Data was collected from the Scopus database. The primary reason for selecting the Scopus database was that it stores a vast amount of scientific research (Falagas et al., 2008; Madzík et al., 2025). The search terms were “blockchain”, “block-chain”, “blockchain”, and “Finance” within the title, abstract, and keywords. The analysis focuses on articles and conference papers, indexed in Scopus between 2016–2025. This resulted in 2,439 articles. We applied two filters, namely “document type” and selected papers published in conference papers and articles; secondly, the English Language was used as another criterion. This resulted in 2,401 articles, which were further analyzed. The file was downloaded from Scopus on 20 September 2025, containing author details, title, publication year, source title, abstract, author keywords, index keywords, etc.

(Şakar and Tan, 2025) used 5,127 articles to perform an STM in the area of gifted education, while (Madzík et al., 2025) used 4,049 articles to study blockchain technologies in the sustainable supply chain. Another study performed by Shukla et al. (2024) employed a dataset of 2,360 articles to perform STM in the area of blockchain technology. Considering the extant literature, the number of articles selected for the present study is justified, especially for the emerging field of blockchain technology in finance.

The textual data (abstracts, keywords, and introductions) have to be pre-processed before the analysis (Hvitfeldt and Silge, 2021). The data were cleaned using natural language processing techniques. It involves tokenization, normalization, stemming, and removal of stop words, punctuation, and domain-irrelevant terms. All the capital letters were converted into small letters as part of normalization, while words were reduced to their root form for stemming. Words such as “the”, “is”, and “at” are classified as “Stop words” in natural language processing, and these words were removed as they do not contribute to the topic identification approach. Additionally, we removed all the punctuation and numbers to make the data clean and usable for analysis. To validate the model, the quantitative validation was evaluated at various levels, denoted by K, from 10 to 15 topics using coherence and exclusivity following Röder et al. (2015).

Topic Modelling: We applied STM to explore evolving themes in existing literature using Python. We included journal name, author affiliation, publication year, etc., to identify the differences in the topic distribution. The major advantage of STM is that it includes covariates such as author, publication year, location, etc., in the topic model, which allows the topic prevalence to be dependent on the covariates rather than observational values (Fresneda et al., 2021).

We used the STM package in Python to label topics with the help of the following metrics:

Highest Probability: It shows the probability of themes existing in the article or group of articles, along with their association with the article or group of articles.

To analyse the data, we have used STM using Python (Fresneda et al., 2021). STM is an advanced method developed on Latent Dirichlet Allocation (LDA). It helps to explore the hidden theme from the large data and visualize the data to explore the themes in depth. STM provides fast, clear, replicable analyzes that need a few priori assumptions (Roberts et al., 2014). STM pinpoints words with the highest per-topic weight, which represents the most pertinent words for each topic. After segments are created by clustering topic prevalence (and covariates), these relevant words can be linked to the identified segments to describe them (Fresneda et al., 2021). On the other hand, unsupervised approaches, such as employing STM followed by cluster analysis, can detect patterns in textual data without any pre-established limitations (Roberts et al., 2014). Using Python genism and scikit-learn, firstly, we estimated the topic proportion. We tested model by estimating 10, 15, and 20 topics, and we followed the recommendations of (Perelman et al., 2010) to find the optimal number of topics for the STM. While analyzing, we have taken care to select topics optimally, as reducing the number of topics may ignore uncommon themes, and increasing the number of topics may reduce semantic consistency (Griffiths et al., 2007).

We found 15 topics, as shown in Figure 2, to fit the STM model, which assigns the highest probability to each topic (Weston et al., 2023). The most important topic is Topic 6, Blockchain, technology, and security, which takes almost 35% of the data, and denotes that most of the studies have focused on blockchain infrastructure and security. The second most important topics are 2 and 13, which cover 15%–20% of the data; in the mid-range are topics 10, 4, and 15, which show that DeFi is the focus after the technical part. The contributions by topics 11, 1, 5, 12, and 7 are in decreasing order, which are niche but emerging topics. The contribution of topics 9, 14, 8 and 3 is very minimal and is less than 5% signifying that these topics are less researched. Overall, technology-related topics, such as technology, infrastructure, and security, account for around 60%; finance-specific topics, including DeFi, risk, and governance lending, comprise around 30%; and the remaining areas, including law, energy, and supply chain, contribute around 10%.

