Crowd Sourcing as an Improvement of N-Grams Text Document Classification Algorithm

摘要

A common task in a world of natural language processing is text classification useful for e.g. spam filters, documents sorting, science articles classification or plagiarism detection. This can still be done best and most accurately by human, on the other hand, we can of ten accept certain error in the classification in exchange for its speed. Here, natural language processing mechanism transforms the text in natural language to a form understandable by a classifier such as K-Nearest Neighbour, Decision Trees, Artificial Neural Network or Support Vector Machines. We can also use this human element to help automated classification to improve its accuracy by means of crowdsourcing. This work deals with classification of text documents and its improvement through crowdsourcing. Its goal is to design and implement text documents classifier prototype based on documents similarity and to design evaluation and crowdsourcing-based classification improvement mechanism. For classification the N-grams algorithm has been chosen, which was implemented in Java. Interface for crowdsourcing was created using CMS WordPress. In addition to data collection, the purpose of interface is to evaluate classification accuracy, which leads to extension of classifier test data set, thus the classification is more successful. We have tested our approach on two data sets with promising preliminary results even across different languages. This led to a real-world implementation started at the beginning of 2019 in cooperation of two universities: VšB-TUO and OSU.