Emotion Analysis From Turkish Tweets Using Deep Neural Networks

Abstract

Text data analysis of social media is becoming more and more important since it includes the most recent information on what people think about. Likewise, emotion is one of the most valuable parts of human communication, emotion analysis is a type of information extraction process which identifies the emotional states of a given text. In this study, we investigated the performance of deep neural networks on emotion analysis from Turkish tweets. For this, we examined three different deep learning architectures including artificial neural network (ANN), convolutional neural network (CNN) and recurrent neural network (RNN) with long short-term memory (LSTM). Besides, we curated a dataset of Turkish tweets and annotated each tweet automatically for six emotion categories using a lexicon-based approach. For the evaluation, we conducted a set of experiments for each architecture. The results showed that the lexiconbased automatic annotation of tweets is valid. Secondly, ANN produced the worst result as expected, and CNN resulted in the highest score of 0.74 in terms of accuracy measure. Experiments also showed that our proposed approach for emotion analysis of tweets in Turkish performs better than state-of-the-art in this topic.

Emotional Analysis using Twitter Data during Pandemic Situation: COVID-19

“…Wang et al [4]have performed an emot ional analysis method on COVID-19 data collected fro m Sina Weibo ,a microblogging site in China using Support Vector Machine, naïve Bayes and Random Forest classifiers. Mansur ALP et al [2] classified Turkish Tweets into basic emotions using Deep Neural Network. In our proposed method analysis is based on the data collected fro m twitter data across the world.…”

“…[1] Our method aims at fetching the t weets related to COVID-19 and performing emot ional analysis of each tweet posted by the twitter users. It aims at classifying the tweets into positive and negative tweets and further classified into six basic emotions given by Ekman i.e joy, sadness, anger, fear, d isgust and surprise [2]. Twitter data is chosen for emotional analysis of people as tweets contain a large number of opin ionated text.…”

Deep Emotion Detection Sentiment Analysis of Persian Literary Text

“…Authors in their study reviewed and compared machine learning and deep learning approaches in analyzing emotions of Persian sentences. Simple Bayes and random reduction gradient and support vector machine are used as machine learning algorithms and two-way short-term memory and torsional neural network as deep learning models [38]. Rohani et al introduced an algorithm based on an unsupervised lexical method to identify six emotional states in Persian texts on social websites [31].…”

Türkçe bilgisayarlı dil bilimi çalışmalarında his analizi

“…Ekman his kategorilerinde yapılan his analizi sonucunda yaklaşık %91 civarında başarı sağlanmıştır. Araştırmacılar (Tocoglu, Ozturkmenoglu ve Alpkocak, 2019) his kategorilerine ait anahtar kelimeleri içeren Twitter gönderilerini toplayarak oluşturdukları his veri kümesi üzerinde his analizi gerçekleştirmiş ve %73 başarı elde etmişlerdir. Görüleceği üzere, Türkçe his analizi alanında oldukça az çalışma yapılmış olup çalışmaların genelde metin sınıflama çalışmalarına öykünerek yapıldığı ve sözlüklü yöntemlerin de denendiği görülmüştür.…”