-modal large language models

MA Qiuping, ZHANG Qi, ZHAO Xiaofan. Review of Research on Chart Question Answering[J]. Frontiers of Data and Computing, 2025, 7(1): 19-37, https://cstr.cn/32002.14.jfdc.CN10-1649/TP.2025.01.002.

Figures/Tables 9

Fig.1

Fig.2

Fig.3

Fig.4

Fig.5

Fig.6

Fig.7

Table 1

Table 2

Evaluation results of chart question answering models on datasets"

		FigureQA		DVQA(No OCR/Oracle)		PlotQA			ChartQA			OpenCQA
		Val1	Val2	Familiar	Novel	Test1	Test2	Avg	Human	Machine	Avg	OpenCQA
基于深度学习的图表问答	RN^[13]	76.39	72.54	-	-	-	-	-	-	-	-	-
	MOM^[14]	-	-	45.03	40.9	-	-	-	-	-	-	-
	ARN^[16]	85.48	82.95	44.5/79.43	44.51/79.58	-	-	-	-	-	-	-
	SANDY^[14]	-	-	36.02/56.48	36.14/56.62	-	-	-	-	-	-	-
	LEAF-Net^[15]	-	81.15	-/72.72	-/72.89	-	-	-	-	-	-	-
	PReFIL^[37]	94.84	93.26	47.7/96.37	47.86/96.53	-	-	-	-	-	-	-
	PlotQA-M^[25]	-	-	57.99	59.54	-	-	22.52	-	-	36.15	-
	CRCT^[50]	94.61	85.04	-	-	79.64	34.44	-	-	-	-	-
	VL-T5^[6]	88.6	88.49	-/94.8	-/77.04	75.9	56.02	-	26.24	56.88	41.56	-
	VisionTaPas^[6]	91.46	91.45	-/95.38	-/95.46	65.3	42.5	-	29.6	61.44	45.52	-
	ChartT5^[45]	-	-	-	-	-	-	-	31.8	74.4	53.16	-
	GoT-CQA^[56]	-	-	-	-	92.8	78.3	-	47.1	87.9	67.5	-
	LWR-RN^[29]	85.91	83.43	44.76/79.83	44.49/79.96	-	-	-	-	-	-	-
	FIBT^[73]	-	-	-	-	-	60.44	-	-	-	-	-
	STL-CQA^[46]	-	-	-/97.35	-/97.51	-	-	-	-	-	-	-
基于大模型的图表问答	QDCHART^[17]	-	-	-	-	-	-	-	34.9	79.4	57.2	-
	Matcha^[18]	-	-	-	-	92.3	90.7	-	38.2	90.2	64.2	-
	Unichart^[19]	-	-	-	-	-	-	-	43.92	88.56	66.24	14.88
	CCM+GPT-3.5+reph PoT SC^[10]	-	-	-	-	62.0	71.4	66.7	67.6	91.4	79.5	-
	ChartLLaMa^[74]	-	-	-	-	-	-	-	48.96	90.36	69.66	-
	ChartAst-D^[68]	-	-	-	-	-	-	-	45.3	91.3	68.3	14.9
	ChartAst-S^[68]	-	-	-	-	-	-	-	65.9	93.9	75.1	15.5
	ChartGemma^[63]	-	-	-	-	-	-	-	64.8	89.44	77.12	-
	ChartInstruct-Flan-T5-XL^[75]	-	-	-	-	-	-	-	50.16	93.84	72	14.81
	ChartMoE^[72]	-	-	-	-	-	-	-	71.36	91.04	81.2	-
	ChartReader-TaPas^[76]	91.12	91.4	-/92.2	-/94.3	74.2	56.2	-	-	-	48.3	-
	EvoChart-4B^[77]	-	-	-	-	-	-	-	-	-	81.5	-
	mChartQA(Intern-LM2)^[78]	96.06	96.3	-	-	78.25	74.79	-	68.24	89.76	79	-
	mChartQA(Qwen)^[78]	90.32	92.75	-	-	78	62.95	-	58.56	93.44	76	-
	ChartReader-T5^[76]	95.5	95.8	-/95.4	-/96.5	78.1	59.3	-	-	-	49.5	-
	SIMPLOT^[79]	-	-	-	-	-	-	-	78.07	88.42	83.24	-
	SynChart^[69]	-	-	-	-	-	-	-	74.24	94.96	84.6	-
	TinyChart@768^[80]	-	-	-	-	-	-	-	73.34	93.86	83.6	-
	ChartPaLI-5B(Conv)^[66]	96.3	96.2	-/86	-/70.7	-	-	-	60.88	93.68	77.28
	VPAgent^[81]	-	-	-/98.5	-/97.1	92.6	78.1	-	48.1	87.4	67.8	-