Topics are labelled using FREX, Semantic Coherence (c_v), and Exclusivity. It helps to find the quality of the topics. FREX is Frequency + Exclusivity. Semantic coherence and exclusivity are the most important tools for understanding whether topics are meaningful and interpretable. The model with K (number of topics) = 15 topics, showing a balanced trade-off between semantic coherence (>0.30) and exclusivity (>0.50), was selected (Roberts et al., 2019). Higher semantic coherence denotes better results (Nunkoo et al., 2023), implying that top words co-occur in documents a greater number of times, and low coherence means top words have less co-occurrence (Mimno et al., 2011). Exclusivity refers to whether the top words are unique or overlap; high exclusivity indicates that the topic is distinct, while low exclusivity means it overlaps with many other topics. Weston et al. (2023) suggested that for that there is no single number threshold for STM. It depends on the number of topics. The increase in the number of topics reduces the coherence, and a smaller number of topics will increase coherence (Mimno et al., 2011). Instead, we have worked on various number of topics and selected the model that balances coherence and exclusivity to extract meaningful themes (Mimno and Lee, 2014; Röder et al., 2015).

A theme with high semantic coherence and exclusivity is the most reliable topic (Mimno et al., 2011). As shown in Table 2, topic 6 is blockchain, a technology that is very frequent in the data, but it is not exclusive. FREX is a smart security system that is distinct from the cluster. FREX explains blockchain technology in technical contexts, such as smart contracts, security, and systems. Semantic coherence is 0.4063, which is moderate, which means the co-occurrence is not very strong. It means this topic is broad. Exclusivity (0.821), which is medium to high, is the blockchain’s most common word, whereas the distinct ones are “security” and “smart.” Junior and Fileto (2025) analyzed short, texted documents and reported coherence (c_v) 0.25 to 0.75 with K = 15. Molavi et al. (2020) applied STM on Open Educational Resources (OER) and reported coherence c_v between 0.36 and 0.52 for K between 2 and 10. Stracqualursi and Agati (2022) have analyzed Tweet topics and sentiments relating to distance learning and found coherence (c_v) between 0.34 and 0.50 for topics between 2–9. Zoya et al. (2021) analyzed Urdu tweets text and testified c_v value between 0.46 and 0.52 for k between 9 and 25. Similarly, M’sik and Casablanca (2020) analyzed the Corpus of COVID-19 and reported a coherence score c_v between 0.40–0.50.

Cluster analysis is a method for grouping topics that often co-occur in the same documents, have similar word distributions, and exhibit correlations in topic proportions. The last method for labelling topics is to observe the correlations between topics (Weston et al., 2023). The correlation among topics is presented in Figure 3. The strongest correlation is between Topic 6 and Topic 13 (0.66), followed by Topic 1 and Topic 13 (0.65), and Topic 2 and Topic 6 (0.65). A high correlation is found between Topic 6 and Topic 11 (0.62), while Topic 15 shows a high correlation with Topic 2 (0.58) and Topic 13 (0.58).

The cluster analysis was executed and presented to find the themes from the dataset using a dendrogram. The dendrogram displays the ‘best’ division into any number of groups (Jolliffe et al., 1989). It is a tree diagram that represents the topics grouped according to their similarity; the height of the line represents the similarity between topics. We have extracted five themes from the dendrogram analysis of the data, as shown in Figure 4. Theme one (Blockchain, AI and Data Intelligence) from topic 12 is centred around artificial intelligence, machine learning, and blockchain technology with direct financial application. Theme two (Blockchain Technology and Supply Chain Financing), gathered from Topics 2 and 7, focuses on the core blockchain technology in supply chain financing and logistics. Theme 3 (Blockchain for Governance, Identity and Regulation), gathered from Topics 14, 9, 3, and 8, focuses on government usage of blockchain, like digital identity, voting, Hyperledger, and legal framework. Theme 4 (Blockchain Infrastructure, Risk and Smart Systems), gathered from topics 15, 6, 11, 1, and 13, focuses on blockchain as a data-driven ecosystem, linked with IoT, security, and decentralisation. Last theme 5 (Blockchain in Finance and Markets), gathered from topics 4, 5, and 10, focuses on blockchain-enabled finance like lending platforms, decentralised markets, and fintech systems.

