Convolutional Neural Networks for text classification

Author: Shalva Jashiashvili
Keywords: neural network, classification, convolution, text
Annotation:

Modern technology enables information to spread so quickly that it becomes necessary to invent new methods of sorting and processing, thus making it easier to analyze the information and put it to use. One of the most widely used forms of information storage and distribution is textual format, which makes text classification problem rather acute. A specific example of text classification is spam filtering. Apart from this, classification of the texts may have other uses, for example: determining product quality according to comments in the online stores. Text classification is different from other types of classification of information, since each word may have a different meaning in different contexts, which complicates the task. The goal of this thesis is to create a system which allows users to create a desired model. The model that can determine the sentiments of textual information (rate is positive or negative). To construct this model a database of films' comments was used. The model can estimate with 97% accuracy whether a comment which rates a film is positive or negative. To solve this problem a specific type of neural network is used, a convolutional neural network. This type of neural network was mainly used to classify multimedia type of information. However, resent research shows that the convolution can be successfully used to classify textual data. Convolution allows us to take into account the context in which the word is used, thus solving the main problem of text classification, loss of context, and improving obtained results.

Lecture files:

პრეზენტაცია [ka]
კონვოლუციური ნეირონული ქსელების გამოყენება ტექსტების კლასიფიკაციისთვის [ka]