Identifying Humor in Reviews using Background Text Sources

Abstract

We study the problem of automatically identifying humorous text from a new kind of text data, i.e., online reviews. We propose a generative language model, based on the theory of incongruity, to model humorous text, which allows us to leverage background text sources, such as Wikipedia entry descriptions, and enables construction of multiple features for identifying humorous reviews. Evaluation of these features using supervised learning for classifying reviews into humorous and non-humorous reviews shows that the features constructed based on the proposed generative model are much more effective than the major features proposed in the existing literature, allowing us to achieve almost 86% accuracy. These humorous review predictions can also supply good indicators for identifying helpful reviews.

Uncertainty and Surprisal Jointly Deliver the Punchline: Exploiting Incongruity-Based Features for Humor Recognition

“…Shahaf et al (2015) investigated funny captions for cartoons and proposed several features including perplexity to distinguish between funny and less funny captions. Morales and Zhai (2017) proposed a probabilistic model and leveraged background text sources (such as Wikipedia) to identify humorous Yelp reviews. Liu et al (2018) proposed to model sentiment association between elementary discourse units and designed features based on discourse relations.…”

Gulu at SemEval-2021 Task 7: Detecting and Rating Humor and Offense

“…Yang et al (2015) first determined the semantic structure behind each structure of the humor and design feature set, and then used a calculation method to identify the humor and their humor recognizer was very effective in automatic distinction humorous and non-humorous text. Morales and Zhai (2017) proposed a generative language model based on the inconsistency theory to model humorous text, so that they can use background text sources such as Wikipedia item descriptions, and can build multiple functions for recognizing humorous comments. Using supervised learning to classify reviews into humorous reviews and non-humorous reviews, these functions showed that the features constructed based on the proposed generative model were more effective than the main features proposed in the existing literature.…”

Computational Modeling of Conversational Humor in Psychotherapy

“…Our first set of features includes simple counts such as number of words, average word length and percentage of uppercase and lowercase characters as suggested by [28]. [17] reported successes in using stylistic features such as number of rhymes or alliterations in the utterance.…”

“…Several works report the use of ambiguity in predicting humor from text [28]. This can be attributed to the theory of subverting expectations which is frequently associated with humor [16,15].…”

“…This can be attributed to the theory of subverting expectations which is frequently associated with humor [16,15]. To measure ambiguity in an utterance, we used Wordnet 2 similar to [28] to get the average number of synsets associated with each word, since higher this number, higher the apparent ambiguity in its meaning.…”