I plan to filter an input document using 4 different labels.

Just for an example, a document discussing about movie summary needs to be labeled with 4 labels (Romance, Drama, Fiction, Hollywood). Only if it is labeled with these specific 4 labels, will it be considered as a "PASS". A lack of either of the labels should mark the document as "Fail"

PASS = ["romance", "drama", "fiction", "hollywood"]

FAIL = ["romance", "drama", "fiction", "chinawood"]

In terms of labels "pass" or "fail", it is a binary classification problem, but in terms of these 4 labels, it is a multi-label document classification problem.

Additionally, I do not have a lot of documents that are tagged with these 4 labels simultaneously. So, I thought of decomposing the problem into a binary classification problem after reading this and using ensemble methods to chain individual binary classifiers together.

However, I have a built a set of 10,000 documents for each class ("Romance" vs "Not romance"), ("Drama" or "Not drama"), ("Fiction" or "Not fiction") and ("Hollywood" or "Not hollywood").

Classifier 1: ("Romance" or "Not romance")

Results of classifier 1 are input to the classifier 2

Classifier 2 ("Drama" or "Not drama")

Results of classifier 2 are input to the classifier 3

Classifier 3 ("Fiction" or "Not fiction")

Results of classifier 3 are input to the classifier 4

Classifier 4 ("Hollywood" or "Not hollywood")

Results of classifier 4 are the final set of set of documents labeled with the four labels.

In thisway, it becomes a hierarchical Multi-label problem.

Note: The movie genre labels are just used for the example. The real problem here connects with very precise medical/health labels (For example ["Lung Cancer", "Adult Female", "Journal Report", "Asian"]). Hence, the training document set that has all the four labels simultaneously is very small. This forced me to think towards decomposing the problem into a binary classification problem.

Additional Note: This is my first machine learning project and I am quite new to this task.

So my question was, if I was heading the right direction or should I read more into it? Any hint or constructive criticism is welcome.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.