Abstract
Data management for large-scale data library services with mining procedures improves the availability and readiness of heterogeneous sources. The heterogeneous data sources are assimilated as a single entity through mining procedures to meet the data demands. This article introduces connectivity-persistent data mining method (CDMM) to improve the data handling precision with boosting availability. The proposed method relies on federated learning for identifying the service demands, thereby providing data mining. The learning paradigm accumulates information on shared data library existence over various services. Based on the availability, further mining demands are forwarded to the data management system. If the existence verified by the federated learning is adaptable, then sharing-enabled mining is endorsed for the connected users. The data management then augments several heterogeneous shared libraries to meet the mining requirements. This process is reversible based on the service mode and existence. Therefore, the proposed method improves data availability with less mining and access time and fewer failures.
1. Introduction
Data mining is a process that extracts certain patterns and useful details from a large set of data. Data mining provides the necessary set of data for the analysis process. Various methods and techniques are used to perform the data mining process. Data mining is a complicated task in every application [1]. Data mining also identifies the problems identified by the data analysis process. A data management system for mining services is a crucial task that manages a huge amount of data. Data management is a process that protects, store, collect, organize, and manage data that provide an appropriate set of data for various processes [2]. Data mining services are a process that converts the raw data into a useful set of data that is used for further processes. Various management services are used for the data mining process using the machine learning (ML) approach [3]. A data management system improves the performance and efficiency rate of the system, improving the accuracy rate in the decision-making process. Data management systems manage the data collected by an application and organization. Storing and managing a data management system is mostly used for data mining. Data mining services and details are handled by a management system [4, 5].
Various data mining types are available to identify the dataset’s important patterns. An organization widely uses the service demand-based data mining method. The data mining process plays a major role in every organization that helps enhance an organization's performance and feasibility [6]. The organization gives requirements and preferences that provide a set of demands over the data mining process. The service demand-based data mining process provides an accurate dataset for the decision-making process that reduces the failure rate [7]. Organizations demand a certain set of services for the data mining process. The real-time data mining process is a complicated task to perform in every management system [8]. The classification method is used in the service demand-based data mining process. The classification method classifies the dataset by combining it with given service demand. Various demands and requests are demanded by an organization for the data mining process. Companies and industries demand a certain set of services that improve the accuracy rate in the data mining process [9, 10].
Machine learning (ML) techniques are widely used for various applications to perform prediction and analysis. ML techniques improve the accuracy rate in both the analysis and prediction process. ML techniques are also used in data mining to enhance the service accuracy rate. ML technique-based data mining process identifies the important features and patterns from a huge set of data [11, 12]. The convolutional neural network (CNN) algorithm is commonly used for data mining. The feature extraction process is used in CNN to extract the features presented in a given raw dataset [13]. The classification process classifies the features extracted from the feature extraction process. CNN predicts the actual data necessary for an application [14]. The support vector machine (SVM) algorithm is also used for data mining. SVM first trains the dataset with an important set of features collected by the analysis. SVM reduces the latency and error rates in the computation process, which improves the efficiency rate of the system. The data analysis process analyzes the raw data stored in the database [15]. The main contribution of CDMM is as follows.(i)The suggested method focuses on federated learning for recognizing the service requests and consequently enabling data mining. The learning paradigm accumulates information about shared data library presence over numerous services.(ii)The data management then augments numerous heterogeneous shared libraries to match the mining needs. This process is adjustable based on the service mode and existence.(iii)Therefore, the suggested strategy improves data availability with less mining and access time and fewer failures.
2. Related Works
Huang et al. [16] introduced a new algorithm for fast mining frequent patterns using a distributed computing system. Frequent pattern mining identifies the important patterns that are presented in a given dataset and reduce the latency rate in the analysis process. The big data analysis process is used here to analyze the huge amount of data and produce an optimal dataset for further data mining. The proposed method improves the accuracy rate in the execution process, enhancing the system’s performance. The proposed method reduces time and energy consumption in the execution process.
Xie et al. [17] proposed an information filtering and mining method for big data analysis. A support vector machine (SVM) algorithm is used here to analyze the data necessary for the mining process. The proposed method is mainly used for the retrieval process that retrieves educational images. Certain features and patterns are identified by filters that produce an optimal dataset for further analysis. The proposed method improves the performance rate and efficiency of the system.
Obregon et al. [18] introduced the data mining information as a flow method for social networking services (SNSs). The proposed discussion flow model identifies the data and provides appropriate details for the data mining process. Data mining captures interaction among communities, producing effective information about discussions. The proposed method enhances the feasibility and reliability of the system. The proposed method reduces the complexity rate and improves the mobility rate of SNS.