Table 2

References 81

[1]	AGRAWAL A, LU J, ANTOL S, et al. VQA: Visual Question Answering[J]. International Journal of Computer Vision, 2015, 123 (1): 4-31. DOI: 10.1109/ICCV.2015.279.
[2]	KIM J, SRINIVASAN A, KIM N W, et al. Exploring Chart Question Answering for Blind and Low Vision Users[J]. Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems: ACM, 2023: (4), Hamburg Germany: 1-15.
[3]	BADAM S K, LIU Z, ELMQVIST N. Elastic Documents: Coupling Text and Tables through Contextual Visualizations for Enhanced Document Reading[J]. IEEE transactions on visualization and computer graphics, 2019, 25(1): 661-671.
[4]	SETLUR V, TORY M, DJALALI A. Inferencing underspecified natural language utterances in visual analysis[C]// Proceedings of the 24th International Conference on Intelligent User Interfaces, Marina del Ray California: ACM, 2019: 40-51.
[5]	KIM D H, HOQUE E, KIM J, et al. Facilitating document reading by linking text and tables[C]// Proceedings of the 31st Annual ACM Symposium on User Interface Software and Technology, Berlin Germany: ACM, 2018: 423-434.
[6]	MASRY A, LONG D X, TAN J Q, et al. Chartqa: A benchmark for question answering about charts with visual and logical reasoning[J]. arXiv preprint arXiv:2203.10244, 2022.
[7]	LUO J, LI Z, WANG J, et al. Chartocr: Data extraction from charts images via a deep hybrid framework[C]// Proceedings of the IEEE/CVF winter conference on applications of computer vision, 2021: 1917-1925.
[8]	RAFFEL C, SHAZEER N, ROBERTS A, et al. Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer[J]. Journal of Machine Learning Research, 2020, 21(140): 1-67.
[9]	HERZIG J, NOWAK P K, MÜLLER T, et al. TAPAS: Weakly Supervised Table Parsing via Pre-training[C]// Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, Online, 2020: 4320-4333.
[10]	LIU Y C, CHU W T. Chart Question Answering based on Modality Conversion and Large Language Models[C]// Proceedings of the 1st ACM Workshop on AI-Powered Q&A Systems for Multimedia, 2024: 19-24.
[11]	LIU Z, LIN Y, CAO Y, et al. Swin Transformer: Hierarchical Vision Transformer Using Shifted Windows[C]. Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV), 2021: 10012-10022.
[12]	LEWIS M, LIU Y, GOYAL N, et al. BART: Denoising Sequence-to-Sequence Pre-training for Natural Language Generation, Translation, and Comprehension[C/OL]//Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. Online: Association for Computational Linguistics, 2020: 7871-7880. https://www.aclweb.org/anthology/2020.acl-main.703.
[13]	KAHOU S E, MICHALSKI V, ATKINSON A, et al. Figureqa: An annotated figure dataset for visual reasoning[J]. arXiv preprint arXiv:1710.07300, 2017.
[14]	KAFLE K, PRICE B, COHEN S, et al. Dvqa: Understanding data visualizations via question answering[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2018: 5648-5656.
[15]	CHAUDHRY R, SHEKHAR S, GUPTA U, et al. Leaf-qa: Locate, encode & attend for figure question answering[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2020: 3512-3521.
