Language models have achieved good results in many tasks. However, there are still some challenges with reasoning due to insufficient knowledge, which leads to poor performance of the models on specific tasks. We proposed a framework based on the existing solution (i.e., GreaseLM), using a large language model to replace some complex modules. Our approach converts the knowledge graph paths into natural language sentences, providing contextual support to enhance the reasoning capability of the large language models. Using the framework based on a large language model is simpler and easier to manage, and it also achieves better accuracy on multi-choice question-answering tasks.

Monitoring the execution of distributed microservices-based software applications is a complex task. As more and more institutions conduct business in a distributed environment, a large amount of user data and transaction data are generated at an accelerated rate in such an environment. As a result, it becomes a big challenge to carry out user analytics with user interaction and transaction data in real-time. Monitoring is one of the most important approaches to getting instant situations of users and transactions. Ideally, the goal of monitoring is to carry out analysis in real-time and be highly scalable with comprehensive analysis and predictions on user interaction and transaction data to gain deeper and facilitate decision-making for the stakeholders. Therefore, we concentrate on what data can be captured to gain insight into users using the platform and how we can capture such data effectively to learn application usage patterns. To achieve this goal, (1) we present how different events data be captured from execution logs of software applications in distributed microservices, and (2) we utilize machine learning methods to predict usage patterns which are generated from interaction between user and system. For example, we show how error types can be predicted in real-time using machine learning and then enable real-time monitoring. The capability to perform analysis based on the continuously growing data volume is considered to be our solution's scalability characteristic. Microservices are used as containers that are managed by Kubernetes. Event logs generated in each microservice are the important data source for executing monitoring. The experiment shows our system can monitor user actions and transactions in real time and that the model capacity is scalable.

Analyzing public sentiment about an entity or issue can be of interest to governments and businesses alike. There is a growing body of research that attempt to devise new sentiment analysis techniques, especially techniques based on machine learning. These machine learning-based techniques typically require large, labeled training data with a large number of instances for training in order to provide reasonable accuracy in sentiment analysis. However, labelling large volumes of data is tedious and expensive. In this paper, we propose a multi-class sentiment analysis technique, named SG-Elect, utilizing cutting-edge transformer based pre-trained models along with more traditional machine learning based approaches in a semi-supervised setting. Our experiments demonstrate that SG-Elect outperforms a recent state-of-the-art baseline for all three datasets.

There is a continuous evolution in the technology industry with different types of devices being produced and connected to the internet. Multiple types of applications run on the different devices, thereby generating a complex and huge amount of traffic (i.e., Big Data) on the internet. This has made it difficult and challenging for different Internet Service Providers (ISPs) to maintain their service quality and keep their networks and services secure. It is important for service providers to have the ability to classify large and complex network traffic to help promote a better Quality of Service (QoS) and manage networks. In this paper, we utilize, apply and evaluate different machine and deep learning techniques for classification of encrypted network traffic to help in managing networks, and thereby, help in improving quality and security of network. A comparison between the different algorithms used is presented. The experiment results show that ANN+XGB, CNN+XGB, and CapsNet+XGB performed better than the LSTM+XGB and Ensemble model in the classification of encrypted network traffic with accuracies of 96%, 96%, 96%, 93%, and 95% respectively using a total of 23 statistical features. More statistical features were considered compared to other existing related works to improve the process of the classification and different hidden patterns associated with the statistical features. The results show the effectiveness of the machine and deep learning algorithms for the classififcation of encrypted network traffic into different categories.