Topic 12 (Data, Financial and AI) drives this theme that connects blockchain, AI and data intelligence. In the present times, there has been unprecedented growth in FinTech due to the development of metaverse along with Generative Artificial Intelligence (Hutson and Banerjee, 2025). These technologies are capable of preventing fraud by improving security of transactions and management of risk to help transform the different operations of the financial sector (Subburayan et al., 2024). The analytics provide deep insights into the origin of the data, predictive planning, uncovering breach of security and fraud and regulatory compliance, using blockchain data and records (Dillenberger et al., 2019). These perspective highlight the synchronization of blockchain with various technologies such as machine learning, artificial intelligences to support automated decision-making and predictive analytics (Li and Wang, 2025; Vella et al., 2025). To explore this perspective further and broaden the knowledge, we propose the following questions:

RQ1: How can blockchain leverage AI to create trust in the financial sector?

This theme evolved by integrating Topic 2 (Blockchain, Technology, Chain) and Topic 7 (Chain, Supply, Supply Chain Finance (SCF). It captures research on core blockchain infrastructure applied to supply chain financing and logistics. Topics in this cluster, which exhibit moderate correlations (e.g., Topic 15 and Topic 2), reflect an emphasis on traceability, transparency, and efficiency across financial and operational networks (Kumari and Devi, 2022; Zheng et al., 2018). Zhang et al. (2025) proposed a solution, which is based on a Convolutional Neural Network-Particle Swarm Optimization, capable of building a warning system for security risk. It is important to note that the role of SCF is crucial for small and medium-sized enterprises for providing financing (Fang et al., 2025). Using the insight from the extant literature, the following Research questions are proposed for future researchers:

RQ1: How can blockchain technology lead to the development of supply chain financing

This theme is much broader and includes Topic 14 (Electricity, Service, Attacks), Topic 9 (Voting, Blockchain, Listed), Topic 3 (Peer, Identity, Buyer) and Topic 8 (Hyperledger, Fabric, Law). The major dimensions are digital voting, Hyperledger frameworks and the legal system. This collection highlights the pivotal role of blockchain for identity management, regulatory compliance, and governance, bridging technological potential with institutional and policy needs (Feng et al., 2019; Yeoh, 2017). Leadership, social norms and community engagement, traits of human behaviour, are instrumental in the governance of an organization; however, using blockchain algorithms, the organization can improve transparency and decentralization (Vella et al., 2025). The capability of blockchain also extends to improve efficiency, security, and transparency, along with enabling users to trust governance structures. All these positive actions led to the reduction of cost and better decision-making (Polcumpally et al., 2024). Many organizations do not possess unified governance, which acts as a threat to blockchain technology adoption (Banerjee and Chandani, 2025). We propose the following research questions:

RQ1: Is blockchain technology capable of handling electricity threats and attacks?

Topic 15 (Financial, System, Risk), Topic 6 (Blockchain, Technology, Security), Topic 11 (Data, System, IoT), Topic 1 (Data, Decentralized, Reserved), and Topic 13 (Blockchain, Smart, Data) constitute theme 4. It represents the most technologically interconnected cluster, with strong correlations between Topic 6 and Topic 13 (0.66) and between Topic 1 and Topic 13 (0.65). Research here focuses on building resilient, secure, and scalable blockchain ecosystems linked to IoT, risk governance, and smart financial systems, emphasizing both technical robustness and practical reliability (Tseng and Shang, 2021; Zheng et al., 2018). The technological infrastructure and regulatory frameworks should be properly aligned for the successful implementation of blockchain technology in financial services aimed at improving efficiency and security (Izzatillayev et al., 2025). Another research performed by Singh and Verma (2025) reported that the use of IoT devices is widespread among household infrastructure, and blockchain technology can lead to greater adoption of IoT. We propose the following research questions to cover the topics within this theme and increase the body of knowledge:

RQ1: How can blockchain infrastructure support a secure system and ensure access and rights are given to the legitimate entities?

The last theme comprises Topic 4 (Defi, Decentralized, Rights), Topic 5 (Lending, Digital, Loan) and Topic 10 (Financial, Research, Study). It centers on the practical financial applications of blockchain, including decentralised finance (DeFi), lending platforms, and investment markets. This cluster demonstrates how blockchain is reshaping real-world financial operations by enhancing transparency, efficiency, and accessibility, reflecting a shift in research from foundational technology toward applied financial innovation (Chang et al., 2019; Lavanya and Kavitha, 2022). Collectively, these clusters reveal that blockchain research in finance is both deepening its technological foundations and expanding into applications across financial markets, governance, and cross-sector domains, while also highlighting interconnections between technology, regulation, and financial practice. Prices in the decentralised markets are usually efficient, while there could be a few frictions related to gas fees and Ether volatility. The prices in the decentralised exchanges are closely linked with the currency markets due to arbitrage and informed trading (Ranaldo et al., 2024). The announcement of blockchain by the firm led to positive market reactions by the small firms in the United States (Ali et al., 2023), highlighting the importance of blockchain in business processes. To extend the literature, we propose the following research questions for future researchers:

RQ1: How can DeFi help and support the liquidity and trading system in the financial markets?

