An Optimal Feature Set for Stylometry-based Style Change detection at Document and Sentence Level

Writing style change detection models focus on determining the number of authors documents with or without known authors. Determining exact contributing in writing a document particularly when contribute short texts form sentence is still challenging because lack standardized feature sets able to discriminate between works Therefore, task identifying best set for all tasks considered important. This paper sought determine tasks; separating several changes (multi-authorship) from any (single-authorship), and location case multi-authorship. We performed exploratory research existing stylometric features level features. Document were extracted used separate single authored multi-authored documents, while answer question To this question, we trained random forest classifier rank separately, applied an ablation test top 15 using k-means clustering algorithm confirm effect these model performance. The study found out that was provided by ensemble including repetitions (num_sentence_repetitions) as most determinant feature, 5-grams, 4-grams, Special_character, sentence_begin_lower, sentence_begin_upper, diversity, automated_readability_index, parenthesis_count, first_word_uppercase, lensear_write_formula, dale_chall_readability, difficult_words, type_token_ratio. These ranked experiment one. On other hand, fifteen based ranks dale_chall_readability grade, check_available_vowel, flesch_kincaid colon_count, verbs, bigrams, alphabets, personal pronouns, coordinating conjunctions, interjections, modals, type_token ratio punctuations_count. Consequently, optimal results features, check_available_vowels, punctuations_counts, parenthesis count, conjunctions colon count.