IKBAL, MUH (2024) ANALISIS SENTIMEN BERBASIS PENGENALAN UCAPAN PADA VIDEO REVIEW PRODUK MENGGUNAKAN METODE BIDIRECTIONAL ENCODER REPRESNTATIONS FROM TRANSFORMERS (BERT) = Sentiment Analysis Based on Speech Recognition in Product Review Videos Using Bidirectional Encoder Representations from Transformers (BERT) Method. Skripsi thesis, Universitas Hasanuddin.
![[thumbnail of Cover]](/46224/1.hassmallThumbnailVersion/D121171317_skripsi_06-03-2024%20cover1.png)

D121171317_skripsi_06-03-2024 cover1.png
Download (174kB) | Preview
![[thumbnail of Bab 1-2]](/style/images/fileicons/text.png)
D121171317_skripsi_06-03-2024 1-2(FILEminimizer).pdf
Download (599kB)
![[thumbnail of Dapus]](/style/images/fileicons/text.png)
D121171317_skripsi_06-03-2024 dp(FILEminimizer).pdf
Download (780kB)
![[thumbnail of Full Text]](/style/images/fileicons/text.png)
D121171317_skripsi_06-03-2024(FILEminimizer).pdf
Restricted to Repository staff only until 16 May 2027.
Download (1MB)
Abstract (Abstrak)
In the current era of information, YouTube serves as a primary social media platform with a vast daily active user base. The popularity of product review content on YouTube facilitates users in searching for desired product reviews. However, these videos often have a lengthy duration due to advertising needs, creating a demand for sentiment analysis approaches to efficiently determine the conclusions of product reviews. This research aims to develop a model capable of accurately analyzing sentiments in product review videos. It leverages speech recognition technology and the pretrained BERT model to analyze sentiments based on speech within product review videos. The methodology employed in this study involves collecting a dataset of product review videos, extracting audio, converting the audio data into text which augmented using random insertion and random replacement techniques, and applying fine-tuning of the IndoBERT model for the sentiment analysis task. The model's testing and evaluation were conducted in four different scenarios with batch sizes of 4, 8, 16, and 32, measuring performance based on precision, recall, and F1- score. The results indicate that the developed model performs well in sentiment analysis, achieving high F1-scores. The scores reached 97.04% for a batch size of 4, and 97.14% for a batch size of 8. For batch sizes of 16 and 32, the scores were 97.23% and 97.41%, respectively, demonstrating good performance consistency across various batch sizes.
Keyword : Sentiment Analysis, Speech Recognition, BERT, Product Review Videos, NLP
Item Type: | Thesis (Skripsi) |
---|---|
Uncontrolled Keywords: | Sentiment Analysis, Speech Recognition, BERT, Product Review Videos, NLP. |
Subjects: | T Technology > T Technology (General) |
Divisions (Program Studi): | Fakultas Teknik > Teknik Informatika |
Depositing User: | Rasman |
Date Deposited: | 09 Jul 2025 01:40 |
Last Modified: | 09 Jul 2025 01:40 |
URI: | http://repository.unhas.ac.id:443/id/eprint/46224 |