A Method for Text Information Extraction Leveraging Large Language Models and Two-Stage Information Filtering

doi:10.12142/ZTECOM.202603011

ZTE Communications ›› 2026, Vol. 24 ›› Issue (3): 92-98.DOI: 10.12142/ZTECOM.202603011

• Industry-Academia Co-Research • Previous Articles

A Method for Text Information Extraction Leveraging Large Language Models and Two-Stage Information Filtering

Lu Ping¹, Qin Bin²(), Wang Yiyang², Xu Yan³, Deng Fangwei¹

^1.ZTE Corporation, Shenzhen 518057, China
^2.Shenzhen University, Shenzhen 518060, China
^3.Beijing Xingyun Digital Technology Co. , Ltd. , Beijing 100176, China

Received:2024-09-20 Online:2026-09-25 Published:2026-09-20
About author:Lu Ping is the vice president of ZTE Corporation. He also serves as the deputy director of the State Key Laboratory of Mobile Network and Mobile Multimedia Communication. His research interests include artificial intelligence, XR, immersive communication, and network-media convergence.
Qin Bin (qinbin@szu.edu.cn) is a professor-level senior engineer of Shenzhen University, China. His research interests focus on artificial intelligence.
Wang Yiyang received her master’s degree in new-generation information technology from Shenzhen University, China in 2026. Her research interests include the applications of large language models and 3D vision.
Xu Yan holds an MS degree in mechanical engineering. He is currently a product manager at Beijing Xingyun Numerical Technology Co., Ltd. His research interest focuses on the digital application of AI technology in the industry.
Deng Fangwei is a chief research engineer of ZTE Corporation. Her research interests include artificial intelligence, XR, immersive communication, and network-media convergence.
Supported by:
ZTE Industry?University?Institute Cooperation Funds(IA20231030016)

Abstract

Abstract:

This paper addresses shortcomings in the current combined application of large language models (LLMs) and vector knowledge bases by proposing a novel method for text information extraction. The method integrates the powerful language understanding capabilities of LLMs with the robust matching ability of two-stage information filtering, aiming to improve both the hit rate for retrieving relevant background knowledge and the accuracy of text information extraction. Experimental results on a real-world accident investigation report dataset demonstrate that the proposed method achieves a significant 37% improvement in extraction accuracy compared to baseline methods. Additionally, comparative results across data from different industries further validate the method’s strong generalization performance, suggesting its potential as an effective technical solution for other long-text information extraction tasks.

Key words: large language models, two-stage information filtering, text information extraction

Lu Ping, Qin Bin, Wang Yiyang, Xu Yan, Deng Fangwei. A Method for Text Information Extraction Leveraging Large Language Models and Two-Stage Information Filtering[J]. ZTE Communications, 2026, 24(3): 92-98.

Figures/Tables 9

Figure 1 Framework of text information extraction method based on LLMs and two-stage information filtering

Table 1 An example of the separation of retrieval and inquiry terms

Information Label	Before Decoupling	After Decoupling
Information Label	Before Decoupling	Retrieval Term		Inquiry Term
Accident location	In which province, city, county (district), and street is the accident site located, and what is the exact address?	1) In which province, city, county (district), and street is the accident site located, and what is the exact address?	2) Accident location	What is the specific location of the accident?
		3) Place of the accident	4) Position of the accident
		5) Accident site	6) Area of the accident
		7) Detailed address of the accident	8) Specific location of the accident
		9) Address of the accident	…

Figure 2 Prompt template for LLM

Table 2 Industry distribution of the dataset

Major Industry Category	Subdivision	Number of Reports
Construction	Residential building	48
Construction	Civil engineering	31
Commercial manufacturing	Trade and industry	100
Commercial manufacturing	Chemical industry	40

Table 3 Information labels for accident investigation reports

Index	Information Label
1	Accident location
2	Casualties
3	Economic loss
4	Legal basis

Figure 3 Flowchart of information extraction process from accident investigation reports

Table 4 Parameter configuration in the method

Parameter	Value
Similarity thresholdS	0.5
First-stage filtering text block threshold $N$	20
Semantic similarity threshold $S'$	0.5
Second-stage filtering text block threshold $K$	8

Table 4 Parameter configuration in the method

Parameter	Value
Similarity thresholdS	0.5
First-stage filtering text block threshold $N$	20
Semantic similarity threshold $S'$	0.5
Second-stage filtering text block threshold $K$	8

Table 5 Comparative experiment results

Legal

Basis

0.76

0.75

0.67

0.75

0.73

Table 6 Experimental results across different industries

Method	Residential Building	Civil Engineering	Trade and Industry	Chemical Industry	Average
Langchain-Chatchat	0.41	0.27	0.40	0.50	0.40
Ours	0.73	0.75	0.79	0.62	0.72