Heart diseases are one of the deadly but are silent killers for humans, which results in the increase in death rate of sufferers every year. The World Health Organization (WHO), in the year 2016, reported that 17.9 million deaths that occur worldwide per year are a result of heart disease. In the health care sector, enormous data are being generated on a daily basis, which contains different types of data, and acquiring knowledge from these data is essential. This knowledge can be acquired using various data mining techniques to mine knowledge by designing models from the medical records dataset. We implement a machine learning based system that can detect and predict heart diseases in patients using the medical records of patients. The proposed solution is based on existing techniques like Random Forest Bayesian Classification and Logistic Regression, which provides a decision support system for medical professionals to detect and predict heart diseases and heart attacks in humans or individuals using risk factors of heart disease. The dataset used in our model consists of 18 features (risk factors) and 1990 observations after performing preprocessing. It was then split into 80% train sets and 20% test sets. Using real medical records of patients, a series of experiments were conducted to examine the performance and accuracy of the proposed system. The system was implemented in RStudio platform which predicts the risk of heart disease in patients. The compared results showed that the system performance and accuracy are acceptable with heart disease classification accuracy of 92.44% for Random Forest, 61.96%, and 59.7% for Naïve Bayes Classifier and Logistic Regression, respectively.

Recent applications like social networks and IoT are the main source of the massive amount of data generated every day. Time series data is a major form where data is sequenced and indexed by timestamps. Multiple data mining techniques are applied to discover the behavior of time series datasets, periodic pattern mining is one of them. Many sequential pattern mining algorithms were presented, some of them built suffix trees and performed early pruning while other algorithms used pattern-growth techniques such as projection. A few algorithms performed Apriori-based techniques where lattice trees were built and traversed. However, most algorithms suffer from time and space issues when mining large scale time series sequences. In our paper, we present a solution that utilizes advanced and sophisticated distributed systems such as MapReduce framework. It splits the original sequence and distributes its segments across thousands of nodes in the MapReduce infrastructure. We use different training datasets to evaluate both traditional pattern mining algorithms and our MapReduce solution. After analyzing our solution in terms of time complexity, efficiency and accuracy, we clarify the advantages of processing data segments using periodic pattern mining along with MapReduce framework.

One of the most commonly used ways to monitor execution of software applications is by analyzing logs. Logs are execution foot-print of software applications that are produced and stored for real-time or post-execution analysis of execution. With the software applications becoming large, complex, distributed, web-scale, also called as big data applications, logs produced by such software applications are also large-scale. That means, such logs are large in volume, velocity and variety. That makes it crucial to have such logs analyzed in an automated, scalable and effective manner to ensure high veracity and have analytics with high value. In this paper, we present our proposed solution of a formal model for organizing and structuring logs. We then present a Bayesian deep learning network based analysis approach that utilizes the formal model for logs to detect and predict any possible faults and consequences of such faults. Moreover, we also present our MapReduce based distributed, parallel, single-pass and incremental approach to build, train and execute the proposed Bayesian deep learning framework. This helps in effective processing of logs on cloud platforms and therefore efficient handling of logs that are produced at the scale of big data by big data applications.

Event segmentation is an important step in monitoring and management applications that categorizes different events into different segments. This is important especially when applications, to be monitored and managed, are large-scale, comprehensive and data-intensive in nature. The process of segmentation is based on data clustering which is one of the key data mining methods used these days. There are several decent algorithms and techniques that exist to perform clustering on small to medium scale data. In the era of Big Data and with applications being large-scale and data-intensive in nature, there is a significant increment in volume, variety and velocity of data in the form of log events produced by such applications. This makes the task of clustering of huge amounts of data more challenging and limited. This paper presents a proposed an effective and efficient approach of event segmentation in logs. It is based on parallel k-means clustering, inherited from MapReduce paradigm, to be used for event segmentation. The proposed approach has been tested and evaluated on large-scale log data derived from real-life case-study. Evaluation includes measuring efficiency and effectiveness of the proposed solution for its usability on log data with large volume, variety and velocity, as well as its applicability on large-scale applications.

