基于智能体的可交互数据结构和算法可视化实现

doi:10.3969/j.issn.1000-5641.2025.05.004

摘要/Abstract

摘要：

数据结构与算法 (Data Structure and Algorithm, DSA) 作为计算机教育的核心课程, 在培养学生的编程能力与算法思维方面起着关键作用. 可视化在DSA教学中可以起到提高教学效率和加深学生理解的重要作用. 然而, 现有的DSA可视化工具大多依赖人工编写可视化代码, 存在覆盖范围有限、更新成本高和缺乏交互性等局限性, 难以满足动态演示与个性化教学的需求. 随着大型语言模型 (Large Language Model, LLM) 在代码生成方面的出色表现, 基于LLM实现自动化的DSA可视化成为可能. 为此, 提出了一种基于ReAct (Reasoning and Acting)智能体的交互式可视化代码生成方法, 旨在解决传统可视化工具自动化程度低、交互性不足的问题. 该方法结合LLM的代码生成能力和DSV (Data Structure Visualization)平台的接口, 能够将基于Python编写的DSA代码转换为可交互、可执行的动态可视化的代码, 从而提升教学直观性和学习体验. 为系统评估该方法的有效性, 构建了包含150对DSA代码及其对应的DSV可视化代码的数据集, 并对比了3种方法 (直接提示、思维链提示、ReAct智能体) 在多种主流LLM上的表现. 实验结果显示, 所提出的基于ReAct智能体的方法在编译通过率 (Compilation Rate, CR)、执行通过率 (Execution Rate, ER)和可用率 (Usability Rate, UR) 这3项指标上均显著优于其他方法, 尤其在DeepSeek-R1模型下表现最优, 显著提升了生成可视化代码的准确性与可交互性, 验证了结合LLM与智能体框架在DSA可视化教学中的可行性与优势, 为构建高效、个性化、自动化的计算机编程教学工具提供了新路径.

关键词: 数据结构与算法可视化, 大语言模型, 智能体, 代码生成

Abstract:

Data structures and algorithms (DSA), as a core course in computer science education, play a key role in cultivating programming skills and algorithmic thinking of students. Visualization can significantly enhance teaching effectiveness and deepen student understanding in DSA education. However, existing DSA visualization tools often rely on manually written visualization codes that lead to limitations such as limited coverage, high maintenance costs, and lack of interactivity; hence, the needs of dynamic demonstrations and personalized teaching are difficult to meet. With the outstanding performance of large language models (LLMs) in code generation, automated DSA visualization has become a promising possibility. Therefore, this study proposed an interactive visualization code generation method based on the reasoning and acting (ReAct) AI agent framework, aiming to address the low automation and insufficient interactivity of traditional visualization tools. By leveraging the code generation capabilities of LLMs and integrating with the data structure visualization (DSV) platform interface, the proposed method transformed Python-based DSA code into interactive, executable, and dynamically visualized code, thereby enhancing teaching clarity and learning experience. To systematically evaluate the effectiveness of the method, we constructed a dataset of 150 pairs of DSA code and corresponding DSV visualization code and compared three approaches—direct prompting, chain-of-thought prompting, and the ReAct AI agent approach—across several mainstream LLMs. The experimental results showed that the proposed ReAct AI agent-based method significantly outperformed the other approaches in terms of the compilation rate, execution rate, and usability rate, with the best performance observed in the DeepSeek-R1 model. This demonstrated notable improvements in the accuracy and interactivity of generated visualization code. This research confirms the feasibility and advantages of integrating LLMs with agent frameworks in DSA visualization teaching, offering a novel path toward building efficient, personalized, and automated tools for computer programming education.

Key words: data structure and algorithm visualization, large language model, AI agent, code generation

中图分类号:

TP391

庞瑞洋, 陆雪松. 基于智能体的可交互数据结构和算法可视化实现[J]. 华东师范大学学报（自然科学版）, 2025, 2025(5): 32-42.

Ruiyang PANG, Xuesong LU. Interactive data structure and algorithm visualization based on AI agents[J]. J* E* C* N* U* N* S*, 2025, 2025(5): 32-42.