References 20

[1]	Mikolov T, Chen K, Corrado G, et al. Efficient estimation of word representations in vector space [PP/OL]. arXiv [2024-09-20].
[2]	Sutskever I, Vinyals O, Le Q V. Sequence to sequence learning with neural networks [C]//The 28th International Conference on Neural Information Processing Systems. ACM, 2014: 3104–3112. DOI:10.5555/2969033.2969173
[3]	Cho K, van Merrienboer B, Gulcehre C, et al. Learning phrase representations using RNN encoder-decoder for statistical machine translation [C]//The 2014 Conference on Empirical Methods in Natural Language Processing. ACL, 2014: 1724–1734. DOI: 10.3115/v1/d14-1179
[4]	Vaswani A, Shazeer N, Parmar N, et al. Attention is all you need [C]//The 31st International Conference on Neural Information Processing System. ACM, 2017: 6000–6010. DOI: 10.5555/3295222.3295349
[5]	Devlin J, Chang M W, Lee K, et al. BERT: pre-training of deep bidirectional transformers for language understanding [C]//Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. Association for Computational Linguistics, 2019: 4171–4186. DOI: 10.18653/v1/N19-1423
[6]	Liu Y H, Ott M, Goyal N, et al. RoBERTa: a robustly optimized BERT pretraining approach [PP/OL]. arXiv (2019-07-26) [2024-09-20].
[7]	Lan Z Z, Chen M D, Goodman S, et al. ALBERT: a lite BERT for self-supervised learning of language representations [PP/OL]. V6. arXiv (2020-02-09) [2024-09-20].
[8]	Li D Y, Tu Y F, Zhou X S, et al. End-to-end Chinese entity recognition based on BERT-BiLSTM-ATT-CRF [J]. ZTE communications, 2022, 20(S1): 27–35. DOI: 10.12142/ZTECOM.2022S1005
[9]	Zhou C, Li Q, Li C, et al. A comprehensive survey on pretrained foundation models: a history from BERT to ChatGPT [PP/OL]. V4. arXiv (2023-05-01) [2024-09-20].
[10]	Brown T B, Mann B, Ryder N, et al. Language models are few-shot learners [PP/OL]. V4. arXiv (2020-07-22) [2024-09-20].
[11]	Yue S B, Chen W, Wang S Y, et al. DISC-LawLLM: fine-tuning large language models for intelligent legal services [PP/OL]. V2. arXiv (2023-09-23) [2024-09-20].
[12]	Chiu J P C, Nichols E. Named entity recognition with bidirectional LSTM-CNNs [J]. Transactions of the association for computational linguistics, 2016, 4: 357–370. DOI: 10.1162/tacl_a_00104
[13]	Qian H J, Liu Z, Zhang P T, et al. MemoRAG: boosting long context processing with global memory-enhanced retrieval augmentation. [PP/OL]. arXiv (2024-09-09) [2024-09-20].
[14]	Guo Z R, Xia L H, Yu Y H, et al. LightRAG: simple and fast retrieval-augmented generation [C]//Proceedings of Findings of the Association for Computational Linguistics. ACL, 2025: 10746–10761. DOI: 10.18653/v1/2025.findings-emnlp.568
[15]	Singh I S, Aggarwal R, Allahverdiyev I, et al. ChunkRAG: novel LLM-chunk filtering method for RAG systems [EB/OL]. V1. arXiv (2024-10-25) [2024-10-30].
[16]	OpenAI. Best practices for prompt engineering with the OpenAI API [EB/OL]. [2024-09-20].
[17]	Du Z X, Qian Y J, Liu X, et al. GLM: general language model pretraining with autoregressive blank infilling [PP/OL]. V2. arXiv (2022-03-17) [2024-09-20].
[18]	Wang Y X, Sun Q X, He S C. M3 e‑base [EB/OL]. (2023‑06‑07)[2024‑09‑20].
[19]	Gao L, Callan J. Condenser: a pre-training architecture for dense retrieval [PP/OL]. V2. arXiv (2021-09-20) [2024-09-20].
[20]	Chatchat-Space. Langchain-Chatchat. [EB/OL]. [2024-09-20].

A Method for Text Information Extraction Leveraging Large Language Models and Two-Stage Information Filtering

RichHTML

PDF

Knowledge

Abstract

Cite this article

share this article

Figures/Tables 9

References 20

Related Articles 2

Recommended Articles

Metrics

[1]	Xie Linhao, Li Fan, Wu Mingxuan, Song Yong, Ouyang Ye. Mitigating Semantic Drift in Multi-Agent Communication: A Dynamic Neuro-Symbolic Approach [J]. ZTE Communications, 2026, 24(2): 64-70.
[2]	DU Linkang, SU Zhou, YU Xinyi. Dataset Copyright Auditing for Large Models: Fundamentals, Open Problems, and Future Directions [J]. ZTE Communications, 2025, 23(3): 38-47.