Text Classification/Labeling using Description

rikin_j_parekh

Hi All,

 

I am new to RapidMiner and would like to perform labeling on a 'Long Description' column using a CSV file. I will be working with 2 columns mainly, 'Long Description' and 'Label'. The 'Label' is applied based on the 'Long Description' value. I have 1000 rows out of which 80% of 'Label' values are already applied as a training set. I wish to populate the remaining 20% 'Label' values using the 'Long Description' value.

All Label Values - 

Cancellation

Price Increase

Normal Payment

Payoff

Price Decrease

Installer Installation Issue

Past Due Payment

Change Order

Incentive Payment

Assumption

Completion Certificate

Interest

Referral

Example -

Long Description - Please review change order in installation phase - loan amount increasing from USD 21;851.00 to USD 24;501.00
Label - Price Increase

Long Description - Cancellation request with SPV Assignment

Label - Cancellation

How should I proceed with this using RapidMiner and what should be the steps to perform the same?

 

Thanks

Telcontar120

You should search the forums for some of the threads on text mining, you will find a lot of helpful information there.  This is a classic classification problem.  You'll use your "long description" as the text, process and tokenize it, and then use the resulting word vectors to predict the label.

However, you may find that you need to consolidate labels.  You have a lot of distinct values, and classification problems increase in complexity when you have have a lot of potential individual label values to predict.  So you may find better success by grouping some of the existing labels together into larger categories.  That's something that you will need to play around with manually, there's not an easy way to automate that in RapidMiner.

 

 

Telcontar120

You should search the forums for some of the threads on text mining, you will find a lot of helpful information there.  This is a classic classification problem.  You'll use your "long description" as the text, process and tokenize it, and then use the resulting word vectors to predict the label.

However, you may find that you need to consolidate labels.  You have a lot of distinct values, and classification problems increase in complexity when you have have a lot of potential individual label values to predict.  So you may find better success by grouping some of the existing labels together into larger categories.  That's something that you will need to play around with manually, there's not an easy way to automate that in RapidMiner.