Automated Detection and Filtering Off-Topic Comments on YouTube

Abstract

Popularity of YouTube has drawn different advertisers and content sharing users to advertise their content and videos, but the individuals will do off-topic comments (“we love Prophet (PBUH), we love Allah”) which are completely off-topic from that content which was uploaded by influencer. In this paper, we had implemented a novel approach for automatically detect & filtered off-topic comments of YouTube videos based on natural language processing methods. In this work, a specific dataset of comments on YouTube has been collected and preprocessed using feature extraction through Bag-of-Words. Machine learning model LLM (Large Language Model) has been trained and tested to obtain various perspectives in their outputs while classifying comments as on-topic or off-topic. We had cull and categorized the off-topic comments into various categories of both sentiment and sentence types that assist YouTubers in discovering appropriate comments for expanding their audience in this paper. Previous work has dealt with sentiment analysis (spam, positive and negative) or classification of sub-categories within the same sentence categories (e.g., question types) on a text corpus. These have limited extension to non-traditional text corpus such as YouTube comments. We tackle this problem of text extraction and classification from YouTube comments with popular statistical metrics and machine learning models. We assessed each statistical metric versus the machine learning model by cross validation and scoring. The results showed that our method incorporating standard methods works well on the classification task, justifying its viability in helping content creators boost viewer engagement on their channel.