1 Introduction

Online Social Networks (OSNs) are becoming a powerful platform for the dissemination of information, enabling the spread of this information in different forms ranging from social behaviours to news and opinions. A proper understanding of information dissemination over OSNs is important for many real-world applications, including recommendation systems, election campaigns, rumour mongering, and viral marketing. The core problem in the domain of information diffusion in OSNs is the identification of seed nodes, that is, a set of influential users that are the sources of information diffusion. The effective identification of the seed users results in impactful information diffusion in OSNs [1,2,3,4,5,6,7,8,9,10,11,12,13,14].

Existing studies in the area have focused on finding these seed nodes using centrality measures that take the topology of the given network into consideration [14,15,16]. These approaches were sometimes less effective because the seed nodes were skewed towards a concentrated locality, which hampers proper and effective information diffusion in OSNs. These centrality-based methods do not take interest groups and communities into consideration. Seed nodes should be handled in a scattered form so that information can have better reachability. Thus, interest groups and seed nodes form a closely related problem and should be handled collectively for better and more effective information dissemination over OSNs [17, 18].

Recent works in the area have highlighted the importance of community structure in forming better information spread, as information often remains confined to communities prior to crossing community boundaries [16, 19]. Existing studies have considered either communities or seed node detection separately. Seed detection remains a challenge when the detected seeds are dominated by a small spatial coverage [20].

Keeping these gaps in mind, in this paper we propose an interest group-based seed node detection framework for effective information diffusion in online social networks. The proposed approach starts with the detection of interest groups using the Leiden community detection algorithm [21]. This method of interest group identification is very effective and result-oriented [20].

Once the interest groups are identified, we identify the seed nodes within each interest group using Netdegree centrality, which is computationally efficient. The effective detection of these seed nodes across the diverse community cover makes the information diffusion successful.

The contributions of our work are as follows:

  • We propose an interest group-aware framework for seed node identification for effective information dissemination.

  • We use the Leiden method for identifying the interest group structures in the given OSNs [21].

  • We utilize a simple strategy for uncovering the seed nodes over the diverse interest group cover.

  • The proposed approach is an improvement over conventional seed node detection in OSNs [14, 16].

The rest of this paper is organized as follows. Section  2 reviews the related literature on influence maximization, community detection and seed node identification in online social networks. Section  3 presents the proposed interest group-based seed node detection framework, along with the Netdegree centrality measure. Section  4 describes the eight benchmark datasets. Section  5 presents the experimental setup, results and performance analysis. Section  6 concludes the paper, and Sect.  7 outlines directions for future research.

2 Related works

The research area of seed node detection and influence maximization has been studied thoroughly in the literature. The foundational work was put forward by Kempe et al., who studied influence maximization using classical diffusion models such as the Independent Cascade (IC) and Linear Threshold (LT) models [14]. Building on this framework, many scalable methods were developed to study information dissemination in large networks [15, 16, 18].

Many studies in the literature have focused on finding seed nodes based on the structural properties of the given Online Social Network (OSN). Centrality-based approaches, such as degree, betweenness, closeness and eigenvector centrality, have been used to uncover the seed nodes of information diffusion in OSNs. These centrality-based approaches are effective in finding seed nodes but neglect their spatial distribution, which can be handled in the form of interest groups.

A major study of influence maximization was carried out by Li et al., who surveyed algorithm designs and diffusion models [17]. The study found that many influence maximization approaches concentrate seed node detection in highly looped regions, which hinders effective information diffusion.

Recent research in this area has not focused much on community structures in the form of interest groups when formulating the process of information diffusion. Tang et al. showed that network topology closely maps to social influence [18]. Qi et al. used community structure for influential user detection and showed that community-based seed detection can make information diffusion more effective across the network [19]. However, the computational complexity of these approaches is very high due to complex influence estimation mechanisms.

Community-based interest group identification plays an important role in Social Network Analysis (SNA). Traditional methods of community detection sometimes generate communities that are loosely connected. The improved Leiden algorithm, which uncovers well-connected communities, was proposed by Traag et al. and can therefore be used for finding high-quality interest groups [21].

Table 1 provides an overview of the most important classes of seed node detection and influence maximization algorithms reported in the literature. Classical influence maximization approaches mostly aim at finding globally influential nodes by applying diffusion models, optimization techniques or topology-based centrality measures. These techniques are successful in maximizing influence but do not necessarily consider the community structure of online social networks, causing redundant selection of seeds in highly connected areas. In recent years, community-aware algorithms, graph-based optimization, deep learning models and reinforcement learning frameworks have been introduced to enhance diffusion performance. These approaches, however, tend to be more complex to compute and require significant model training or complex optimization procedures, which limits their use in large-scale networks.