图/表 9

图1

图2

图3

表1

表2

图4

图5

图6

图7

参考文献 26

1	GE Y Q, HUA W Y, MEI K, et al.. OpenAGI: When LLM meets domain experts. Advances in Neural Information Processing Systems, 2023, 36, 5539- 5568.
2	UMER F, BATOOL I, NAVED N.. Innovation and application of large language models (LLMs) in dentistry–a scoping review. BDJ Open, 2024, 10, 90.
3	WEBER I. Large language models as software components: A taxonomy for LLM-integrated applications [EB/OL]. (2024-06-13)[2025-06-05]. https://arxiv.org/abs/2406.10300.
4	MTAHO A B, MSELLE L J.. Difficulties in learning the data structures course: Literature review. The Journal of Informatics, 2024, 4 (1): 26- 55.
5	SU S, ZHANG E, DENNY P, et al. A game-based approach for teaching algorithms and data structures using visualizations [C]// Proceedings of the 52nd ACM Technical Symposium on Computer Science Education. ACM, 2021: 1128-1134.
6	ZINGARO D, TAYLOR C, PORTER L, et al. Identifying student difficulties with basic data structures [C]// Proceedings of the 2018 ACM Conference on International Computing Education Research. ACM, 2018: 169-177.
7	CHEN E, HUANG R, CHEN H S, et al. GPTutor: A ChatGPT-powered programming tool for code explanation [C]// Artificial Intelligence in Education. Posters and Late Breaking Results, Workshops and Tutorials, Industry and Innovation Tracks, Practitioners, Doctoral Consortium and Blue Sky. AIED 2023. Communications in Computer and Information Science, vol 1831. Cham: Springer, 2023: 321-327.
8	JAMIE P, HAJIHASHEMI R, ALIPOUR S. Utilizing ChatGPT in a data structures and algorithms course: A teaching assistant’s perspective [EB/OL]. (2025-03-02)[2025-06-05]. https://arxiv.org/abs/2410.08899.
9	YAO S Y, ZHAO J, YU D, et al. ReAct: Synergizing reasoning and acting in language models [EB/OL]. (2023-03-10)[2025-06-05]. https://arxiv.org/abs/2210.03629.
10	MU F W, SHI L, WANG S, et al.. ClarifyGPT: A framework for enhancing LLM-based code generation via requirements clarification. Proceedings of the ACM on Software Engineering, 2024, 1 (FSE): 2332- 2354.
11	GU Q H. LLM-based code generation method for golang compiler testing [C]// Proceedings of the 31st ACM Joint European Software Engineering Conference and Symposium on the Foundations of Software Engineering. ACM, 2023: 2201-2203.
12	DU X Y, LIU M W, WANG K X, et al. Evaluating large language models in class-level code generation [C]// Proceedings of the IEEE/ACM 46th International Conference on Software Engineering. IEEE, 2024: 982-994.
13	ZHAO X L, LUO X Z, SHI Q, et al. ChartCoder: Advancing multimodal large language model for Chart-to-Code generation [C]// Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). ACL, 2025: 7333–7348.
14	XIAO J Y, WAN Y X, HUO Y T, et al. Interaction2Code: How far are we from automatic interactive webpage generation? [EB/OL]. (2024-11-05)[2025-06-05]. https://arxiv.org/html/2411.03292v1.
15	JI H N, QIU S, XIN S Y, et al. From EduVisBench to EduVisAgent: A benchmark and multi-Agent framework for pedagogical visualization [EB/OL]. (2025-03-27)[2025-06-05]. https://arxiv.org/abs/2505.16832.
16	LIU Z J, ZHANG Y Z, LI P, et al. Dynamic LLM-agent network: An LLM-agent collaboration framework with agent team optimization [EB/OL]. (2024-02-11)[2025-06-05]. https://openreview.net/forum?id=i43XCU54Br.
17	ZHAO A, HUANG D, XU Q, et al. Expel: LLM agents are experiential learners [C]// Proceedings of the AAAI Conference on Artificial Intelligence. 2024, 38(17): 19632-19642.
18	SHAO Y F, LI L Y, DAI J Q, et al. Character-LLM: A trainable agent for role-playing [EB/OL]. (2023-11-14)[2025-06-05]. https://arxiv.org/abs/2310.10158.
19	LI Y, ZHANG Y X, SUN L C. MetaAgents: Simulating interactions of human behaviors for LLM-based task-oriented coordination via collaborative generative agents [J]. (2023-10-10)[2025-06-05]. https://arxiv.org/abs/2310.06500.
20	TANG H, KEY D, ELLIS K. WorldCoder, a model-based LLM agent: Building world models by writing code and interacting with the environment [C]// Proceedings of the 38th International Conference on Neural Information Processing System . Red Hook, NY, USA: Curran Associates Inc., 2024: 70148-70212.
21	WEI J, WANG X Z, SCHUURMANS D, et al. Chain-of-thought prompting elicits reasoning in large language models [C]// Proceedings of the 36th International Conference on Neural Information Processing System. Red Hook, NY, USA: Curran Associates Inc., 2022: 24824-24837.
22	HUI B Y, YANG J, CUI Z Y, et al. Qwen2.5-Coder technical report [EB/OL]. (2024-11-12)[2025-06-05]. https://arxiv.org/abs/2409.12186.
23	GUO D Y, ZHU Q H, YANG D J, et al. DeepSeek-Coder: When the large language model meets programming-The rise of code intelligence [EB/OL]. (2024-01-26)[2025-06-05]. https://arxiv.org/abs/2401.14196.
24	GUO D Y, YANG D J, ZHANG H W, et al. DeepSeek-R1: Incentivizing reasoning capability in LLMs via reinforcement learning [EB/OL]. (2025-01-22)[2025-06-05]. https://arxiv.org/abs/2501.12948.
25	BALOG M, GAUNT A L, BROCKSCHMIDT M, et al. DeepCoder: Learning to write programs [EB/OL]. (2017-03-08)[2025-06-05]. https://arxiv.org/abs/1611.01989.
26	CHEN M, TWOREK J, JUN H, et al. Evaluating large language models trained on code [EB/OL]. (2021-07-14)[2025-06-05]. https://arxiv.org/abs/2107.03374.

