Expressive modeling for trusted big data analytics: techniques and applications in sentiment analysis
Erik Tromp, Mykola Pechenizkiy, Mohamed Medhat Gaber · Big Data Analytics · 2017
Sentiment analysis becomes ubiquitous for a variety of applications used in marketing, commerce, and public sector. This has been raising a natural interest within the academic research and industry to develop approaches and solutions for ubiquitous sentiment analysis. However, we can observe that most of the academic research focuses on adopting state-of-the-art machine learning techniques for sentiment classification and elements of natural language processing for feature construction and evaluate them on benchmark datasets not regarding much the actual application settings. In industry the focus is on developing platforms, services and customized solutions for certain applications and for different domains. In this work we propose a generic framework for ubiquitous sentiment classification. We discuss the Rule-Based Emission Model (RBEM) algorithm that we employ for polarity detection. We show with the experimental results on benchmark datasets and real case studies that the proposed framework and RBEM approach for polarity detection are indeed generic and extendable. As the state-of-the-art machine learning techniques produce black-box models for sentiment analysis, they are hard to fine-tune, debug and adapt to new domains. The necessity to collect lots of labeled data from a particular domain is another obstacle. Therefore, in industry rather simplistic approaches are adopted resulting potentially in poor accuracy. The proposed framework for sentiment analysis allows to develop different solutions that are scalable, transparent, and easy to maintain.