Is it a good idea to have a spam filter rule that rejects messages containing more than one exclamation mark in the subject line? This paper considers the authorial sentiments ascribed to communications containing one or more exclamation marks, and the consequences of those sentiments for automated methods of determining frame of mind from written text. Results reported by the paper suggest that exclamation mark frequency of occurrence proportionally amplifies a reader’s subjective impression of an author’s sentiments.

Sentiment analysis involves text, punctuation, and other forms of (usually folk) orthography. This paper hypothesizes the significance of exclamation marks in accurate sentiment determination, and reports on experiments set up to test the hypothesis. The experiments compared the sentiment analysis of six online tools to the sentiment analysis made by humans. The tools analyzed a set of approximately 1,000 website comments in various areas, and the humans analyzed a 30-comment subset. Experiment results showed that exclamation mark occurrence had almost no effect on automatic sentiment analysis, while it caused strong and significant bias in human sentiment judgments.

Alas, this paper doesn’t directly answer the question posed at the start of this review, but it does suggest amplifying the results of spam filter, and sentiment, analysis in proportion to the number of exclamation marks used. The paper is a short and straightforward read, although apparently only half of the experiments conducted are described. Reference is made to other work providing more details.