Table 1 Comparative analysis of existing seed node detection and influence maximization techniques in online social networks
Full size table

The Leiden algorithm has improved community detection by creating well-connected communities, but it does not directly address the identification of influential seed nodes. These observations have led to the proposed framework, which combines Leiden-based interest group detection with the proposed Netdegree Centrality to identify influential users independently in each community. This allows for a balanced distribution of seeds, good diffusion coverage and high computational scalability, so that the proposed approach is feasible for large and heterogeneous online social networks. Although much work has been done in the field, gaps remain, as some approaches focus on global influence mechanisms or propose computationally expensive methods for uncovering influential users. Therefore, this study focuses on seed node detection using Netdegree centrality based on interest group identification with the Leiden algorithm. The proposed approach offers a scalable and computationally favourable solution for seed node detection.

3 Proposed methodologies

In the proposed approach, the given social network is modelled as a graph G (V, E), where V represents the nonempty set of users of the network and E represents the set of edges modelling the relationships and interactions among the users. The graph G (V, E) can be directed or undirected depending on the network; in our case, it is directed.

The interest group cover is obtained using the efficient Leiden algorithm. Leiden is based on RB Configuration quality optimization, which has shown efficiency over many other techniques, and we therefore use it in our work. This method favours interest groups that are more densely connected internally than with external social agents. For a network represented as a graph G, the Leiden method divides the nodes V into a number of disjoint interest groups IGs = {IG1, IG2, …, IGk}, where each IGi is a subset of V representing an interest group of densely interconnected members. The use of the Leiden approach is mainly motivated by its robustness and scalability.

The Leiden method of community detection is used to identify structurally tightly knit communities within the given social network. In this study, these Leiden-based communities are subsequently referred to as interest groups, as the nodes within a given community exhibit relatively stronger connectivity pattern. Thus, the term interest group denotes an application-level interpretation of a Leiden-based community; it does not directly represent the users’ semantic interests.

Once the cover of interest groups is found, the seed nodes are detected using the Netdegree centrality of each node within each interest group. The Netdegree Centrality of a node v in an interest group is defined in Eq. (1).

$$NDC\left(v\right)={D}_{out}\left(v\right)-{D}_{in}\left(v\right)$$
(1)

where Dout(v) is the out-degree of node v, Din(v) is the in-degree of node v, and NDC(v) is the Netdegree Centrality of node v.

For each node v in the given interest group, if NDC(v) > 0, node v is classified as a seed node; otherwise, if NDC(v) ≤ 0, the node is classified as a non-seed node. This process of seed node detection covers all the interest groups in the given network. The pseudocode of the whole approach is shown in Algorithm 1, and the diffusion simulation used for evaluation is shown in Algorithm 2.

Algorithm 1
Algorithm 1
Full size image

Interest group-based seed node detection

Algorithm 2
Algorithm 2
Full size image

IC-based simulation

$$\:Coverage\:\left(\%\right)=\frac{{N}_{activated}}{{N}_{total}}\times\:100$$
(2)

The computational complexity of the proposed framework is analysed according to its stages. A given social network is represented as a graph G (V, E), with V as the nonempty set of n nodes and E as the set of m edges. The NDC of all n nodes is computed in O (n + m) time, and the selection of nodes satisfying NDC(v) > 0 needs O(n) time. Ranking the nodes according to their NDC values requires O (n log n) time in the worst case; otherwise, in the best case, it requires O (c log c), where c is the number of seed nodes that need to be ranked. Thus, the computational complexity of the proposed framework is TLeiden + O (n + m) + O(n). The diffusion simulation is considered separately and is evaluated empirically.

4 Datasets

The proposed approach was implemented in Python using Google Collaboratory and tested on eight datasets collected from the Pajek and SNAP dataset repositories. The details of the datasets are summarised below.

  • Student Government [22,23,24]: A small interaction network dataset from the University of Ljubljana with 11 nodes and 41 edges. This dataset is directed.

  • ModMath [25, 26]: A directed social network dataset with 38 nodes representing school superintendents and 76 edges representing friendship relationships.

