Text clustering from author attributes

Text clustering is applicable in a wide range of tasks such as topic detection and semantic extraction of documents. For this purpose, diverse models and features has been used in literature, where TF-IDF and Vector space model are of high sparsity and sentimental, structural or syntactic features are not be taken into account in recent researches. In this study, it is proposed a novel multi-objective clustering method, which incorporates feature selection and clustering process. Features are not selected, but it is retained graph interconnections based on integrated feature values with higher degree. Utilized features are distinguished in five categories-lexicological, syntactic, sentimental, structural and semantic-to be formed five vectors of each feature category values thus it is not mandatory to belong to a specific feature category like TF-IDF and Vector space models. Textual data based on financial reports labeled by their author. Multi-objective clustering method is examined for author clustering task and as pre-processing step it is utilized k-partite graph to be represented the degree of separateness between texts based on their textual features. It is found that both in case where 100 random and 500 random financial cases are used, multi-objective clustering method is high accurate in author clustering task.