In this work we propose a novel predictive model for improving investment capability that uses structured and unstructured data to predict stock price movements. We adopt deep learning techniques that have already been used successfully for natural language processing tasks, along with traditional data retrieval, to analyze and predict trends in the Taiwan stock market, and conduct experiments on both structured and unstructured data. Machine learning and data preprocessing techniques such as word2vec are used to train prediction models. Our experiments show that using deep learning on structured data yields improved accuracy, which attests the suitability of deep learning for structured data, especially for long short-term memory (LSTM) models. Finally, we combined structured and unstructured data using a combined approach to achieve improved accuracy with lower investment risks. The models in this work are thus suitable for real-world applications, including day trading strategy planning as well as long or short transaction strategy planning.