[16]	ZOU J, WU G, XUE T, et al. An affinity-driven relation network for figure question answering[C]// 2020 IEEE International Conference on Multimedia and Expo (ICME), IEEE, 2020: 1-6.
[17]	ZHENG H, WANG S, THOMAS C, et al. Advancing Chart Question Answering with Robust Chart Component Recognition[J]. arXiv preprint arXiv:2407.21038, 2024.
[18]	LIU F, PICCINNO F, KRICHENE S, et al. MatCha: Enhancing Visual Language Pretraining with Math Reasoning and Chart Derendering[J]. arXiv preprint arXiv:2212.09662, 2022.
[19]	MASRY A, KAVEHZADEH P, DO X L, et al. Unichart: A universal vision-language pretrained model for chart comprehension and reasoning[J]. arXiv preprint arXiv:2305.14761, 2023.
[20]	WU Q, TENEY D, WANG P, et al. Visual question answering: A survey of methods and datasets[J]. Computer Vision and Image Understanding, 2017, 163: 21-40.
[21]	王虞, 孙海春. 视觉问答技术研究综述[J]. 计算机科学与探索, 2023, 17(7): 1487-1505. doi: 10.3778/j.issn.1673-9418.2303025
[22]	张一飞, 孟春运, 蒋洲, 等. 可解释的视觉问答研究进展[J]. 计算机应用研究, 2024, 41(1): 10-20.
[23]	WU A, WANG Y, SHU X, et al. Ai4vis: Survey on artificial intelligence approaches for data visualization[J]. IEEE Transactions on Visualization and Computer Graphics, 2021, 28(12): 5049-5070.
[24]	HOQUE E, KAVEHZADEH P, MASRY A. Chart question answering: State of the art and future directions[C]// Computer Graphics Forum, 2022, 41(3): 555-572.
[25]	METHANI N, GANGULY P, KHAPRA M M, et al. Plotqa: Reasoning over scientific plots[C]// Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision, 2020: 1527-1536.
[26]	YANG Z, HE X, GAO J, et al. Stacked attention networks for image question answering[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2016: 21-29.
[27]	KIM J H, JUN J, ZHANG B T. Bilinear attention networks[C]// Proceedings of the 32nd International Conference on Neural Information Processing Systems, 2018: 1571-1581.
[28]	SINGH A, NATARAJAN V, SHAH M, et al. Towards vqa models that can read[C]// Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2019: 8317-8326.
[29]	黎颖, 吴清锋, 刘佳桐, 等. 引导性权重驱动的图表问答重定位关系网络[J]. 中国图象图形学报, 2023, 28(2): 510-521.
[30]	RONNEBERGER O, FISCHER P, BROX T. U-net: Convolutional networks for biomedical image segmentation[C]// Medical image computing and computer-assisted intervention-MICCAI 2015: 18th international conference, Munich, Germany, October 5-9, 2015, proceedings, part III 18. Springer International Publishing, 2015: 234-241.
[31]	SANTORO A, RAPOSO D, BARRETT D G T, et al. A simple neural network module for relational reasoning[C]// Proceedings of the 31st International Conference on Neural Information Processing Systems, Red Hook, NY, USA: Curran Associates Inc., 2017: 4974-4983.
[32]	MASRY A, PRINCE E H. Integrating image data extraction and table parsing methods for chart question answering[C]// Chart Question Answering Workshop, in conjunction with the Conference on Computer Vision and Pattern Recognition (CVPR), 2021: 1-5.
[33]	REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE transactions on pattern analysis and machine intelligence, 2016, 39(6): 1137-1149.
[34]	SMITH R. An overview of the Tesseract OCR engine[C]// Ninth international conference on document analysis and recognition (ICDAR 2007), IEEE, 2007, 2: 629-633.
