With the development of society, lots of people are using social media to entertain, record life, connect people around the world, and so on. Youtube is one of the most popular platforms in the thousands of social media and everyone can share their videos on Youtube to let others watch it. The goal of the project is to create a user interface for people to use and help them become influencers on Youtube by getting more followers and more views for their videos and channel. Becoming an influencer can let more people watch her or his videos and can help them to get self-satisfaction. Moreover, in this special time and circumstances due to the COVID-19, people’s daily lives are significantly affected. Therefore, our project can also help people reduce pressure, keep mental health, entertain themselves, find their self-value, and even keep the social stabilization by sharing videos with more people.
We have used NLP and random forest models to train separate models for “title” and “description”. Our algorithm provides a robust solution to predict the popularity of youtube videos based on “titles” and “description”. By combining NLP (CountVectorizer, Tfidf) with the RandomForest model, we achieved 79% accuracy in classifying the title into 5-star levels (our baseline model has only 30% accuracy). In order to avoid re-training when others use the interface, we have saved our models to pickles and then linked the pickles to the interface. Moreover, we have designed the interface by considering the user-friendly and interpretable requirements from users, thus people can easily use our interface to identify the potential popularity of the title and description for their videos. In our user interface, users can directly input the titles and descriptions for their videos on Youtube based on their categories, and then our UI will provide the scores towards their input. Then the user will know how well their title and description are and can adjust them to attract more views.
Watch the demo here!
Project by: Fengbin Wang, Hanmi Zou, Yangyi Li, Xiaoyan Wang