Metadata Generation for Multi-Text Classification in Structured Data

dIn today's information-saturated world, text analysis has become an indispensable resource to extract useful data from massive amounts of texts. A large portion of this information is unstructured. Hence, it has created a need for methodologies -Named Entity Recognition (NER), Part-of-Speech (PoS) Tagging, N-grams, Term Frequency - Inverse Document Frequency (TF-IDF)- which can read and understand information based on their meaning, context and linguistic cohesion. However, these approaches on their own fall short if applied in already structured data. The idea of generating metadata which can simultaneously provide situational information from structured text data is proposed in this paper. The abstraction of text as a "group of concepts" can boost the relevance of a word in a collection of documents, which allows a more refined separation of classes and a better performance in multi-text classification tasks.