Sarcasm as a Contradiction Between a Tweet and its Temporal Facts : A Pattern Based Approach

In the context of Indian languages, sarcasm detection in Hindi is a tedious job as it is rich in morphology and complex in structure. The annotated resources for sarcastic Hindi sentences are almost negligible for machine learning analysis. Here, we propose a pattern-based framework for sarcasm detection in Hindi tweets. It has been observed that a tweet is sarcastic if it contradicts its temporal facts intentionally. The temporal fact is a collection of time-dependent facts which may change over the period. We used Hindi news with timestamp as a corpus of temporal facts. The timestamp describes the fact period of any entity. In this research, a temporal fact is represented as a <key, value> pair. To form a <key, value> pair, one need to extract triplets i.e. subject, verb and object for every sentence. Next, a key is formed using subject and verb together. The value is formed using object and timestamp together. To predict the sarcastic tweet; one needs to extract the triplets from input tweet and form a <key, value> pair. Now, the <key, value> pair of the input tweet is mapped with related <key, value> pair in the corpus of temporal facts and are checked if they coincide. If they contradict, the input tweet is considered as sarcastic. The achieved accuracy of the proposed approach outperforms the state-of-the-arts techniques for Hindi sarcasm detection as it attains an accuracy of 82.8%.