Tik Tok Video Content Classification

Random Forest and Gradient Boosting

Classification Model

Introduction

TikTok users can report videos that they believe violate the platform's terms of service. There are millions of TikTok videos made and viewed every day, which means that many of them receive reports—too many for a human moderator to individually review. Analysis indicates that when authors do violate the terms of service, they're much more likely to be presenting a claim than an opinion. Therefore, it is useful to be able to determine which videos make claims and which videos are opinions. TikTok wants to build a machine learning model to help identify claims and opinions. A human moderator will be less likely to review videos with the label "opinion." Videos that are labeled as claims will be further sorted by a downstream process to determine whether they should get prioritized for review.

All the files related to this project are available at Github.com/nitin6753/TikTok_Content_Classification

Summary

Objective:

Develop a machine learning model to classify TikTok videos as either claims or opinions, with a focus on enhancing the efficiency of content moderation by prioritizing videos likely to violate terms of service.

Tasks Performed:

Conducted extensive exploratory data analysis (EDA) to understand the dataset, address outliers, and manage missing values, preparing the data for modeling.
Developed and evaluated a logistic regression model to predict whether a user is verified, finding that verified users are more likely to post opinions.
Engineered features, particularly tokenizing video transcription text using NLP techniques, to enhance model performance.
Built and tuned Random Forest and Gradient Boosting models using GridSearchCV, prioritizing recall due to the higher cost of false negatives (misclassifying claims as opinions).
Selected Random Forest as the champion model after evaluating its performance on test data, where it achieved near-perfect classification with only five misclassifications out of 3,817 samples.

Key Result Area:

The Random Forest model was chosen as the final model, demonstrating superior performance in classifying videos, with a focus on minimizing false negatives.
Key predictors identified were related to video engagement metrics, such as video view count, like count, share count, and download count, which contributed significantly to the model's predictive accuracy.

Recommended Actions:

Implement the Random Forest model in TikTok's moderation pipeline to prioritize content more likely to be claims.
Further investigate and optimize features related to video engagement for potential improvements in classification accuracy.
Continue monitoring the model's performance, particularly focusing on recall, to minimize the risk of misclassifying claims as opinions.

Skills Demonstrated: EDA, Logistic Regression, Random Forest, Gradient Boosting, Hyperparameter Tuning, Cross-Validation, Natural Language Processing (NLP), Model Evaluation.

Home

Page updated

Google Sites

Report abuse

Tik Tok Video Content Classification

Introduction

Summary

Develop a machine learning model to classify TikTok videos as either claims or opinions, with a focus on enhancing the efficiency of content moderation by prioritizing videos likely to violate terms of service.

Tasks Performed:

Conducted extensive exploratory data analysis (EDA) to understand the dataset, address outliers, and manage missing values, preparing the data for modeling.

Developed and evaluated a logistic regression model to predict whether a user is verified, finding that verified users are more likely to post opinions.

Engineered features, particularly tokenizing video transcription text using NLP techniques, to enhance model performance.

Built and tuned Random Forest and Gradient Boosting models using GridSearchCV, prioritizing recall due to the higher cost of false negatives (misclassifying claims as opinions).

Selected Random Forest as the champion model after evaluating its performance on test data, where it achieved near-perfect classification with only five misclassifications out of 3,817 samples.

Key Result Area:

The Random Forest model was chosen as the final model, demonstrating superior performance in classifying videos, with a focus on minimizing false negatives.

Key predictors identified were related to video engagement metrics, such as video view count, like count, share count, and download count, which contributed significantly to the model's predictive accuracy.

Recommended Actions:

Implement the Random Forest model in TikTok's moderation pipeline to prioritize content more likely to be claims.

Further investigate and optimize features related to video engagement for potential improvements in classification accuracy.

Continue monitoring the model's performance, particularly focusing on recall, to minimize the risk of misclassifying claims as opinions.

Skills Demonstrated: EDA, Logistic Regression, Random Forest, Gradient Boosting, Hyperparameter Tuning, Cross-Validation, Natural Language Processing (NLP), Model Evaluation.