Artificial intelligence for literature reviews: opportunities and challenges

Artificial Intelligence Review, Год журнала: 2024, Номер 57(10)

Опубликована: Авг. 17, 2024

Abstract This paper presents a comprehensive review of the use Artificial Intelligence (AI) in Systematic Literature Reviews (SLRs). A SLR is rigorous and organised methodology that assesses integrates prior research on given topic. Numerous tools have been developed to assist partially automate process. The increasing role AI this field shows great potential providing more effective support for researchers, moving towards semi-automatic creation literature reviews. Our study focuses how techniques are applied semi-automation SLRs, specifically screening extraction phases. We examine 21 leading using framework combines 23 traditional features with 11 features. also analyse recent leverage large language models searching assisting academic writing. Finally, discusses current trends field, outlines key challenges, suggests directions future research. highlight three primary challenges: integrating advanced solutions, such as knowledge graphs, improving usability, developing standardised evaluation framework. propose best practices ensure robust evaluations terms performance, transparency. Overall, offers detailed overview AI-enhanced researchers practitioners, foundation development next-generation solutions field.

Язык: Английский

---

Collaborative large language models for automated data extraction in living systematic reviews

Journal of the American Medical Informatics Association, Год журнала: 2025, Номер unknown

Опубликована: Янв. 21, 2025

Abstract Objective Data extraction from the published literature is most laborious step in conducting living systematic reviews (LSRs). We aim to build a generalizable, automated data workflow leveraging large language models (LLMs) that mimics real-world 2-reviewer process. Materials and Methods A dataset of 10 trials (22 publications) LSR was used, focusing on 23 variables related trial, population, outcomes data. The split into prompt development (n = 5) held-out test sets 17). GPT-4-turbo Claude-3-Opus were used for extraction. Responses 2 LLMs considered concordant if they same given variable. discordant responses each LLM provided other cross-critique. Accuracy, ie, total number correct divided by responses, computed assess performance. Results In set, 110 (96%) concordant, achieving an accuracy 0.99 against gold standard. 342 (87%) concordant. 0.94. 0.41 0.50 Claude-3-Opus. Of 49 25 (51%) became after cross-critique, increasing 0.76. Discussion Concordant are likely be accurate. instances cross-critique can further increase accuracy. Conclusion Large models, when simulated collaborative, workflow, extract with reasonable performance, enabling truly “living” reviews.

Язык: Английский

---

Extracting Fruit Disease Knowledge from Research Papers Based on Large Language Models and Prompt Engineering

Applied Sciences, Год журнала: 2025, Номер 15(2), С. 628 - 628

Опубликована: Янв. 10, 2025

In China, fruit tree diseases are a significant threat to the development of industry, and knowledge about is most needed professional for farmers other practitioners in industry. Research papers primary sources that represent cutting-edge progress disease research. Traditional engineering methods acquisition require extensive cumbersome preparatory work, they demand high level background information technology skills from handlers. This paper, perspective industry dissemination, aims at users such as farmers, experts, communicators, gatherers. It proposes fast, cost-effective, low-technical-barrier method extracting research paper abstracts—K-Extract, based on large language models (LLMs) prompt engineering. Under zero-shot conditions, K-Extract utilizes conversational LLMs automate extraction knowledge. The has constructed comprehensive classification system and, through series optimized questions, effectively overcomes deficiencies LLM providing factual accuracy. tests multiple available Chinese market, results show can seamlessly integrate with any model, DeepSeek model Kimi performing particularly well. experimental indicate have accuracy rate handling judgment tasks simple Q&A tasks. simple, efficient, accurate, serve convenient tool agricultural field.

Язык: Английский

---

Named entity recognition for construction documents based on fine-tuning of large language models with low-quality datasets

Automation in Construction, Год журнала: 2025, Номер 174, С. 106151 - 106151

Опубликована: Март 31, 2025

Язык: Английский

---

A discrete convolutional network for entity relation extraction

Neural Networks, Год журнала: 2025, Номер 184, С. 107117 - 107117

Опубликована: Янв. 6, 2025

Язык: Английский

---

A feature curve-based method for balancing brand identity and emotional imagery in automobile frontal form design

Advanced Engineering Informatics, Год журнала: 2025, Номер 65, С. 103118 - 103118

Опубликована: Янв. 18, 2025

Язык: Английский

---

SampleLLM: Optimizing Tabular Data Synthesis in Recommendations

Опубликована: Фев. 7, 2025

Tabular data synthesis is crucial in machine learning, yet existing general methods-primarily based on statistical or deep learning models-are highly data-dependent and often fall short recommender systems. This limitation arises from their difficulty capturing complex distributions understanding complicated feature relations sparse limited data, along with inability to grasp semantic relations. Recently, Large Language Models (LLMs) have shown potential generating synthetic through few-shot understanding. However, they suffer inconsistent distribution lack of diversity due inherent disparity the target dataset. To address these challenges enhance tabular for recommendation tasks, we propose a novel two-stage framework named SampleLLM improve quality LLM-based recommendations by ensuring better alignment. In first stage, employs LLMs Chain-of-Thought prompts diverse exemplars generate that closely aligns dataset distribution, even when input samples are limited. The second stage uses an advanced attribution-based importance sampling method refine relationships within reducing any biases introduced LLM. Experimental results three datasets, two online deployment illustrate significantly surpasses methods tasks holds promise broader range scenarios.

Язык: Английский

---

Meta-PKE: Memory-Enhanced Task-Adaptive Personal Knowledge Extraction in Daily Life

Information Processing & Management, Год журнала: 2025, Номер 62(4), С. 104097 - 104097

Опубликована: Фев. 17, 2025

Язык: Английский

---

Single-stage overlapping entity and relation extraction based on relation-specific heterogeneous graph neural network

Engineering Applications of Artificial Intelligence, Год журнала: 2025, Номер 147, С. 110288 - 110288

Опубликована: Фев. 20, 2025

Язык: Английский

---

Reliability of large language models as a tool for knowledge extraction from biographical dictionaries: the case of the Polish Biographical Dictionary

Digital Scholarship in the Humanities, Год журнала: 2025, Номер unknown

Опубликована: Фев. 13, 2025

Abstract Large language models are tools with great potential for text processing. This study aims to assess the reliability of models’ results in extracting structured knowledge from unstructured textual sources, particularly biographies Polish Biographical Dictionary. The task model was extract information about individuals, such as date and place birth, death burial, family relationships, important people, related settlements institutions well occupied positions. test conducted on a sample 250 biographies. texts were written 1930s onwards described lives individuals various historical periods. show that large (LLM) is very effective identifying basic personal data, occupations, or offices held by characters. Weaker obtained when attempting find places associated protagonists. outcome suggests LLMs can efficiently assist digitizing structuring biographical data offer promising tool improving bases speeding up work compared manual extraction information.

Язык: Английский

---