Wish 107.5, a YouTube channel renowned for its live music performances, has attracted a large and active audience. Understanding viewer sentiments and the topics discussed in the comments section is crucial for enhancing audience engagement and refining content strategy. This study employs Natural Language Processing (NLP) techniques to analyze the sentiments and topics of YouTube comments on Wish 107.5 videos, using a dataset from Kaggle covering the period from December 2019 to December 2020. Google Collab was used for data processing, with sentiment analysis performed using a binary classification tool, and Long Short-Term Memory (LSTM) networks applied for topic modeling. The sentiment analysis model achieved notable performance metrics, including an accuracy of 89%, precision of 87%, recall of 90%, F1-score of 88%, and an ROC AUC of 0.92, demonstrating its effectiveness in classifying YouTube comments. The results revealed a predominantly positive reception of the content, with 70% of comments classified as positive, 20% as neutral, and 10% as negative. Common topics included appreciation for artists, song requests, and feedback on technical aspects. While the model exhibited a training accuracy nearing 1.0, the validation accuracy was 0.78, indicating some overfitting. These outcomes provide valuable insights for content creators and marketers to tailor their strategies according to audience preferences, thereby enhancing overall engagement and satisfaction. By focusing on positive feedback and addressing common requests and technical concerns, content creators can improve their offerings and foster a more engaged and loyal audience. KEYWORDS: binary cross entropy, long short-term memory (LSTM), natural language processing (NLP), sentiment analysis, topic modeling.
Read full abstract