[35]	PASUPAT P, LIANG P. Compositional semantic parsing on semi-structured tables[J]. arXiv preprint arXiv:1508.00305, 2015. http://aclweb.org/anthology/P15-1142.
[36]	JAIN H, JAYARAMAN S, SOORYANATH I T, et al. TapasQA-Question Answering on Statistical Plots Using Google TAPAS[C]// International Conference on Image Processing and Capsule Networks, Cham: Springer International Publishing, 2022: 63-77.
[37]	KIM D H, HOQUE E, AGRAWALA M. Answering questions about charts and generating visual explanations[C]// Proceedings of the 2020 CHI conference on human factors in computing systems, 2020: 1-13.
[38]	SATYANARAYAN A, MORITZ D, WONGSUPHASAWAT K, et al. Vega-lite: A grammar of interactive graphics[J]. IEEE transactions on visualization and computer graphics, 2016, 23(1): 341-350.
[39]	MAZRAEH FARAHANI A, ADIBI P, EHSANI M S, et al. Chart Question Answering with Multimodal Graph Representation Learning[A/OL]. Rochester, NY, 2023[2024-10-05]. https://www.ssrn.com/abstract=4655474. DOI:10.2139/ssrn.4655474.
[40]	HE K, GKIOXARI G, DOLLÁR P, et al. Mask r-cnn[C]// Proceedings of the IEEE international conference on computer vision, 2017: 2961-2969.
[41]	HOWARD A G, ZHU M, CHEN B, et al. Mobilenets: Efficient convolutional neural networks for mobile vision applications[J]. arXiv preprint arXiv:1704.04861, 2017.
[42]	DEVLIN J, CHANG M W, LEE K, et al. Bert: Pre-training of deep bidirectional transformers for language understanding[C]// Proceedings of naacL-HLT. 2019, 1: 2.
[43]	KAFLE K, SHRESTHA R, PRICE B, et al. Answering questions about data visualizations using efficient bimodal fusion[C]// Proceedings of the IEEE/CVF Winter conference on applications of computer vision, 2020: 1498-1507.
[44]	HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]// Proceedings of the IEEE conference on computer vision and pattern recognition, 2017: 4700-4708.
[45]	ZHOU M, FUNG Y R, CHEN L, et al. Enhanced chart understanding in vision and language task via cross-modal pre-training on plot table pairs[J]. arXiv preprint arXiv:2305.18641, 2023.
[46]	SINGH H, SHEKHAR S. STL-CQA: Structure-based transformers with localization and encoding for chart question answering[C]// Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2020: 3275-3284.
[47]	GUPTA A, GUPTA V, ZHANG S, et al. Enhancing Question Answering on Charts Through Effective Pre-training Tasks[J]. arXiv preprint arXiv:2406.10085, 2024.
[48]	HOCHREITER S, SCHMIDHUBER J. Long short-term memory[J]. Neural Computation, 1997, 9 (8) : 1735-1780. doi: 10.1162/neco.1997.9.8.1735 pmid: 9377276
[49]	LIPTON Z C, BERKOWITZ J, ELKAN C. A Critical Review of Recurrent Neural Networks for Sequence Learning[J/OL]. arXiv Preprint, CoRR, abs/1506.00019, 2015: 1-1. http://arxiv.org/abs/1506.00019.
[50]	LEVY M, BEN-ARI R, LISCHINSKI D. Classification-regression for chart comprehension[C]// European Conference on Computer Vision. Cham: Springer Nature Switzerland, 2022: 469-484.