  • HiTech [27, 28]: A directed organizational communication network from a small company with 36 nodes representing employees and 147 edges.

  • Wiki-Vote [29, 30]: A directed social network of Wikipedia elections (2008) with 7115 nodes and 103,689 edges.

  • Email URV [31]: A directed organizational communication network of the University of Rovira i Virgili with 1133 nodes and 10,902 edges.

  • Email-EU Core [32]: A directed institutional email network with 1005 nodes and 25,571 edges.

  • Congress Twitter [33, 34]: A directed Twitter communication network of the 117th US Congress with 475 nodes and 13,289 edges.

  • Slashdot [35]: A directed large-scale friendship network on the Slashdot platform with 82,168 nodes and 948,464 edges.

5 Results and discussion

We implemented and tested the proposed interest group (IG)-based seed node detection framework in Python via Google Colaboratory on a wide range of benchmark social network datasets from the Pajek and SNAP repositories. The datasets are networks with different structural properties, ranging from very small interaction graphs to online social networks with tens of thousands of users. This variety makes it possible to thoroughly test the scalability and robustness of the proposed framework and its capacity to find influential seed nodes in several network topologies. The datasets include educational, organizational, communication, political and online friendship networks: Student Government, ModMath, HiTech, Wiki-Vote, Email URV, Email-EU Core, Congress Twitter and Slashdot. The proposed framework starts by dividing each network into interest groups using the Leiden community detection algorithm and then finds influential seed nodes in each community by measuring the proposed Netdegree Centrality. This two-stage approach enables influential users to be selected in all major network areas, not just in one high-density region.

5.1 Results on the student government dataset

The proposed methodology was first tested on the Student Government dataset, and the results are shown in Fig. 1.

Fig. 1
Fig. 1
Full size image

Interest groups and NDC-selected seed nodes detected by the proposed framework on the Student Government network

Figure 1 shows the experimental results for the Student Government network, one of the smallest networks studied. The network has only 11 vertices and 41 relationships, but it is a good benchmark for visualizing the behaviour of the proposed framework. The Leiden algorithm divides the network into three well-connected interest groups, which match the naturally connected members well. A Leiden resolution parameter of 1.0 was used to keep the interest groups sufficiently large and to prevent excessive fragmentation. In contrast to traditional graph partitioning methods, which tend to yield loosely connected communities, the Leiden algorithm produces smaller, more cohesive and less fragmented communities. This behaviour is important, as information tends to spread quickly within a tightly knit community before moving across community lines.

After detecting communities, the Netdegree Centrality measure finds the most influential users in each community. The chosen seed nodes are located in various communities of the graph rather than concentrated in one area. Balanced seeding allows multiple communities to be activated at the same time in the diffusion process. Despite the small size of this network, the proposed strategy shows that it is possible to achieve greater initial coverage than by simply choosing the globally highest-degree nodes. Thus, information started at these distributed seed nodes can spread efficiently through the entire network with minimal overlap between activated users. Moreover, the visualization shows that each detected community has at least one node that can influence others in the initial dissemination of information, preventing isolated communities from remaining idle at the beginning.

The distribution of seed nodes in each interest group is shown in Fig. 2.

Fig. 2
Fig. 2
Full size image

Distribution of NDC-selected seed nodes across the identified Leiden communities of the student government network

The seed nodes were used to uncover the information diffusion in each interest group using Independent Cascade (IC) model-based simulation. The simulation begins by activating the detected seed nodes in each interest group as sources of information dissemination; the IC model then gives each active node in an interest group a single opportunity to activate each of its inactive neighbours with a predefined probability of 0.10. The dissemination process continues until no more activations can take place.

To handle the stochastic nature of the diffusion process, the IC model-based simulation was repeated for 100 Monte Carlo runs using random seeds. The total diffusion spread and diffusion coverage were recorded for every simulation. The diffusion spread was recorded as the number of activated nodes, and the diffusion coverage was recorded as the percentage of activated nodes relative to the total number of nodes in the network.

The mean and standard deviation of information spread and coverage over 100 simulations were used to quantify the performance of NDC-based seed nodes. The non-seed activation and cascade depth were recorded as supplementary diffusion statistics to provide further insight into the information propagation process.

The results obtained from the simulation using 100 Monte Carlo runs are presented in terms of diffusion spread and diffusion coverage. Figure 3 depicts the diffusion spread obtained by the NDC-based seed nodes over the simulation runs, while Fig. 4 shows the corresponding diffusion coverage of the network. These results provide a quantitative assessment of the extent to which information disseminates from the seed nodes under the Independent Cascade (IC) model.