Bhattacharya et al. [19] proposed a mobile blockchain (MB) based data mining method as-a-service (MB-MaaS) for the Industrial Internet of Things (IIoT). MB is used here to enhance the effectiveness of rata in the analysis process. The proposed method identifies the group discussion and interaction of users in IIoT. The experimental results show that the proposed method achieves a high accuracy rate in the mining process, which improves the system's performance.
Zhang et al. [20] introduced a massive data mining-based method for mobile libraries. The filtering technique is used here to filter the candidate's available datasets in mobile libraries and produce a feasible set of data for further process. The Apriori algorithm is used here to provide optimal rules for the candidates, reducing unnecessary problems in the management system. The proposed method reduces energy and time consumption in the computation process. The proposed mining method also improves the execution time of the system.
Dhelim et al. [21] proposed a personality-aware hybrid filtering-based mining method for a social network. The personality filter first filters the traits and personalities of users and produces necessary information for the mining process. The data analysis process collects the data available in a social network that provides appropriate data for the mining process. The proposed method maximizes the accuracy rate in the data mining process that provides appropriate services to the users.
Wang et al. [22] introduced a new framework for library services and immigrant needs. The proposed framework identifies the cause of problems that are occurred in libraries. Social networks provide necessary information about the candidates, reducing the time consumption rate in the searching process. Finally, the proposed framework provides various guidelines and rules for libraries that improve the appropriate services to the users.
Xiao et al. [23] proposed a fine-grained sentiment analysis-based preference mining method. The sentiment analysis approach finds out the important emotions and characteristics of users. User features are identified by a pretraining language model that produces a feasible set of data for preference mining. Both numerical and text-relation information is analyzed by preference mining, reducing the execution process’s latency rate. The proposed method achieves a high-performance rate in providing services for the users.
Peng et al. [24] introduced a fuzzy convolutional neural network (FCNN) based on big data mining and analysis (BDMA). The feature extraction approach is used here to extract the important features available in the dataset. The feature extraction method collects an appropriate dataset for the big data analysis. The FCNN algorithm is mostly used for the recognition process that enhances the system’s feasibility. The proposed method maximizes the system's effectiveness and efficiency rate, improving the accuracy rate in the big data analysis process.
Alkathiri et al. [25] proposed a multidimensional data mining method using the MapReduce technique for a distributed environment. The MapReduce technique is used here for the ecosystem data analysis process that finds the features presented in a given dataset. Machine learning (ML) techniques are also used here to enhance the system’s feasibility. The proposed method reduces the error rate in the data mining process, improving the system’s performance.
Deng et al. [26] introduced a jointed neural network-based multimedia data stream is an information mining model. The soft clustering technique is used here to cluster the huge data available in the database. A joint neural network is implemented here to train the dataset necessary for the data mining process. The proposed data mining approach addresses the problems presented in an application. The proposed model achieves high efficiency and effectiveness rate in the mining process.
Ju et al. [27] proposed a data mining-based commodity recommendation method for online shopping. The proposed method is mostly used in e-commerce and online shopping applications. The commodity recommendation method identifies users' preferences, requests, and browsing history that provide relevant details for an application. The data mining approach analyzes the given set of data and produces a feasible set of data for the recommendation process. The proposed method improves the performance and feasibility rate of the online shopping system.
Zhou et al. [28] introduced a new data mining approach using particle swarm optimization (PSO)-based backpropagation (BP) neural network. Internet of Things (IoT) is used here to enhance the communication process among users and organizations. PSO is used here to train the dataset necessary for the data mining process. IoT collects real-time data that users produce. The proposed method increases the accuracy rate in the prediction and analysis process.
2.1. Proposed Connectivity-Persistent Data Mining Method
The data source repository is the maintenance of databases by collecting data from multiple sources meeting the objective function. It is a database infrastructure that aggregates, manages, and stores datasets mined for data analysis. It makes sharing data easier by managing it and maintaining metadata for the study of data. The aggregated data are reviewed for the type of data based on which the data are stored. The data in the repository are loaded with an increasing volume of data. In Figure 1, the proposed method is illustrated.