[51]	VASWANI A, SHAZEER N, PARMAR N. Attention is all you need[C]. NIPS’17: Proceedings of the 31st International Conference on Neural Information Processing Sys-tems, Red Hook, NY, USA: Curran Associates Inc., 2017: 6000-6010.
[52]	LIU Y, GU J, GOYAL N, et al. Multilingual denoising pre-training for neural machine translation[J/OL]. Transactions of the Association for Computational Linguistics, 2020, 8: 726-742. https://api.semanticscholar.org/CorpusID:210861178.
[53]	MANNING C, SURDEANU M, BAUER J, et al. The Stanford CoreNLP natural language processing toolkit[C]// Proceedings of 52nd annual meeting of the association for computational linguistics:system demonstrations, 2014: 55-60.
[54]	ZHANG Y, PASUPAT P, LIANG P. Macro grammars and holistic triggering for efficient semantic parsing[C]. Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing. Copenhagen, Denmark: Association for Computational Linguistics, 2017: 1214-1223.
[55]	LIU C, YU J, GUO Y, et al. Breathing New Life into Existing Visualizations: A Natural Language-Driven Manipulation Framework[J]. arXiv preprint arXiv:2404.06039, 2024.
[56]	ZHANG L, HUANG M, WANG Q, et al. GoT-CQA: Graph-of-Thought Guided Compositional Reasoning for Chart Question Answering[J]. arXiv preprint arXiv:2409.02611, 2024.
[57]	REDDY R, RAMESH R, DESHPANDE A, et al. Figurenet: A deep learning model for question-answering on scientific plots[C]// 2019 International Joint Conference on Neural Networks (IJCNN), IEEE, 2019: 1-8.
[58]	CORTES C, VAPNIK V. Support-Vector Networks[J]. Machine Learning, 1995, 20(3): 273-297.
[59]	SHARMA M, GUPTA S, CHOWDHURY A, et al. Chartnet: Visual reasoning over statistical charts using mac-networks[C]// 2019 International Joint Conference on Neural Networks (IJCNN), IEEE, 2019: 1-7.
[60]	FENG J. Chart Understanding with Large Language Model[J/OL]. Engineering Archive, 2023[2024-10-05]. https://engrxiv.org/preprint/view/3401/version/4747. DOI: 10.31224/3401.
[61]	RADFORD A, KIM J W, HALLACY C, et al. Learning transferable visual models from natural language supervision[C]// International conference on machine learning, PMLR, 2021: 8748-8763.
[62]	CHIANG W L, LI Z, LIN Z, et al. Vicuna: An open-source chatbot impressing gpt-4 with 90%* chatgpt quality[J]. See https://vicuna.lmsys.org (accessed 14 April 2023), 2023, 2(3): 6.
[63]	MASRY A, THAKKAR M, BAJAJ A, et al. Chartgemma: Visual instruction-tuning for chart reasoning in the wild[J]. arXiv preprint arXiv:2407.04172, 2024.
[64]	CHEN X, WANG X, BEYER L, et al. Pali-3 vision language models: Smaller, faster, stronger[J]. arXiv preprint arXiv:2310.09199, 2023.
[65]	FORD J, ZHAO X, SCHUMACHER D, et al. Charting the Future: Using Chart Question-Answering for Scalable Evaluation of LLM-Driven Data Visualizations[J]. arXiv preprint arXiv:2409.18764, 2024.
[66]	CARBUNE V, MANSOOR H, LIU F, et al. Chart-based reasoning: Transferring capabilities from llms to vlms[C]// Findings of the Association for Computational Linguistics: NAACL 2024, 2024: 989-1004.
[67]	WU Y, YAN L, SHEN L, et al. ChartInsights: Evaluating Multimodal Large Language Models for Low-Level Chart Question Answering[C]// Findings of the Association for Computational Linguistics: EMNLP 2024, 2024: 12174-12200.
