Large language models (LLMs) are rapidly reshaping business and management, yet existing reviews remain fragmented, with limited integration of bibliometric structure and operational insights. This study addresses that gap by combining bibliometric analysis with a systematic thematic synthesis, linking publication trends, intellectual structure, and real-world deployment patterns across 84 application-focused studies selected from an initial pool of 210 records identified from Scopus, Web of Science, and Google Scholar (up to August 2025). The analysis shows a sharp post-2023 expansion of research alongside a shift from general-purpose conversational use toward domain-specific and architecture-aware implementations. In contrast to prior reviews, this study identifies how application patterns differ across domains: knowledge-intensive contexts increasingly adopt retrieval-augmented generation (RAG) for grounding, while finance applications more frequently rely on fine-tuned and domain-specific models. Keyword co-occurrence analysis highlights a technology-centric core, dominated by ChatGPT, foundation models, and RAG, linked to key business domains including operations, finance, marketing, and supply chain management. While organisations are increasingly adopting customised LLM pipelines and governance-aware workflows, challenges persist in accuracy, integration, ethical risk, and data quality. By consolidating dispersed evidence, this study offers a structured understanding of emerging application patterns and provides guidance for future research and responsible enterprise adoption of LLMs.

Ontology learning from unstructured text remains a complex challenge, particularly for large and intricate textual sources. This paper introduces CoA-Text2OWL, a multi-agent framework that leverages Large Language Models (LLMs) within a Chain-of-Agents to improve ontology generation. Unlike traditional single-LLM approaches, CoA-Text2OWL distributes the task across multiple worker agents, each processing a chunk of the input text, while a manager agent synthesizes their outputs into a coherent ontology. We evaluate our approach against a baseline single-LLM-based Text2OWL method, demonstrating improvements in object property extraction and ontology completeness. However, challenges remain in preserving hierarchical structures. Our results highlight the potential of multi-agent AI for ontology learning and suggest future enhancements, including specialized agent roles for term extraction, classification, and validation. We further validate CoA-Text2OWL by applying it to construct ontologies from real-world TRACES data related to urban systems in Geneva, achieving strong semantic alignment with source documents. This research contributes to the evolving field of LLM-powered multi-agent systems and their application in knowledge representation.

Neuro-symbolic artificial intelligence (NSAI) represents a transformative approach in artificial intelligence (AI) by combining deep learning’s ability to handle large-scale and unstructured data with the structured reasoning of symbolic methods. By leveraging their complementary strengths, NSAI enhances generalization, reasoning, and scalability while addressing key challenges such as transparency and data efficiency. This paper systematically studies diverse NSAI architectures, highlighting their unique approaches to integrating neural and symbolic components. This study then evaluates these architectures against comprehensive set of criteria, including generalization, reasoning capabilities, transferability, and interpretability, therefore providing a comparative analysis of their respective strengths and limitations. Notably, the Neuro $\rightarrow$ Symbolic $\leftarrow$ Neuro model consistenty outperforms its counterparts across all evaluation metrics. This result aligns with state-of-the-art research that highlight the efficacy of such architectures in harnessing advanced technologies like multi-agent systems. Moreover, our NSAI framework using retrieval-augmented illustrates how the 4D printing ontology can be systematically enriched with additional classes, object properties, data properties and individuals.

In the fast-paced realm of digital transformation, businesses are increasingly pressured to innovate and boost efficiency to remain competitive and foster growth. Large Language Models (LLMs) have emerged as game-changers across industries, revolutionizing various sectors by harnessing extensive text data to analyze and generate human-like text. Despite their impressive capabilities, LLMs often encounter challenges when dealing with domain-specific queries, potentially leading to inaccuracies in their outputs. In response, Retrieval-Augmented Generation (RAG) has emerged as a viable solution. By seamlessly integrating external data retrieval into text generation processes, RAG aims to enhance the accuracy and relevance of the generated content. However, existing literature reviews tend to focus primarily on the technological advancements of RAG, overlooking a comprehensive exploration of its applications. This paper seeks to address this gap by providing a thorough review of RAG applications, encompassing both task-specific and discipline-specific studies, while also outlining potential avenues for future research. By shedding light on current RAG research and outlining future directions, this review aims to catalyze further exploration and development in this dynamic field, thereby contributing to ongoing digital transformation efforts.