Service mining is influenced by federated learning to validate its existence for further sharing. This learning further operates on different service demands. If any deficiency is found, a data management system ensures data existence and availability for varying users (refer to Figure 1). The request-based services from users are generated in a particular time slot where the total number of requests r from the users is denoted as. The request from the users allocated to the data source repository s is denoted as. be the number of maximum requests from users to the data source repository as shown in the following equations:
To handle the requests, the capacity of the data source repository with its pricing of data to be provisioned is calculated. Thus, the cost of the data source repository for the request is obtained from
The delay in addressing the request to the data source repository based on the quality of experience is calculated considering the network and queuing delay. The following equation denotes the delay of the network:
The network delay and the queuing delay to fulfil the request depending on the factors such as transmission delay and propagation delay. The queuing delay is obtained from the workload network delay on the distance between the user and the data source repository. The delay in making a decision incurs further delay, which is represented as:
From 5the above equations, is the request allotted to the data source repository. The network delay for the request is as. is the distance between the user request to the data source repository. p, v are the parameters considered to scale the distance and maintain the function’s convex property. The decision-making based on the delay factors for data existence verification is presented in Figure 2.
The user requests are influenced by is sustained for the entire allocation intervals. The data availability and existence are verified is satisfied. The learning process relies on are distinguished for their existence (Figure 2). The queuing delay for the request allocated to the data source repository is obtained using the following equation:
The workload to be processed is represented as for the data source repository at a particular time slot. The service time provisioned by the data source repository allotted for the request is given by. From (7), the deficiency of the service time at the data source repository is obtained by. Thus, the upcoming request from the users must have to wait for their request to be processed. To maintain fair processing of request based on the heterogeneity of the data, the quality of experience by the users is calculated by considering the tolerable delay and the actual delay. Thus, it can be defined as in the following equations:
From the above equations, the tolerable delay and the actual delay of the request in a particular time slot are represented with. The above equations denote the quality of experience by parameter. The user is processed before a tolerable delay, and then, the requests from the users are mentioned with. If the request is not processed within the tolerable delay, then it is considered that the users are not fulfilled and the waiting time of the users is expired, and a is the parameter that mentions the rate of declination representing the quality of experience. Based on the conditions above, the quality of experience by the user request in the data source repository within the time slot is defined by
Based on the above estimation, the validations (8), (9), and (10) are performed using the federated learning model. This is depicted in Figure 3.
The learning induces multiple as defined in (8), (9), and (10) for different. Based on the sharing output, user service mining and allocations are performed. This requirement is fulfilled based on the availability factor. The delay and existence impacts are mitigated using the maximum sharing ratio and learning implication (refer to Figure 3).
2.2. Learning Implications for Data Management
Federated learning is a technique where devices are decentralized with collaboration processing service demands considering user requests. The networks with several users have been partitioned based on their interests. This number of users share the data among themselves. Data resembling common interests among the users have been identified to verify the available data. If similar data are available, then the data are shared in a decentralized manner. The model with users and their data is . These users with the data information collaborate to identify the existence of data. The users in the network are combined with the data , which is used to train the model M.
The users in the network share this model used for training; for each new data in the network among the users, a common interest procedure is to be followed. In the proposed, horizontal federated learning is considered where the users in the network communicate with each other to update the model M.
Based on the data from the users, a common interest group is created, enhancing network efficiency. The users in this group have their data. By aggregating the data from all the members, a dataset is generated. Thus, each common interest group maintains its own set of data within it. If the user in the network wishes to leave the common interest group, the user may leave with the data. Contrarily, if a user wishes to join the common interest group, the data are verified by some users in the common interest group for the relevant data. Each common interest group has its reputation for maintaining relevant or accurate data. The rewards are shared among the users in the common interest group based on the size and the relevant data they offer.
The common interest group in the network with model M., the number of users in the common interest group, update their model . The proposed model improves when the users join the common interest group, so the data availability for the users also improves. Each user in the common interest group is provided to access the shared model M. The users use the model to calculate the existence of data by finding the similarities between the requested data by the users and the availability of data in the common interest group. A cosine similarity index is used to find the potentiality of the data by identifying the similarities between the requested data and the available data, as shown in the following:
The availability of data is accepted only when most users in the common interest group find the resemblance of data between the requested data and the available data. The users in the common interest group are provided with rewards based on the amount of data that is being made available. The users in the common interest group must be made available with some sort of data by generating the data and updating the model . Else, the user in the group might be expelled from the group. The users are requested to maintain some sort of space for the data allocation. The data from other users in the common interest group are stored in the maintained space. The subset of the data sent to the other users in the joint interest group is checked for relevancy. The data are verified whether it remains fixed to maintain the data within the common interest group. It asks for recommendations from common interest groups to ensure the availability of the data. Each user is provided with some functions to maintain the reliability of the users in the common interest group. Each user interacts with a common interest group; the data are shared with its functions key. Suppose these function key does not match with the available function key list. In that case, a warning update is provided, which is shared with model M. On receiving this model update, all the users in the common interest group verify its function key. If the function key fails, the corresponding user is removed from the common interest group. The learning process for in maximizing data sharing for different mining requests is presented in Figure 4.