Fig. 3
Fig. 3
Full size image

Independent Cascade diffusion spread across 100 Monte Carlo simulations on the Student Government dataset

Fig. 4
Fig. 4
Full size image

Independent Cascade diffusion coverage across 100 Monte Carlo simulations on the Student Government dataset

The overall results on the Student Government dataset show that NDC-based seed selection initiated measurable information dissemination across the network, as indicated by the observed diffusion spread and coverage.

Under the same IC-based simulation settings and an equal seed budget, the proposed NDC-based seed selection produced higher mean diffusion spread and diffusion coverage on the Student Government dataset than other centrality methods in the literature, namely degree, betweenness, PageRank and eigenvector centrality, as shown in Table 2.

Table 2 Comparison of NDC with standard centrality-based seed selection methods under the Independent Cascade model on the Student Government dataset
Full size table

To evaluate the contribution of NDC to the proposed Leiden-based seed selection framework, an ablation study was conducted by replacing Netdegree centrality with degree centrality while keeping the Leiden-based interest groups, seed budget, dissemination probability and Monte Carlo settings unchanged. The two approaches were evaluated using diffusion spread, diffusion coverage, non-seed activation and cascade depth under the IC model of information diffusion. The ablation results on the Student Government dataset are shown in Table 3.

Table 3 Ablation comparison of Leiden-based seed selection using Netdegree Centrality and degree centrality under the Independent Cascade model on the Student Government dataset
Full size table

The ablation results show that the proposed Leiden + NDC achieved higher diffusion spread and coverage than Leiden + Degree, illustrating the contribution of NDC to the Leiden-based seed selection framework.

5.2 Results on the ModMath and HiTech datasets

Two directed datasets, ModMath with 38 nodes and 76 directed edges and HiTech with 36 nodes and 147 edges, were tested using the proposed methodology. The experimental settings were kept the same as for the Student Government dataset to maintain a consistent evaluation. The performance of NDC and the baseline centrality measures is compared in Table 4.

Table 4 Comparative analysis of NDC with other well-known centrality measures on the ModMath and HiTech datasets
Full size table

The experimental results for the diffusion spread are depicted in Fig. 5.

Fig. 5
Fig. 5
Full size image

Comparative analysis of diffusion spread on the ModMath and HiTech datasets

The results for the diffusion coverage on these two datasets are depicted in Fig. 6.

Fig. 6
Fig. 6
Full size image

Diffusion coverage comparison on the ModMath and HiTech datasets

The comparative ablation analysis of Leiden + NDC and Leiden + Degree is presented in Table 5.

Table 5 Ablation study of Leiden + NDC and Leiden + Degree on the ModMath and HiTech datasets
Full size table

Overall, Tables 4 and 5; Figs. 5 and 6 provide a comparative evaluation of the proposed methodology on two well-known directed datasets.

The results on the ModMath social network show that the proposed framework is an effective way to unveil the underlying community structure even when the network connections are not uniformly distributed. The Leiden algorithm identifies multiple interest groups, represented by clusters of closely interacting people. Communication networks tend to have local interactions, which is why it is crucial to identify these communities to maximize the reach of information.

The proposed Netdegree Centrality identifies seed nodes independently in each interest group after partitioning the network. This gives a much more balanced distribution of seeds than conventional centrality measures, which tend to pick nodes adjacent to high-degree nodes in the same highly connected part of the network. Instead, the proposed framework ensures that the influential nodes are spread structurally across the network. The visualization also shows that the algorithm picks some of the bridge nodes between different communities when they have a sufficiently high Netdegree. Bridge nodes enable information to spread between communities and therefore beyond local neighbourhoods much more rapidly. Thus, the proposed framework ensures better diffusion ability without redundantly selected seed nodes.

The results on the HiTech dataset show that the Leiden algorithm discovers natural communication communities without creating disconnected partitions. The proposed Netdegree Centrality identifies the locally influential users within each of the detected communities. Unlike traditional degree-based measures, which favour globally connected people, the proposed measure maintains community diversity while considering local neighbourhood interactions. This means that influential users can come from a number of communication clusters instead of one dominant department.

This distributed seed configuration greatly improves network coverage, as information originates from multiple organizational units at once. This property is especially advantageous for organizational communication, where announcements, emergency alerts or policy changes must reach many people at once. The experimental visualization thus verifies that combining community detection with local influence estimation yields a markedly better spatial distribution of seeds than topology-based global centrality methods.

