Text analysis with deep learning and big data approaches

Big data analytics and deep learning are two significant areas of research and study that data science has focused on in the developing digital world over the last few years. Analyzing and managing large amounts of text data using a variety of traditional software tools and technologies is a difficult problem. In this thesis, big data technologies and deep learning architectures have been analyzed in detail and four basic applications have been proposed as academic contributions. First, a cloud based distributed performance analysis and evaluation system was developed for call centers. The proposed system aims to provide significant contribution in terms of customer satisfaction, sales and marketing, high quality of service and performance management by offering a cloud based performance measurement system that handles both internal and external call records in a distributed manner. Second, a distributed readability analysis system for the Turkish language was developed using big data technologies. There is no readability application used by educational institutions in Turkey and due to this need, a readability system has been developed to analyze Turkish reading books in a short time. Third, using various deep learning models which are created with different architectures, methods, layers and hyper parameters sentiment analysis and multi-category text classification on news datasets studies are performed. Lastly, a novel Average Document Embeddings (ADE) approach is presented which can be used for multi-category language independent text classification. The proposed method has been tested for sentiment classification in Turkish and English movie reviews and has performed well. There is no large scale benchmark dataset that can be used in Turkish text classification studies. The other main contribution of this thesis is that the Turkish news data set containing about 1 million unique words to meet this need and the creation of the 150,000 labeled Turkish movie reviews dataset, which is made available for academic use.