Creative Commons

This thesis reports on a series of novel approaches enabling knowledge acquisition through exploiting various machine learning capabilities. Interdisciplinary approaches may expose new possibilities for data analytics. Two theoretical frameworks are conceptualized reporting on findings that relate to the research domains of Social Media (SM) and Energy. Common methods/algorithms/tools may be utilized for knowledge extraction considering specific mining tasks. The first theoretical framework presents and combines three novel approaches in Social Media domain elaborating in mining tasks related with Social Media Types (SMTs), Social Media Topic Extraction (SMTE) and Social Media Sentiment Analysis (SMSA). SMTs are evaluated through a novel hypothesis-based data driven methodology that analyses Social Media Platforms (SMPs) and categorizes SMPs based on their services proposing new SMTs. The proposed methodology evaluates a new taxonomy, based on a mixture of hypothesis and data driven approach utilizing association rules and clustering algorithms. As a result, three new SMTs emerge, namely Social, Entertainment and Profiling networks, that update and capture emerging SMP services. Regarding SMTE, this study utilizes Twitter data to mine association rules and extract knowledge about public attitudes. COVID-19 pandemic acts as the use case, analysing crawled tweets. The approach incorporates topic extraction and visualization techniques, to form word clusters that infer to themes of opinions. Association rule mining is utilized to improve the process of extracted topics, producing more accurate and generic results. For the examined period, out of 50 initially retrieved topics with common SMTE methods, the proposed novel approach manages to reduce topics to just a few ones. SMSA relates to the identification and analysis of sentiment polarity in microblogging data. Such a mining task enables new possibilities for knowledge extraction and evaluation of public sentiment in response to global events, producing valuable insights. COVID-19 is the use case, gathering data from Twitter. The main objective in this topic is the evaluation of a possible correlation between public sentiment and the number of cases and deaths attributed to COVID-19. Findings iv “Interdisciplinary data science methods using machine learning for enhanced knowledge acquisition” correlate sentiment polarity with announced deaths, starting 41 days and expanding up to three days prior to the count. Also a strong correlation is identified, between COVID-19 Twitter conversation polarity and reported cases, but a weak correlation between polarity and reported deaths. The second theoretical framework presents and combines three novel approaches in Energy domain elaborating in mining tasks related with Energy Balancing (EB), Energy Load Forecasting (ELF) and Energy Optimal Day-Ahead Scheduling (EODS). Energy management may be improved by performing EB in both Peer-to-Peer (P2P) and Virtual Microgrid-to-Virtual Microgrid (VMG2VMG) level. This task yields an interdisciplinary analytics-based approach for the formation of VMGs achieving EB. Computer Science methods are incorporated for addressing an Energy sector problem, utilizing data preprocessing techniques and Machine Learning concepts. Each prosumer is perceived as a peer, while VMGs are perceived as clusters of peers. This approach incorporates clustering and binning algorithms for preprocessing Energy data (for 94 prosumers) producing options for generating VMGs. Then, a customized Exhaustive brute-force Balancing Algorithm (EBA) balances at the cluster-to-cluster level (VMG2VMG balancing) reporting outcomes and prospects for scaling up and expanding this work. A novel approach in the task of ELF exposes improvements for residential house energy requirements. This task is crucial for Energy sector stakeholders (e.g., DSO, aggregators etc.) since they are able to plan in more efficient manner their Demand Response (DR) management strategies. The experimentation includes the retrieval of energy readings from a state-of-the-art nearly Zero Energy Building (nZEB). Focus is made on one step ahead ELF, producing an approach regardless the time resolution of available data while yielding high accuracy results. Ensemble methods and forecasting algorithms are utilized while the evaluation of forecasting results is performed with popular accuracy metrics (MAPE, SMAPE and RMSE) and an Execution Time (ET) metric. Optimal energy management relates with the task of EODS. A novel approach is proposed in the form of a framework/tool for a multi-objective analysis comprising a decision-making system. Two distinct optimization problems for two actors (consumers and aggregators) are considered, with each solution completely or partly interacting with the other in the form of DR signal exchange. The overall optimization “Interdisciplinary data science methods using machine learning for enhanced knowledge acquisition” v is formulated by a bi-objective optimization problem for the consumer's side aiming at cost minimization and discomfort reduction; and a single objective optimization problem for the aggregator's side aiming also at cost minimization. Experimentation is conducted on a real pilot (Terni Distribution System portfolio). The framework performs decision making by forecasting the day-ahead energy management requirements while aiming at optimal management of energy resources considering both aggregator's and consumer's preferences and goals. Achievements of this thesis highlight prospects for enhanced knowledge acquisition through the conception of two theoretical frameworks in the domains of Social Media and Energy while envisioning an interdisciplinary research design. The theoretical frameworks, “A Multi-Functional Framework for defining Social Media Types, extracting Topics and Inferences, and discovering Correlations based on Public Sentiment” and “A Novel Framework for P2P and VMG2VMG Energy Balancing, Incorporating One Step Ahead Load Forecasting and Optimization for Day-Ahead Energy Scheduling” incorporate common data mining methods/algorithms elevating the necessity for interdisciplinary novel approaches in multi-domain data analytics along with benefits they might yield.