The two datasets represent two different real-world contexts. ModMath represents the friendship ties among school superintendents and was originally used to study the diffusion of a new mathematics teaching method across school systems. HiTech represents the friendships among the employees of a small high-technology company and is associated with the diffusion of support and opposition during an attempted unionization process. In the current study, the proposed methodology identifies information diffusion sources as seed nodes using the Leiden method with Netdegree centrality and evaluates the associated diffusion process under the Independent Cascade model. Thus, the experiments on these two social networks show how the strategic selection of seed nodes can facilitate information dissemination in educational and organizational social network settings.

5.3 Sensitivity analysis

To evaluate the robustness of the proposed framework, a sensitivity analysis was conducted on the directed HiTech dataset. The effect of the Leiden resolution parameter on the number of detected communities is depicted in Fig. 7, and its effect on the RB Configuration quality is shown in Fig. 8; Table 6 summarises the community structure and seed selection for each resolution value. The diffusion performance of Leiden + NDC and Leiden + Degree over the different resolution values and propagation probabilities under the IC model is reported in Table 7, and the mean diffusion spread of the two approaches across resolution values is compared in Fig. 9. In these experiments, propagation probabilities from 0.05 to 0.25 were used, with 100 Monte Carlo simulations and a fixed seed budget of five nodes.

Fig. 7
Fig. 7
Full size image

Number of communities detected by the Leiden algorithm as a function of the resolution parameter on the HiTech dataset

Fig. 8
Fig. 8
Full size image

Leiden objective (RB Configuration quality) as a function of the resolution parameter on the HiTech dataset

Table 6 Sensitivity of community structure and seed selection to the Leiden resolution parameter on the HiTech dataset
Full size table
Table 7 Diffusion performance of Leiden + Degree and Leiden + NDC across Leiden resolution values and propagation probabilities under the Independent Cascade model on the HiTech dataset
Full size table
Fig. 9
Fig. 9
Full size image

Mean diffusion spread of Leiden + NDC and Leiden + Degree across Leiden resolution values on the HiTech dataset (propagation probability p = 0.15)

5.4 Results across all datasets

The number of interest groups detected in each benchmark dataset is summarized in Fig. 10. The results clearly show that the number of communities increases as the size and structural complexity of the network increase. Small networks such as Student Government, ModMath and HiTech naturally have a limited number of communities because their interaction patterns are relatively compact. Larger communication networks such as Email URV, Email-EU Core, Wiki-Vote, Congress Twitter and Slashdot, on the other hand, have much more complex interactions and connections among users, and thereby much larger numbers of interest groups.

Fig. 10
Fig. 10
Full size image

Number of interest groups detected in each benchmark dataset

The results highlight several significant points. First, the framework scales from a network with a handful of nodes to one with over eighty thousand users. The scalability of the Leiden algorithm ensures that the community detection process remains computationally efficient. Second, the identified communities accurately represent the local structure of real-world social networks. Communities are often a good foundation for the selection of influential seed nodes because information tends to spread faster within communities than between them. Third, the differences in the number of communities observed across the datasets are not intrinsic properties of the proposed algorithm but rather inherent features of the datasets. Dense communication networks are known to generate more recognisable communities than sparse interaction networks. In general, the results presented in Fig. 10 confirm that the proposed framework handles various network topologies effectively and consistently discovers meaningful interest group structures for effective seed selection.

The most important result of this study is shown in Fig. 11, where the distribution of the detected seed nodes is plotted across all discovered interest groups in each dataset. Unlike other influence maximization approaches, which often pick multiple influential nodes within one dense community, the proposed framework spreads the seed nodes across all detected communities. This diversification greatly enhances the initial diffusion coverage, since each of the major communities has at least one influential member able to start an independent diffusion cascade.

Fig. 11
Fig. 11
Full size image

Number of seed nodes detected across the interest groups of each benchmark dataset

As the size of the dataset increases, more communities naturally need local representatives, which, as the experimental results show, increases the number of seed nodes. On the other hand, smaller networks can be covered with fewer seed nodes. It is noteworthy that the selected seed nodes in Fig. 11 are neither evenly distributed nor densely clustered. Rather, they are located according to the underlying community organization identified by the Leiden algorithm. This means that information can spread within multiple communities while at the same time crossing community boundaries via bridge nodes.

