Distinguishing between facts and opinions for sentiment analysis: Survey and challenges

Sentiment analysis requires a lot of information coming from different sources and about different topics to be retrieved and fused. For this reason, one of the most important subtasks of sentiment analysis is subjectivity detection, i.e., the removal of 'factual' or 'neutral' comments that lack sentiment. It is possibly the most essential subtask of sentiment analysis as sentiment classifiers are often optimized to categorize text as either negative or positive and, hence, forcefully fit unopinionated sentences into one of these two categories. This article reviews hand-crafted and automatic models for subjectivity detection in the literature. It highlights the key assumptions these models make, the results they obtain, and the issues that still need to be explored to further our understanding of subjective sentences. Lastly, the advantages and limitations of each approach are compared. The methods can be broadly categorized as hand-crafted, automatic, and multi-modal. Hand-crafted templates work well on strong sentiments, however they are unable to identify weakly subjective sentences. Automatic methods such as deep learning provide a meta-level feature representation that generalizes well on new domains and languages. Multi-modal methods can combine the abundant audio and video forms of social data with text using multiple kernels. We conclude that the high-dimensionality of n-gram features and temporal nature of sentiments in long product reviews are the major challenges in sentiment mining from text.