Abstract:
As technology advances, more people are being persuaded to use the internet to
acquire information. On the internet, people may find a wide range of documents,
including scholarly papers, academic books, reports, research articles, etc.
However, in general, web papers are not logically organized, which makes it
difficult and time-consuming to get pertinent information from a website.
Therefore, a particular study was accomplished to classify the documents based
on formal and informal writing styles considering their characteristics. There are
linguistic variations that are specific to each style that may be used to determine
if a document is formal or informal. Before creating this model perceived the
characteristics of the informal and formal styles. In this study, we focused on 15
characteristics, and currently, 6 characteristics are considered namely,
Colloquialism, Abbreviation, Contraction, Voice, Modal Verbs, and Phrasal
Verbs. Used 5000 data sets for this experiment as formal news articles, informal
letters and personal blogs. Pre-processed them using four steps, such as
tokenization, stop word removal, lowercasing, and lemmatization, and used four
feature extractions methods: Tf-Idf, Word2Vec, Doc2Vec, and Glove. For
contraction create an algorithm to find out how many contractions are included
using seven rules. Modal verbs and phrasal verbs are also counted in every
document and identify the passive voice and active voice separately. And also
identifies the abbreviation and colloquialism by classifying the documents using
two target variables. For the classification process Artificial Neural Network
(ANN) and Long Short-Term Memory (LSTM) algorithms. Considering the
abbreviation characteristic doc2vec showed highest accuracy in the ANN
algorithm and for the colloquialism characteristic doc2vec showed the highest
accuracy in the LSTM algorithm. In this approach, six features have been
completed, while the remaining nine are being worked on. Finally, planning to
classify documents as formal or informal, combines all of the findings.