This well-balanced distribution significantly decreases redundant influence overlap, which is often a problem when multiple globally influential nodes lie in the same neighbourhood. Minimizing the overlap between the selected seeds maximizes the utility of each seed and enhances the information dissemination process. The results also show that the computation of Netdegree Centrality scales well. The framework efficiently finds representative influential nodes for all detected communities, even for very large datasets such as Slashdot and Wiki-Vote, without the costly influence simulations or iterative optimization procedures used by many influence maximization algorithms.

Overall, the experimental results demonstrate the effectiveness of combining community detection with local influence estimation. Current centrality-based methods tend to identify globally important nodes based on structural properties such as degree, betweenness, closeness or eigenvector centrality. While these methods work well for finding highly connected users, they tend to choose several neighbouring nodes from the same highly connected region, which restricts diffusion coverage [17, 19, 20].

The proposed framework, on the other hand, follows a community-aware approach that first uses the Leiden algorithm to partition the network into meaningful interest groups and then applies the proposed Netdegree Centrality within each community. This hierarchical structure ensures that influential nodes are spread throughout the network, so that diffusion cascades start simultaneously in different regions. This diversification significantly enhances information propagation throughout the network and limits redundant activation of neighbouring users. Another benefit is computational efficiency: the proposed approach requires community detection and a local Netdegree computation, followed by a diffusion simulation using the IC model of information diffusion. For this reason, the framework can be used for small organizational networks as well as very large online social networks with tens of thousands of users.

5.5 Computational complexity

The computational complexity of the major processing stages of the proposed framework was analysed. The analysis considers interest group/community detection, NDC computation, seed node selection, NDC ranking and diffusion-based simulation separately. The complexity characteristics are given in Table 8. This stage-wise analysis provides a clear understanding of the computational requirements of the proposed framework across networks of different sizes and structures.

Table 8 Computational complexity and runtime characteristics of the proposed framework
Full size table

Overall, the analysis shows that the proposed framework maintains manageable computational requirements, with NDC computation and seed selection remaining computationally efficient as the network size grows.

6 Conclusions

In this paper, we proposed a Leiden interest group-based seed node detection framework for efficient information diffusion in Online Social Networks (OSNs). In contrast to traditional approaches to influence maximization that use global centrality measures, the proposed framework combines Leiden-based interest group detection with a new centrality metric, Netdegree Centrality (NDC), to find influential seed nodes within each community. By selecting seed nodes separately for each detected interest group, the proposed approach prevents the seed nodes from being concentrated in one area while retaining high influence. Experimental evaluation was performed on eight benchmark networks, ranging from small organizational networks to large social media graphs, with varying sizes and structural complexity. The experimental results showed that the proposed approach was able to uncover significant community structures and to find representative influential users in all interest groups. Therefore, the framework increased the diffusion readiness of the network with reasonable overall computational efficiency. The proposed methodology is scalable and robust, can be adapted to diverse real-world network topologies, and can be applied in many real-world applications such as viral marketing, organizational communication, recommendation systems, emergency information dissemination, public awareness campaigns and influence maximization on online social platforms. Overall, the proposed method of combining community-aware analysis with local influence estimation is feasible and efficient for identifying seeds in large networks.

7 Future scope

The proposed framework shows good performance in detecting seed nodes with a community-aware approach, but several research directions remain to be explored. First, the current work does not consider the dynamic nature of network structures, whereas real-world online social networks constantly evolve through the addition and removal of users and relationships. The proposed methodology could be extended to dynamic and temporal social networks, allowing adaptive seed selection in continuously changing environments. Second, future work can leverage edge weights, user activity, trust relationships and interaction frequencies to create a more realistic influence estimation model. The proposed Netdegree Centrality can also be combined with graph neural networks, transformer-based node embeddings, reinforcement learning or evolutionary optimization algorithms to further improve the accuracy of seed selection. Combining multiple diffusion models, such as the Independent Cascade and Linear Threshold models, is another promising approach for capturing diffusion behaviour under various propagation mechanisms. Moreover, comparing the performance of the proposed framework with the most recent deep learning and community-aware influence maximization algorithms on larger benchmark datasets would give a better understanding of its impact. Finally, the methodology can be applied in areas such as misinformation containment, epidemic awareness, personalized recommendation systems, smart cities and cyber threat intelligence, where efficient information dissemination and influential user identification are crucial for timely and effective decision-making.