Explorations in Identifying and Summarizing Subjective Content in Text

The extraction and summarization of opinions in text is a useful task that lends itself naturally to connectionist models of sequential data analysis, such as RNNs and LSTMs. Recent work by Irsoy et al. [11] has shown that RNNs (particularly bidirectional deep RNNs) can be quite successfully applied to the task of identifying opinionated phrases in text. Further, Wang et al. [19] have worked on the problem of generating abstractive summaries of opinionated text using an attention-based model and an LSTM neural network. In this work, we explore both these tasks by training various models to perform them, thereby advancing our understanding of the implementation, advantages and challenges of the current approaches to identifying and summarizing opinionated text.