A New Text Representation Technique-Based Approach for Authorship Verification

Author verification predicts whether the given text is written by the suspected author or not. Researchers used different types of stylistic features to differentiate the writing style of the author. PAN is an organization conducting competitions on different tasks every year by providing a suitable dataset. The author verification task has been included in several years of PAN competition. In this article, we conducted an experiment on the PAN competition 2022 author verification task. In this task, we need to predict whether the given two texts that belong to two different discourse types are written by the same author or different authors. The dataset contains an English corpus, which contains the pairs of texts written using four discourse types, such as business memos, text messages, emails and essays. We proposed a new text representation technique for the task of authorship verification. In this approach, we used word embedding techniques for representing words as vectors. The word embedding techniques consider the importance of a word in the total dataset of documents to generate the word vectors. The TFIDF measure is used to identify the importance of a word within a document. Each word is represented by combining the TFIDF weight of word and word embedding vector. The documents are represented as vectors by aggregating the word vectors that are contained in the document. These document vectors are trained with machine learning algorithms for predicting the accuracy of the proposed approach for authorship verification. We identified that the proposed text representation technique attained the best accuracies when compared with various solutions for authorship verification.