Business events signify crucial activities within a company, indicating growth opportunities and investment prospects. They encompass various developments such as recruitment drives, market expansions, mergers, and product launches. Understanding these events is vital for businesses seeking to stay updated with market dynamics, as they provide real-time insights into a company’s trajectory. Moreover, comprehending the business events of one company can offer strategic advantages to others, facilitating informed decision-making and fostering collaboration within the business ecosystem. Extracting information about these events involves diverse structured, semi-structured, and unstructured data sources, posing challenges for traditional extraction methods. Despite the promise shown by existing openly available LLMs driven by Generative Artificial Intelligence (GenAI), they face challenges when dealing with domain-specific queries. Retrieval-Augmented Generation (RAG) addresses this challenge by seamlessly integrating multiple external data sources of varying structures. In our study, we demonstrate how RAG with LLM facilitates precise extraction of business events, ensuring adaptability in dynamic business environments where datasets are constantly evolving.

Businesses heavily rely on data sourced from various channels like news articles, financial reports, and consumer reviews to drive their operations, enabling informed decision-making and identifying opportunities. However, traditional manual methods for data extraction are often time-consuming and resource-intensive, prompting the adoption of digital transformation initiatives to enhance efficiency. Yet, concerns persist regarding the sustainability of such initiatives and their alignment with the United Nations (UN)’s Sustainable Development Goals (SDGs). This research aims to explore the integration of Large Language Models (LLMs) with Retrieval-Augmented Generation (RAG) as a sustainable solution for Information Extraction (IE) and processing. The research methodology involves reviewing existing solutions for business decision-making, noting that many systems require training new machine learning models, which are resource-intensive and have significant environmental impacts. Instead, we propose a sustainable business solution using pre-existing LLMs that can work with diverse datasets. We link domain-specific datasets to tailor LLMs to company needs and employ a Multi-Agent architecture to divide tasks such as information retrieval, enrichment, and classification among specialized agents. This approach optimizes the extraction process and improves overall efficiency. Through the utilization of these technologies, businesses can optimize resource utilization, improve decision-making processes, and contribute to sustainable development goals, thereby fostering environmental responsibility within the corporate sector.

In the contemporary digital landscape, media content stands as the foundation for political news analysis, offering invaluable insights sourced from various channels like news articles, social media updates, speeches, and reports. Natural Language Processing (NLP) has revolutionized Political Information Extraction (IE), automating tasks such as Event Extraction (EE) from these diverse media outlets. While traditional NLP methods often necessitate specialized expertise to build rule-based systems or train machine learning models with domain-specific datasets, the emergence of Large Language Models (LLMs) driven by Generative Artificial Intelligence (GenAI) presents a promising alternative. These models offer accessibility, alleviating challenges associated with model construction from scratch and reducing the dependency on extensive datasets during the training phase, thus facilitating rapid implementation. However, challenges persist in handling domain-specific tasks, leading to the development of the Retrieval-Augmented Generation (RAG) framework. RAG enhances LLMs by integrating external data retrieval, enriching their contextual understanding, and expanding their knowledge base beyond pre-existing training data. To illustrate RAG’s efficacy, we introduce the Political EE system, specifically tailored to extract political event information from news articles. Understanding these political insights is essential for remaining informed about the latest political advancements, whether on a national or global scale.

The fast development of wireless location acquisition technologies has led to a significant increase in the availability of mobility data, specifically spatio-temporal trajectory data, which includes information about the movements (locations) of objects over time. This data has proven valuable for a wide range of applications, including predicting travel patterns, discovering routes, analyzing social interactions, and managing resources in urban environments. However, using trajectory datasets can also be challenging, particularly in different countries. This article aims to explore the challenges of using trajectory datasets, specifically using focus group discussions. Focus groups are a qualitative method of gaining a deeper understanding of a particular topic and have not been previously used to examine the challenges related to trajectory datasets. The information gathered through these discussions is augmented by a review of existing literature for a comprehensive understanding of data challenges.

Managing large volumes of business news data is important for companies. However, identifying relevant data can be challenging due to the dynamic nature of business environments and varying market needs. The article acknowledges this challenge and suggests that the lack of a standardized definition for relevant data makes it even more difficult for companies to prioritize and manage their data effectively, which can lead to missed opportunities. To address this challenge, the article proposes a text relevance hierarchy framework consisting of five levels that assess the relevance of business news texts to a company's specific operations and interests. The framework uses criteria such as specific topics, organizations, people, locations, and financial figures involved in the article to evaluate the importance of texts. Natural Language Processing (NLP) approaches such as entity recognition, topic modeling, and similarity analysis can be leveraged to implement the text relevancy hierarchy. By using this framework, companies can prioritize and manage their business news information efficiently, focus on the most important and relevant texts, and identify new areas of interest based on changing market needs.

