Document Classification: Model Training and Deployment

The goal of this workflow is to do spam classification using YouTube comments as the dataset. The workflow starts with a data table containing some YouTube comments taken from the YouTube Spam Collection Data Set at the UCI ML Repository[1] . The data is available in the workflow directory. The comments are divided into two categories, spam and ham (non-spam). The distribution of the values in both categories is roughly equal.

First, the comments are converted into documents, whose category is the class spam or ham. The documents are then preprocessed by filtering and stemming. After that, the documents are transformed into a bag of words, which is filtered again. Only terms that occur at least in 1% of the documents (at least in 3 documents) will be used as features and not be filtered out. Then the documents are transformed into document vectors.
The document vectors are a numerical representation of documents and are in the following used for classification via a support vector machine.

The lower part contains the deployment workflow.

External resources

• Sentiment Classification of Documents
• YouTube Spam Collection Dataset