What’s in a Gist? Towards an Unsupervised Gist Representation for Few-Shot Large Document Classification

The gist can be viewed as an abstract concept that represents only the quintessential meaning derived from a single or multiple sources of information. We live in an age where vast quantities of information are widely available and easily accessible. Identifying the gist contextualises information which facilitates the fast disambiguation and prediction of related concepts bringing about a set of natural relationships defined between information sources. In this paper, we investigate and introduce a novel unsupervised gist extraction and quantification framework that represents a computational form of the gist based on notions from fuzzy trace theory. To evaluate our purposed framework, we apply the gist to the task of semantic similarity, specifically to few-shot large document classification where documents on average have a large number of words. The results show our proposed gist representation can effectively capture the essential information from a text document while dramatically reducing the features used.