The management system eyes on M for different such as sharing for availability and existence. This M is modified based on such that service responses are granted with better mining outcomes. The allocated requests are granted from the mining demands (Figure 4). Once the existence of data is verified with the cosine similarity, it is allocated for the request placed by the users in a particular time slot. Suppose the requested service is not available within the common interest group. In that case, a service demand is placed upon the request based on which, using the federated learning, the service demand is addressed. The data management system using federated learning enables the storage facility to enhance the network performance minimizing the access time. It is designed to operate asynchronously, making the proposed technique more flexible. It maintains a model at both the data source repository and the user side. On registration of users with the data source repository, a global model is designed. Using this global model, the data management makes arrangements for a suitable space to store the data based on the request. If the available storage space is enough, then the version information with the model data is stored. If the higher data version is available, then the model is updated. It gets updated to the advanced version from the existing version of data. The data source repository has the privilege of designing its global model and sharing it with the users. On sharing the global model, the local model gets updated. The version of the global model is initially checked before updating the local model.
The user requests the data source repository by importing the global model data based on which the local model data are generated. In the absence of a request from the user regarding the version update, the data management system monitors the version and sends the update to the users. The data source repository aggregates the updated local model from users concerning the corresponding version of the global data. It maintains the aggregation until adequate quality is obtained. This proposed method, namely, connectivity-persistent data mining method (CDMM) managed by storing the characteristics of data from the users where the requests are stored. The data source repository uses this information to process the service demands and allocate the required data. The requests from the clients include parameters such as request ID, version of the models, and device ID. The request ID represents the identity of the request where the users and the data source repository perform a task.
The data source repository addresses the request of the users by allocating the data. The data management system searches for the data using federated learning and provides the data to the users. The version of the global model is used to update the local model; the global model gets updated by connectivity between the users and the clients. Based on the model updates, the users and data source repository manage the service demands based on the requests. The device ID represents the specific ID given to the users. A particular user can be provided with the requested data using this device ID. The corresponding data communication based on the global and local model is performed with these parameters. The proposed method achieves better data available to the users by reducing the dependency on the traditional request/service demand with minimal access time and fewer failures by providing asynchronous peer-to-peer communication between the users and the data source repository. The self-analysis for varying capacity, similarity index, and demand factors is presented in Figures 5, 6, and 7, respectively.
Figure 5 presents the analysis of cost factors and requests allocated for the varying . This method allocates is performed for the increasing mining requests. The is validated based on is accounted for maximizing . Therefore, the allocations are maximized in intervals . The learning segregates existence and availability for the requests such that is reduced. This single factor maximizes the responses by reducing the wait time for which is reduced. Based on the and learning output, the further is performed. In the process, is used for maximizing the allocation.
The analysis of availability and failure rate for the varying similarity index is presented in Figure 6. The proposed method maximizes availability by reducing cost and s. In the federated learning for is maximized for which existence is verified. If the verification fails, then is analyzed, and hence, availability is maximized. Therefore, the allocations are performed to improve the allocations post and . The failures based on is rectified by assigning such that new allocations are performed. The demands are supported in achieving fair sharing depending on the available sources. As the sharing increases, the availability is maximized by reducing failures.
Figure 7 presents an analysis of the existence and availability of the varying service demands factors. This analysis relies on and such that is performed. However, the existence is high compared to the availability such that determines its allocation. This is required by the for further sharing and analysis. Based on this, further, allocation is performed to improve availability.
2.3. Performance Assessment
This section discusses the comparative analysis results of assessing the proposed CDMM using the dataset [29]. This dataset contains Flipkart product data classified under 16 fields for 30 K products. The mining process is performed by searching a product by its “ID,” “Category,” “Title,” and “Price Range.” Such queries are reverted with appropriate “Purchase,” “Description,” “Offers,” and “Availability” information for 160 users. Similarly, the services are held for 12–20 mins for a user. From this detailing, the metrics of data availability, mining time, access time, failure rate, and sharing ratio are compared with the existing UIMS [21], FCNN-DM [24], and DMFM [16] methods.
2.4. Availability Comparison
The comparative analysis for availability is presented in Figure 8 for the varying services and users. The proposed method identifies for improvements. This improvement is analyzed in the preallocation and for the tolerable level verification. Based on these assessments, the federated learning validates and individually. The common outputs are merged across different such that availability is maximized. In particular, the availability is maximized using between two successive intervals. In the repeated assessment, is satisfied using minimization. Therefore, the conventional request allocation and service assignments are maximized. In the mining process, the available resources are shared across satisfied intervals. Therefore, the user accessing intervals are maximized with based on . This is carried forward for all based on learning outputs. Hence, this proposed method maximizes data/resource availability.
