IOWA & cross-ratio uninorm operators as aggregation tools in sentiment analysis and ensemble methods

Abstract

In the field of Sentiment Analysis, a number of different classifiers are utilised to attempt to establish the polarity of a given sentence. As such, there could be a need for aggregating the outputs of the algorithms involved in the classification effort. If the output of every classification algorithm resembles the opinion of an expert in the subject at hand, we are then in the presence of a group decision-making problem, which in turn translates into two sub-problems: (a) defining the desired semantic of the aggregation of all opinions, and (b) applying the proper aggregation technique that can achieve the desired semantic chosen in (a). The objective of this article is twofold. Firstly, we present two specific aggregation semantics, namely fuzzy-majority and compensatory, which are based on Induced Ordered Weighted Averaging and Uninorm operators, respectively. Secondly, we show the power of these two techniques by applying them to an existing hybrid method for classification of sentiments at the sentence level. In this case, the proposed aggregation solutions act as a complement in order to improve the performance of the aforementioned hybrid method. In more general terms, the proposed solutions could be used in the creation of semantic-sensitive ensemble methods, instead of the more simple ensemble choices available today in commercial machine learning software offerings.