Monitoring and management of large scale applications is already a complex task because of syntactic and unstructured nature of execution data. Traditional application monitoring and management solutions focused on employing analysis techniques on unstructured and syntactic log information become limited as unstructured information cannot be well utilized to find out related events information or correlate such information with other related information from applications. Our proposed solution of semantically formalized logging fills this gap by bringing formal semantics and combining it in a meaningful way to enable automated monitoring and management of applications. Such formalized and well-structured log information helps analytical solution to maximally automate the process of monitoring and management of applications. However, while formalizing and structuring the log information, we came across several missing and incomplete data which causes hindrance in this process. In this paper, we tackle this problem and propose a social network analysis based solution to handle incomplete and missing data from application execution, possibly compute it and use it by our proposed solution of semantically formalizing and structured logs with adapted data mining techniques to enable automated and effective application monitoring and management. We demonstrate from an industrial use-case application that how historical data from application execution is stored using semantic logging and utilized with standard social-network analysis techniques to find out missing values in incomplete data and perform application monitoring and management.

Matchmaking services connecting consumers and providers on the internet have become phenomenally important in today's world. Whereas the matchmaking was by traditional media such as print and television in the past it is now expected to include the internet. Consumers expect these services to be readily available as this can only be accomplished via the internet Providers that do not have an internet presence are therefore severely disadvantage in the competition for customers (consumers). The decline of the traditional physical music store and the ascendancy of the virtual iTunes store is a perfect example of this. As the traditional stores go bankrupt and go out of business has also become the largest music retailer in the United States of America. Consumers simply do not want to spend extra energy to get what they want. If the music can be purchased and downloaded from the comfort of their own home, the consumers (customers) will do just that. Therefore online matchmaking services are a hot topic of discussion for many companies, as they are finding ways to provide the fastest, cheapest and most efficient methods for consumers to easily access their services.

Searching for the right information over the Web is not straight-forward. In the era of high speed internet, high capacity networks, and interactive Web applications, it has become even easier for the users to publish data online. A huge amount of data is published over the internet; every data is in the form of web pages, news, blogs and other material, etc. Similarly, for search engines like Google and Yahoo, it becomes rather hard to find out the right information, i.e., as per user's preferences; search results for same query differ in priority for different users. In this paper, we proposed a way to prioritize search results of search engines like Google, based on the personal interests and context of users. In order to find out personal interest and context, we follow a unique approach of (1) finding out activities of a user of his/her social-network, (2) finding out what information does the social networks (i.e., friends and community) provide to the user. Based on this information, we have developed a methodology that takes into account the information about social networks and prioritize search results from Web search engine.

Triple space computing is a communication and coordination paradigm that allows communication of semantic technologies in general, by publish and read of semantic data. It has also been provided as an underline communication middleware for semantic Web services. While focusing on scalability and openness of triple space computing to bring it into its full potential in a global view, the possibility of heterogeneity among different users communicating over triple space is very likely to arise. This paper focuses on providing the triple space computing with data mediation to enable easy integration of data, information, and knowledge. Mediation is a technique to overcome heterogeneity issues in a system, i.e. to remove differences in the syntactic representation and the intended semantics of data that is exchanged [2]. The paper introduces an abstract mapping language and shows how mapping rules can be created using this mapping language. It further proposes mediation APIs for users and internal system. It also explains the grounding of mediation mapping rules to triple space. It further provides a refined version of architecture of the mediation engine along with its bindings with other components of triple space computing paradigm.

The application of semantics in Web services as semantic Web services for dynamic discovery, composition, invocation and monitoring has been very helpful in enabling Enterprise Application Integration and E-Commerce. There are many initiatives that aim to realize the semantic Web services to enable effective exploitation of semantic annotations, and two major of them are Web service modeling ontology (WSMO) and ontology Web language for services (OWL-S). Several tools have been developed to realize both the conceptual models i.e. Web services execution environment (WSMX) is the reference implementation for WSMO, on the other side OWL-S reference implementation exists in the form of loose collection of individual tools like OWL-S Editor, OWL-S matchmaker, OWL-S virtual machine, OWL-S IDE, WSDL20WL-S converter and OWL-S2UDDI converter etc. In this paper, we have conducted a comparison of both the reference implementations to identify similarities and differences between them and to evaluate their potential to become widely accepted implementation recommendations.