Some clues on irony detection in tweets

Microposts on Twitter allow users to express ideas and opinions dynamically, although in a very limited space. The high volume of data available provides relevant clues about the judgment about certain products, events, services etc. While in sentiment analysis the common task is to classify the utterances according to their polarity, the detection of ironic statements represent a big challenge for this task. In this study, we analyze and implement some patterns that may be associated to ironic statements in Brazilian Portuguese. A common ground between the author of the tweets and their audience is required in order to establish some background information on the text; thus, contextual features, such as the specificity of a domain, the period of time, the textual support and genre (Twitter and tweets, for example), are considered. Irony may be seen as a complex mechanism of communication that is governed by pragmatic principles, and it is often confused with sarcasm, satire or parody. In this study, we will base the task of capturing irony on a general concept for this phenomenon, since there are no consensus opinions on a rigid definition. We base the implementation of the patterns in the works of [1], [3], [2], [4]. We follow a pragmatic view, in which the context is organized as the reader is in touch with the text. This is based on information the reader knows about the textual genre (tweet format, time, domain and type of content) and on what is written. The elaboration of the patterns that would involve possible evidences of ironic messages considers the following: syntactic rules, POS tagging, some static expression, list of laugh-