Analysis of Turkish audio recording data labeled with three emotions popular machine learning algorithms

Abdülkadir Tepecik; Engin Demir

Makale detayları

Analysis of Turkish audio recording data labeled with three emotions popular machine learning algorithms

Yazarlar:

Abdülkadir Tepecik | Engin Demir

Yayıncı:

Gazi University

Ana Başlıklar:

Mühendislik

Dergi:

Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi Vol.39 Issue 2, pp.709-716

Belge Tipi:

Makale

Yayın Tarihi:

11/2023

Belge Dili:

Türkçe

Document ID (Org No.):

2001301438

pISSN:

1300-1884

Anahtar kelimeler:

Özet:

 Emotions are physical changes in a persons mood resulting from his interaction with internal and environmental influences. Individuals can convey their feelings to other individuals by means of voice communication as well as body language. Especially in situations and times when body language is insufficient, voice communication becomes important for individuals. In our study, the emotion labels of the data were firstly determined by the Python programming language on the data set containing the Turkish voice recordings, and then the analyzes were carried out with the five most used machine learning algorithms in the literature studies. Analyses were conducted through both Rapid Miner and the Pyhton programming language. In the study, both CountVectorizer and TF-IDF vectorization methods were used in analyses performed through the Pyhton programming language, and TF-IDF vectorization method was used in analyses performed with Rapid Miner. As a result, the best accuracy rate in the Pyhton programming language was achieved by the Naive Bayes machine learning algorithm CountVectorizer method with 70. At Rapid Miner, the Support Vector Machine machine learning algorithm achieved the best accuracy rate of 69.60.with our study, a new Turkish emotion dataset has emerged. Our study is also an original study in which emotion detection was done with the BERT model of the data obtained from Turkish voice recordings.

 Emotions are physical changes in a persons mood resulting from his interaction with internal and environmental influences. Individuals can convey their feelings to other individuals by means of voice communication as well as body language. Especially in situations and times when body language is insufficient, voice communication becomes important for individuals. In our study, the emotion labels of the data were firstly determined by the Python programming language on the data set containing the Turkish voice recordings, and then the analyzes were carried out with the five most used machine learning algorithms in the literature studies. Analyses were conducted through both Rapid Miner and the Pyhton programming language. In the study, both CountVectorizer and TF-IDF vectorization methods were used in analyses performed through the Pyhton programming language, and TF-IDF vectorization method was used in analyses performed with Rapid Miner. As a result, the best accuracy rate in the Pyhton programming language was achieved by the Naive Bayes machine learning algorithm CountVectorizer method with 70. At Rapid Miner, the Support Vector Machine machine learning algorithm achieved the best accuracy rate of 69.60.with our study, a new Turkish emotion dataset has emerged. Our study is also an original study in which emotion detection was done with the BERT model of the data obtained from Turkish voice recordings.

 Emotions are physical changes in a persons mood resulting from his interaction with internal and environmental influences. Individuals can convey their feelings to other individuals by means of voice communication as well as body language. Especially in situations and times when body language is insufficient, voice communication becomes important for individuals. In our study, the emotion labels of the data were firstly determined by the Python programming language on the data set containing the Turkish voice recordings, and then the analyzes were carried out with the five most used machine learning algorithms in the literature studies. Analyses were conducted through both Rapid Miner and the Pyhton programming language. In the study, both CountVectorizer and TF-IDF vectorization methods were used in analyses performed through the Pyhton programming language, and TF-IDF vectorization method was used in analyses performed with Rapid Miner. As a result, the best accuracy rate in the Pyhton programming language was achieved by the Naive Bayes machine learning algorithm CountVectorizer method with 70. At Rapid Miner, the Support Vector Machine machine learning algorithm achieved the best accuracy rate of 69.60.with our study, a new Turkish emotion dataset has emerged. Our study is also an original study in which emotion detection was done with the BERT model of the data obtained from Turkish voice recordings.

 Duygu, insanın ruh halinde içsel ve çevresindeki etkilerle etkileşiminden doğan fiziksel değişimlerdir. Bireyler duygularını, beden dilinin yanı sıra sesli iletişim vasıtalarıyla da diğer bireylere aktarabilirler. Özellikle beden dilinin yetersiz olduğu durum ve zamanlarda bireyler için sesli iletişim önem kazanmaktadır. Çalışmamızda da Türkçe ses kayıtlarını içeren veri seti üzerinde Pyhton programlama dili aracılığıyla öncelikle verilerin duygu etiketlerinin tespiti yapılmış olup, sonrasında literatür çalışmalarında en çok kullanılan beş makine öğrenim algoritmasıyla analizler gerçekleştirilmiştir. Analizler hem Rapid Miner hem de Pyhton programlama dili aracılığıyla gerçekleştirilmiştir. Çalışmada, Pyhton programlama dili aracılığıyla yapılan analizlerde hem CountVectorizer hem de TF-IDF vektörizasyon yöntemleri, Rapid Miner ile yapılan analizlerde TF-IDF vektörizasyon yöntemi kullanılmıştır. Sonuç kısmında ise Pyhton programlama dilinde en iyi doğruluk oranını 70 oranla Naive Bayes makine öğrenim algoritması CountVectorizer yöntemiyle elde etmiştir. Rapid Minerda ise en iyi doğruluk oranını 69,60 oranla Support Vector Machine makine öğrenim algoritması elde etmiştir. Çalışmamızla beraber ortaya yeni bir Türkçe duygu veri seti çıkmıştır. Çalışmamız ayrıca Türkçe ses kayıtlarından elde edilen verilerin BERT modeli ile duygu tespiti yapılan özgün bir çalışmadır.

Makale detayları

Analysis of Turkish audio recording data labeled with three emotions popular machine learning algorithms

Yazarlar:

Abdülkadir Tepecik | Engin Demir

Yayıncı:

Gazi University

Ana Başlıklar:

Mühendislik

Dergi:

Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi Vol.39 Issue 2, pp.709-716

Belge Tipi:

Makale

Yayın Tarihi:

11/2023

Belge Dili:

Türkçe

Document ID (Org No.):

2001301438

pISSN:

1300-1884

Sentiment Analysis

Turkish Sentiment Analysis

Machine Learning

Turkish Auido Recordings

Rapid Miner

Python.Duygu Analizi

Makine Öğrenmesi

Türkçe Ses Kayıtları

Türkçe Duygu Analizi

Rapid Miner

Pyhton

Sayı

Makale detayları

Yazarlar: Abdülkadir Tepecik | Engin Demir

Yayıncı: Gazi University

Ana Başlıklar: Mühendislik

Dergi: Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi Vol.39 Issue 2, pp.709-716

Belge Tipi: Makale

Yayın Tarihi: 11/2023

Belge Dili: Türkçe

Document ID (Org No.): 2001301438

pISSN: 1300-1884

Sayı

Yazarlar:

Abdülkadir Tepecik | Engin Demir

Yayıncı:

Gazi University

Ana Başlıklar:

Mühendislik

Dergi:

Gazi Üniversitesi Mühendislik Mimarlık Fakültesi Dergisi Vol.39 Issue 2, pp.709-716

Belge Tipi:

Makale

Yayın Tarihi:

11/2023

Belge Dili:

Türkçe

Document ID (Org No.):

2001301438

pISSN:

1300-1884