Automatic Prediction of Aesthetics and Interestingness of Text Passages

Proceedings of COLING 2014, the 25th International Conference on Computational Linguistics: Technical Papers, pages 905–916, Dublin, Ireland, August 23-29 2014. Automatic Prediction of Text Aesthetics and Interestingness Debasis Ganguly CNGL, School of Computing, Dublin City University, Dublin 9, Ireland dganguly@computing.dcu.ie Johannes Leveling CNGL, School of Computing, Dublin City University, Dublin 9, Ireland jleveling@computing.dcu.ie Gareth J.F. Jones CNGL, School of Computing, Dublin City University, Dublin 9, Ireland gjones@computing.dcu.ie Abstract This paper investigates the problem of automated text aesthetics prediction. The avail- ability of user generated content and ratings, e.g. Flickr, has induced research in aesthet- ics prediction for non-text domains, particularly for photographic images. This problem, however, has yet not been explored for the text domain. Due to the very subjective nature of text aesthetics, it is difficult to compile human annotated data by methods such as crowd sourcing with a fair degree of inter-annotator agreement. The availability of the Kindle “popular highlights” data has motivated us to compile a dataset com- prised of human annotated aesthetically pleasing and interesting text passages. We then undertake a supervised classification approach to predict text aesthetics by constructing real-valued feature vectors from each text passage. In particular, the features that we use for this classification task are word length, repetitions, polarity, part-of-speech, semantic distances; and topic generality and diversity. A traditional binary classification approach is not effective in this case because non-highlighted passages surrounding the highlighted ones do not necessarily represent the other extreme of unpleasant quality text. Due to the absence of real negative class samples, we employ the MC algorithm, in which training can be initiated with instances only from the positive class. On each successive iteration the algorithm selects new strong negative samples from the unlabeled class and retrains itself. The results show that the mapping convergence (MC) algorithm with a Gaussian and a linear kernel used for the mapping and convergence phases, respectively, yields the best results, achieving satisfactory accuracy, precision and recall values of about 74%, 42% and 54% respectively. 1 Introduction Since their inception, Amazon Kindle device1 and Apps for other general purpose hand-held devices, have led to a massive increase in the trend of reading e-books over paper printed ones. The Amazon Kindle and the Kindle Apps provide a very simple mechanism for highlighting a piece of text and sharing it on social media. The most popular highlighted pieces of text are shown in the Kindle device with an intention to help readers focus on passages that are pleasing or interesting to the greatest number of people. Every month, Kindle customers highlight millions of book passages that are meaningful to them2. The general trend among Kindle readers, while reading the classic English literary works, is to highlight text passages that are associated with a high aesthetic quality. An example highlighted passage is shown in Figure 1. With the availability of such highlighted text, which may be considered as text passages which most readers find pleasing to read, an interesting research problem is to attempt automatic prediction of …

Content truncated to 500 words