2.5. Mining Time Comparison
The proposed CDMM achieves less mining time for the varying services and users, as presented in Figure 9. First, the influencing factors for for and is estimated. Based on the delay estimation, is assigned using maximization. In the consecutive allocations, - and-based federated learning influences the delay causing factors such that is reduced. In the available allocation intervals, is the deciding factor for preventing increasing mining time for multiple resources. If the service and user concentration increase, then the factor as in(8), (9), and (10) is assessed for different conditions. These conditions are based on the time factor for preventing additional delay, and therefore, the allocation consecutively aids existence. This is unanimously pursued for and such that is improved by reducing delay. Contrarily, for the varying users, is varied such that all is allocated from the available resources. Therefore, the wait time, , is reduced, preventing additional mining time.
2.6. Access Time Comparison
The access time for the proposed method's varying users and services is less than the other methods (refer to Figure 10). The queuing and mining time in the proposed method is reduced by assigning based on . This is required to improve the allocation and processing rate. Based on the allocation capacity and accessing intervals, the availability is maximized. First, the is reduced by mining concurrent resources across varying such that is achieved. Depending on and the further access grant is provided. In particular, using the federated learning is improved for such that is increased. This is pursued to improve the existence, wherein is reduced. However, in the varying user concentration, varies across multiple preventing the balance in . Therefore, is also reduced balancing interval. The successful is increased for achieving less access time for any service users in the same interval.
2.7. Failure Rate Comparison
The resource allocation failure in the proposed method is less than in other methods. Following the varying services, is maximized by reducing in the pre- assessment. In the learning-based validation, assessment achieves fair across the varying users. This is confined between 1 to intervals such that is the same. If the similarity index is high, then -based allocations are performed. Therefore, the maximum requests are assigned with a resource in the interval. For the varying services, is varied for admitting in the continuous intervals. The proposed method identifies in all the assigned such that is maximized. This is required for maximizing , wherein the learning operates independently. From the and designed by the learning model, further allocations are performed, preventing reduction. This is required for mitigation and balancing between and . In the learning output assessment, -based allocations are prevented from interfering the a decision such that existence is updated. Therefore, the sharing (shared) resource augments the demand suppression and reduces failures (refer to Figure 11).
2.8. Sharing Ratio Comparison
The proposed method achieves a fair sharing ratio compared to the other methods, as presented in Figure 12. The sharing is enabled by reducing the delay in queuing, access, and mining, as discussed earlier. The -varying users and services are streamlined using the deviating delay and mining time to prevent additional failures. The process is responsible for performing the allocation across different tolerance factors. In the consecutive resource allocation, is the estimating factor for maximizing the sharing ratio. The data management is performed for the above factor and independently to maximize the mining process. In this process, the learning for similar features is streamlined to achieve a high repository allocation level. The process is prevented from avoiding requesting fewer allocations in the consecutive repository mining process. The collaborative allocations are performed for varying services such that is maximized. In this process, the cost suppression is maintained such that the delay is also confined. The learning process further augments the data management system for improving the availability and retaining its existence until the interval . Therefore, the repository is available for varying users and requests to improve the sharing ratio. This is not repeated until the next allocation prevents additional access time. The above analysis is summarized for varying services and users in Tables 1 and 2 respectively.
3. Conclusion
This article introduced a connectivity-persistent data mining method for improving the sharing and allocation of a library of resource-based services. The proposed method relies on federated learning for validating data existence and availability for diverse user services. The mining process is performed for heterogeneous resources based on capacity-based allocation and delay mitigation. The user service demands are satisfied using experience and tolerance-based mining assimilations for improving resource availability. Besides, the available data are shared between the users and requests based on their existence. This existence is provided by maximizing request allocation and mining between connected users. The distinct service modes through existence and allocations are performed using the federated learning process through precise decisions from the data management system. Therefore, the proposed method maximizes existence and sharing regardless of the demands across the various intervals. The proposed method maximizes the availability and sharing ratio for the varying services by 9.43% and 10.16%, respectively. It reduces the mining time, access time, and failure rate by 10.47%, 12.91%, and 6.09%, respectively.
Data Availability
Data cannot be made available due to restrictions.
Conflicts of Interest
The authors declare that there are no conflicts of interest regarding the publication of this article.
Acknowledgments
This work was supported by the research project of China Academic Library & Information System (CALIS) National Agricultural Literature Information Center in 2022 “The Research about the Service Mode and Construction Path of the Scientific Research Date in Universities by the Era of Big Data.” Project Number: 2022012.