This research has made two theoretical contributions. First, it broadens the perspective of blockchain as only a technical or finance-related phenomenon. We offer an integrative framework through five interrelated clusters, depicting blockchain in finance as a complex socio-technical system of co-evolving technology, market applications and governance (Beck et al., 2018). This framework could be used to expand theory-building to include interdependence and for scholars to create models referring to interconnections. Second, the importance of themes, such as “Blockchain for AI and Data Intelligence” and “Blockchain for Governance, Identity and Regulation”, reflects the need for new theoretical lenses that combine principles of computer science with theories from institutional economics, corporate governance and regulation (Davidson et al., 2018).

Our cluster analysis presents an avenue for financial institutions and FinTech companies to expand into blockchain. The investment into foundational infrastructure for the system’s security, data integrity and resilient systems will remain a must-have condition for any applications to be trusted, as theme 4 signifies foundational applications are reaching saturation. Likewise, there exists a stage of maturity presented by the applications within theme 5 for practitioners to take advantage of market opportunities with products that could range from productizing user-friendly DeFi interfaces (Schär, 2021), to building tokenisation platforms for real-world assets, to deploying automated lending protocols (Gudgeon et al., 2020). As the next revolutionary frontier for firms and a prospective competitive differential in the U.S. and Europe, the outputs of theme 1 are advancing firms to stake their claim on being early adopters in integrating AI with blockchain for prediction analytics, advanced fraud detection and algorithm-based automated asset management.

Our research indicates a significant gap between technological innovation and regulatory oversight, which will require a substantial change in public policy thought and action. Regulators need to transition their position from reactive to proactive and consider implementing regulatory sandboxes for testing new products (Zetzsche et al., 2018), as well as partnering to develop standards in areas such as smart contract auditing and digital identification. Additionally, regulators will need to adopt ‘Suptech’ (Supervisory Technology), utilising the grounds for transparency provided by blockchain and creating tools for the real-time, in-the-galley monitoring of systemic risk and nefarious information (Arner et al., 2016), thus taking regulators out of the existing self-reported audit framework. Lastly, because public blockchains are borderless, this new regulatory approach requires unprecedented coordination on a global level to develop harmonized frameworks that reduce the ability to create jurisdictional arbitrage and establish a consistent, stable playing field for the future of finance.

While this study provides an in-depth outline of the study of blockchain in the finance context, it is accompanied by some limitations that can also be made into future research opportunities. First, reliance on the Scopus database and focus on abstracts, keywords and introductions was limiting; while this is a favourable approach, analyzing other databases, or the full text of the repeated articles would provide an even wider range of literature and further thematic detail. Second, STM itself was limiting the 15 topics structure required a level of researcher judgement, which could provide an alternative aggregate understanding if a different number of topics were chosen, and the method is less suited to very fractured and fledgling sub-fields of research (Roberts et al., 2019), where there are limited publications. Finally, although the identification processes of topic labels and the clustering process were supplemented by the STM, and these processes are subjective, and so whilst our five-cluster framework is compelling, it is not the only theme framework that could organise the research landscape.

Building directly from our findings, we present multiple targeted proposals for future research. First, a move to applied solutions (rather than solely conceptual exploration) is required, highly relevant for the intersection of Blockchain, AI, and IoT in adopting agile risk management in Decentralized Finance (DeFi) and architecture for scalable trade finance. Second, the DeFi phenomenon requires interdisciplinary research into practical governance and regulatory solutions, including “Regulatory Tech” in addition to legal personhood for Decentralized Autonomous Organizations (DAOs) (Hassan Collado and De Filippi, 2021). Also, a comparative or empirical approach highlighting requests to continue to explore and understand scalability would have to explicitly track real-world financial ecosystem Layer-2 solutions and implementation (Gans and Halaburda, 2015), beyond discussions of theory. Third, we recommend using qualitative approaches, like case studies, alongside quantitative methods to clarify the organizational and operational realities of blockchain adoption, to the data generated by topic modelling in our review.