[68]	MENG F, SHAO W, LU Q, et al. Chartassisstant: A universal chart multimodal language model via chart-to-table pre-training and multitask instruction tuning[J]. arXiv preprint arXiv:2401.02384, 2024.
[69]	LIU M, LI Q, CHEN D, et al. SynChart: Synthesizing Charts from Language Models[J]. arXiv preprint arXiv:2409.16517, 2024.
[70]	LI Z, JASANI B, TANG P, et al. Synthesize Step-by-Step: Tools Templates and LLMs as Data Generators for Reasoning-Based Chart VQA[C]// Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2024: 13613-13623.
[71]	ZENG X, LIN H, YE Y, et al. Advancing Multimodal Large Language Models in Chart Question Answering with Visualization-Referenced Instruction Tuning[J]. IEEE Transactions on Visualization and Computer Graphics, 2024 (1): 1-11.
[72]	XU Z, QU B, QI Y, et al. ChartMoE: Mixture of Expert Connector for Advanced Chart Understanding[J]. arXiv preprint arXiv:2409.03277, 2024.
[73]	BHAISAHEB S, PALIWAL S, PATIL R, et al. Program Synthesis for Complex QA on Charts via Probabilistic Grammar Based Filtered Iterative Back-Translation[C]// Findings of the Association for Computational Linguistics: EACL, 2023: 2501-2515.
[74]	HAN Y, ZHANG C, CHEN X, et al. Chartllama: A multimodal llm for chart understanding and generation[J]. arXiv preprint arXiv:2311.16483, 2023.
[75]	MASRY A, SHAHMOHAMMADI M, PARVEZ M R, et al. ChartInstruct: Instruction Tuning for Chart Comprehension and Reasoning[J]. arXiv preprint arXiv:2403.09028, 2024.
[76]	CHENG Z Q, DAI Q, HAUPTMANN A G. Chartreader: A unified framework for chart derendering and comprehension without heuristic rules[C]// Proceedings of the IEEE/CVF International Conference on Computer Vision, 2023: 22202-22213.
[77]	HUANG M, HAN L, ZHANG X, et al. EvoChart: A Benchmark and a Self-Training Approach Towards Real-World Chart Understanding[J]. arXiv preprint arXiv:2409.01577, 2024.
[78]	WEI J, XU N, CHANG G, et al. mChartQA: A universal benchmark for multimodal Chart Question Answer based on Vision-Language Alignment and Reasoning[J]. arXiv preprint arXiv:2404.01548, 2024.
[79]	KIM W, PARK S, IN Y, et al. SIMPLOT: Enhancing Chart Question Answering by Distilling Essentials[J]. arXiv preprint arXiv:2405.00021, 2024.
[80]	ZHANG L, HU A, XU H, et al. Tinychart: Efficient chart understanding with visual token merging and program-of-thoughts learning[J]. arXiv preprint arXiv:2404.16635, 2024.
[81]	HUANG M, ZHANG L, HAN L, et al. VProChart: Answering Chart Question through Visual Perception Alignment Agent and Programmatic Solution Reasoning[J]. arXiv preprint arXiv:2409.01667, 2024.

数据集	创建日期	图表类型	问题类型	说明
FigureQA	2018年	折线图、条形图、饼图	结构查询	数据来源于随机抽样，图表由Bokeh绘图库绘制，问题基于模板生成，答案仅包含Yes/No两种
DVQA	2018年	条形图	结构理解、数据检索、推理	数据来源于随机抽样，图表由Matplotlib绘图库绘制，问题基于模板生成
LEAF-QA	2019年	条形图、饼图、箱形图、散点图、折线图	结构理解、推理	数据来源各种在线的统计表，然后使用Matplotlib库生成各种风格的统计图，问题基于模板生成
PlotQA	2020年	条形图、折线图、散点图	结构理解、数据检索、推理	图表数据来自真实的统计数据，问题和答案基于众包收集和人工模板生成，涵盖了多种图表类型、复杂问题和开放词汇答案
ChartQA	2022年	条形图、折线图、饼图	结构理解、数据检索、推理	图表来源与真实场景，问题由人工标注和T5模型生成两种方式生成