Natural Language Processing (NLP) is a rapidly growing field with immense potential to transform Management Information Systems (MIS) and drive digital transformation in businesses. To explore the possible applications of NLP for MIS, we first investigated current state-of-the-art research by examining the top-tier journals in the field. Our analysis revealed that while there has been some research exploring the benefits of NLP for MIS, the full potential of this powerful technology has not yet been fully realized in the literature. Therefore, this study aims to fill this gap by discussing the potential applications of NLP for MIS. We cover several scenarios, including boosting marketing campaigns, enhancing supplier relationship management, detecting misinformation, using competitive analysis to gain insights, identifying related contextual terms, and enhancing BI systems. The main goal of presenting various scenarios is to demonstrate the vast potential of NLP in transforming MIS and supporting digital transformation. Through this exploration of various scenarios, we hope to inspire researchers and practitioners to further explore the potential of NLP for MIS applications and unlock new possibilities for enhancing business operations, decision-making, and strategy development in the context of digital transformation.

Taxonomies are widely used by various business organizations for document classification and organization. Business models built using taxonomies have the potential to reduce their efficiency with the arrival of new business ideas and concepts in the market over time. This happens because outdated business taxonomies get insufficient to fully capture the intended meaning of business documents to perform classification. Experts may need time to understand and engineering to place new data into the taxonomy to increase the classification accuracy of business documents. Here, the idea of automatic semantic enrichment of taxonomies came into consideration. This study introduces a taxonomy enrichment approach based on Natural Language Processing (NLP) technique, i.e. BERTopic, a Neural topic modeling using contextualized Bidirectional Encoder Representations from Transformers (BERT) to automatically augment a given business taxonomy with many additional concepts by leveraging a corpus of online news documents. The experiments show that augmenting topics from the text corpus into the taxonomy substantially increases the classification accuracy of business documents.

Taxonomies are crucial for executing Business Intelligence (BI) applications by preventing users from being overwhelmed with information. The business applications require knowledge of the key concepts and their organization to perform the classification of news articles. To ensure and maintain reliable information classification quality, it is crucial to keep the same definitions and organization of these concepts. This indicates the major significance of BI taxonomies in organizations. However, their development in business information systems follows an ad hoc process in most cases. Compared to many other domains, e.g. environmental and life sciences research, no mature and updated BI taxonomies are available in the literature. Existing studies cover BI taxonomies, but these are excessively generic and domain-specific. As a result, the BI domain suffers from many immature, incorrect, and incomplete notions of concepts. New BI-related concepts emerge rapidly, making it essential to include them in existing taxonomies during the enrichment process. The contribution of our research is the exploration of the possibilities of taxonomy enrichment using existing datasets. The expansion of the existing business taxonomy using multifaceted data sources to capture new concepts comprising 1) lexical datasets, 2) pre-trained word embeddings, 3) linked open data vocabulary, and 4) corpus-based relevant thematic extraction of features from news articles using Natural Language Processing (NLP) techniques. The highest semantic enrichment rate of a taxonomy got on a combination of these 4 methods. Eventually, enriched business taxonomy will contribute to the improved classification of news articles.

Business Intelligence (BI) requires the collection and organization of important pieces of information (i.e. entities) from multiple sources to provide valuable insights (e.g. business trends) as events (i.e. a specific happening linked with a specific location and time) to users. Online news articles are one of the important information sources that present business news offered by various companies in the market every day all around the world. These news articles often cover the same events and report redundant information. Existing news platforms aim at collecting the key entities from news articles and providing a mechanism to view the latest and relevant business events based on user interest. However, they do not provide a method to model business events and understand them temporally, spatially, and contextually (i.e. changes in the event). For instance, it is crucial to know for how long a business event has been active? How important is its evolution locally, or worldwide? Or how did different companies come up with this event as competitors in the market? The contribution of this research is the exploration of the possibilities of modeling spatial, temporal, and contextual information evolution related to business events through the application of knowledge graphs and text analytics, more specifically, Natural Language Processing (NLP) methods. The constructed knowledge graphs through Named-Entity Recognition (NER), i.e., an NLP technique, present a compact news representation that tells the key entities of the business event at one glance using linked open data concepts. It enables the assessment of other related news events as well as provides the means for analysis of the influence and evolution of business events.

