Project Plan for Final Year Project Deep Learning for Text Classification in Azure Infrastructure

Increasing computing power and exploding volumes of data have driven us into the era of machine learning in the past few decades. With the aid of deep neural networks, researchers have equipped computers with eyes that can identify different objects [1], brains that can diagnose diseases [2], and hands that can create astonishing works of art [3]. Besides these remarkable achievements, natural language processing (NLP for short) is also a promising field for deep learning to explore its potential. Traditionally, researchers need to summarize features of natural languages themselves based on their linguistic knowledge and then teach computers how to understand human languages [4]. Nowadays with emerging technologies of deep learning, much research has been done to let computers learn by themselves to recognize human speeches [5], translate between languages [6], and even speak in human languages [7]. Computers’ better understanding of natural languages has also enabled numerous useful real-world applications such as Siri and Google Translation. However powerful NLP is with the help of deep learning, the majority of current studies focus on English processing. There are few research studies and applications of NLP with Chinese. In addition to the fact that English is the major language used in academia, Chinese language processing faces several other challenges such as word segmentation [8]. With great potential for real-world applications as well as research directions, it is desirable to adapt existing deep NLP models originally designed for English to Chinese processing and evaluate their performance, and also to design and develop new models. Having this motivation, we will cooperate with Microsoft to develop a system which leverages the approach of deep learning for Chinese sentiment analysis in Azure infrastructures. The significance of this project is three-fold. First, we will explore different options and possibilities for Chinese NLP and identify their advantages and drawbacks. Second, sentiment analysis for domain-specific Chinese corpora may yield significant value in that domain. For example, analyzing sentiments in customers’ feedbacks may guide companies to improve their products or services. Last, we will get our hands on state-of-the-art knowledge and technologies about deep learning and NLP, and acquire valuable experiences of developing deep learning NLP systems.