模型	方法	CR@1/%	CR@3/%	ER@1/%	ER@3/%	UR@1/%	UR@3/%	NLD/%
Qwen2.5-Coder-7B-Instruct	Direct	57.45	78.67	37.33	62.07	9.56	31.98	76.39
	CoT	60.98	85.41	45.56	65.60	11.73	34.07	75.31
	ReAct	71.81	91.93	47.24	74.21	12.08	35.62	75.12
Qwen2.5-Coder-14B-Instruct	Direct	68.43	83.48	41.17	64.18	14.31	33.29	75.34
	CoT	69.56	89.97	47.78	71.30	15.19	34.67	74.50
	ReAct	83.42	93.54	49.63	79.81	17.87	35.93	73.81
Qwen2.5-Coder-32B-Instruct	Direct	78.68	95.43	46.47	78.67	25.24	51.31	73.59
	CoT	81.29	96.01	49.12	80.14	26.67	52.49	72.88
	ReAct	87.53	97.17	57.83	85.48	30.29	53.63	72.62
DeepSeek-Coder-6.7B-Instruct	Direct	43.56	70.88	21.59	32.63	4.05	10.52	91.44
	CoT	48.11	78.69	24.30	39.41	5.33	13.26	89.85
	ReAct	49.47	81.32	28.13	41.09	5.98	15.72	88.15
DeepSeek-Coder-33B-Instruct	Direct	46.74	82.67	23.77	40.18	4.67	12.07	88.04
	CoT	49.27	84.33	28.12	46.00	5.56	16.71	86.92
	ReAct	50.56	87.26	30.57	48.42	6.12	17.32	84.81
DeepCoder-14B-Preview	Direct	48.97	85.34	26.33	46.32	8.11	14.38	88.60
	CoT	51.56	88.27	28.67	49.65	10.65	17.46	86.21
	ReAct	53.32	89.02	29.81	51.18	11.17	19.21	85.02
DeepSeek-R1	Direct	83.43	96.77	63.11	86.53	45.98	72.61	71.93
	CoT	87.89	97.12	65.37	91.34	48.31	78.96	71.86
	ReAct	91.52	98.28	68.53	93.14	50.46	82.75	69.84

模型	方法	简单			中等			困难
模型	方法	CR@3/%	ER@3/%	UR@3/%	CR@3/%	ER@3/%	UR@3/%	CR@3/%	ER@3/%	UR@3/%
Qwen2.5-Coder-32B -Instruct	Direct	96.57	90.17	77.78	95.18	90.17	47.73	93.48	40.47	1.96
	CoT	97.26	93.65	79.37	95.84	93.28	53.28	94.01	45.81	4.27
	ReAct	98.41	94.92	87.48	96.23	94.09	62.72	94.97	50.40	5.65
DeepSeek-Coder-33B -Instruct	Direct	87.31	80.14	43.81	73.75	37.27	4.55	60.11	3.54	0
	CoT	89.82	86.25	48.57	75.27	43.19	6.81	64.18	3.87	0
	ReAct	91.53	88.73	50.24	79.52	47.67	7.72	66.74	4.02	2.33
DeepCoder-14B -Preview	Direct	85.43	79.60	42.27	76.82	51.82	5.92	65.63	14.17	0
	CoT	87.62	84.96	47.46	78.63	59.10	7.61	72.09	20.23	1.04
	ReAct	92.15	86.81	51.62	80.26	62.43	11.36	75.13	23.26	2.57
DeepSeek-R1	Direct	99.34	90.04	84.16	89.81	86.36	68.19	88.15	49.27	10.47
	CoT	100	93.18	89.44	92.14	87.52	70.24	91.45	52.09	13.22
	ReAct	100	95.47	91.65	94.71	90.41	75.13	93.02	59.57	16.51

[1]	何德鑫, 韩凡宇, 王伟. 大语言模型在开源项目主题标注中的应用与评估研究[J]. 华东师范大学学报（自然科学版）, 2025, 2025(5): 14-24.
[2]	陈郅睿, 陆雪松. 基于开源代码大语言模型提示的学生代码修复[J]. 华东师范大学学报（自然科学版）, 2024, 2024(5): 93-103.
[3]	寇思佳, 闫凤云, 马晶. 国内大语言模型在学科知识图谱自动标注上的应用——以道德与法治和数学学科为例[J]. 华东师范大学学报（自然科学版）, 2024, 2024(5): 81-92.
[4]	刘佳, 孙新, 张宇晴. 知识图谱与大语言模型协同的教育资源内容审查[J]. 华东师范大学学报（自然科学版）, 2024, 2024(5): 57-69.
[5]	葛进, 陆雪松. 基于UI图像的Web前端代码自动生成[J]. 华东师范大学学报（自然科学版）, 2023, 2023(5): 100-109.
[6]	朱仁龙;赵俊逸;黄勇. 嵌入式装置USB接口程序的设计[J]. 华东师范大学学报(自然科学版), 2006, 2006(3): 126-133.