The paper proposes a new type of probabilistic description logics (p-DLs) with a different interpretation of uncertain knowledge. In both approaches (classical state of the art approaches and the approach of this paper), probability values are assigned to axioms in a knowledge base. While In classical p-DLs, the probability value of an axiom is interpreted as the probability of the axiom to be true in contrast to be false or unknown, the probability value in this approach is interpreted as the probability of an the axiom to be true in contrast to other axioms being true. The paper presents the theory of that novel approach and a method for the treatment of such data. The proposed description logic is evaluated with some sample knowledge bases and the results are discussed.

This paper presents a data model to capture moving and changing objects in the context of dynamic built environment. Building elements are subject to change which represents semantic trajectories crossing trajectories of users. These semantic trajectories in dynamics built environment permit to capture fine-grained activities and behaviors of users and objects. The data model is based on ontology and description logics to capture logic constraints on semantic trajectories.

International audience

Analyzing Big Data can help corporations to improve their efficiency. In this work we present a new vision to derive Value from Big Data using a Semantic Hierarchical Multi-label Classification called Semantic HMC based in a non-supervised Ontology learning process. We also propose a Semantic HMC process, using scalable Machine-Learning techniques and Rule-based reasoning.

International audience

The identification of objects in 3D point cloud data has always presented a real challenge. Such a process highly depends on human interpretation of the scene and its objects. Actual approaches are numerical based; in best cases, static models are used as a template for the detection process. By the presented work, we aim at extending the detection process by bringing the human expert knowledge about the scene, the objects, their characteristics and their relations onto the processing chain. To do, we present in this paper a knowledgedriven method for the detection of object and its qualification using OWL ontology. The knowledge contained by the ontology defines the constraints about the objects. Logic programs are used as rules to define constrains between objects. The processing of the scene is an iterative annotation process that combines 3D algorithms, geometric analysis, spatial analysis and especially specialist’s knowledge. The created platform takes a set of 3D point clouds as input and produces as output a populated ontology corresponding to an indexed scene. The context of the study is the detection of railway objects materialized within the Germany Railway scene. Thus, the resulting enriched and populated ontology contains the annotations of objects in the point clouds, and can be used further on to feed a GIS system or an IFC file for architecture purposes.

This work aims at bound geometrical detection of 3D objects from a point cloud using semantic descriptors to improve reusability of architectural building reconstruction and aid automatic reasoning in building information modeling (BIM). Based on exploring cognitive origins of spatial semantics representations, semantics conceptualization and classification is proposed for management of architectural objects. The knowledge classification is formalized with transformations among closed world assumption (CWA) and open world assumption (OWA). Initial case study of a building prototype complying with the IFC standard reveals the organization of empirical knowledge rules and semantics scopes both in a bottom up manner of geometry→topology→semantics, and vice versa.

This paper presents an ongoing research which aims at combining geometrical analysis of point clouds and semantic rules to detect 3D building objects. Firstly by applying a previous semantic formalization investigation, we propose a classification of related knowledge as definition, partial knowledge and ambiguous knowledge to facilitate the understanding and design. Secondly an empirical implementation is conducted on a simplified building prototype complying with the IFC standard. The generation of empirical knowledge rules is revealed and semantic scopes are addressed both in the bottom up manner along the line of geometry → topology → semantic, and a vice versa top down manner. Concrete implementation is on the platform of protégé with Semantic Web Rule Language (SWRL).

The "Industrial Foundation Classes" (IFC) are an ISO norm to define all components of a building in a civil engineering project. IFC files are textual files whose size can reach 100 megabytes. Several IFC files can coexist on the same civil engineering project. Due to their size, their handling and sharing is a complex task. In this paper, we present an approach to automatically identify business objects in the IFC files and simplify their visualization and manipulation on the Internet. We construct an IFC Viewer which transforms the IFC file into a XML IFC tree manipulated through the 3D visualization of the building. The IFC Viewer composed a web-based platform called ACTIVe3D BUILD SERVER. This platform lets geographically dispersed project participants-from architects to electricians-directly use and exchange project documents in a centralized virtual environment during the life cycle of a civil engineering project.