# Classification of skills based on job ads

I have around 1,000 job ads in the filed of IT (in excel file). I want to find the skills which are mentioned in each of ads. and then find the similar jobs based on skills.

My method: I created 12 categories Such as programming skills, testing skills, communication skills, network skills, ... . Each advertisement may belong to 3-4 categories. In this case, some said multi-variate classification or Multi label classification is useful. But I don't know how to do this kind of classification in RapidMiner.

1- Does anyone know how to do multi-variate classification or Multi label classification in RapidMiner? or is there another way?

2- Do you recommend "classification" in order to analysis required job skills? or another technique?

3- Is there any better way to classify the skills which are stated in job ads?

I'm new in the field of text mining. Please let me know if you have any idea. Thanks

• Have you already manually assigned these 12 category labels to some or all of your dataset? If you haven't, then this becomes more a problem of unsupervised clustering than supervised classification. – Ryan J. Smith Apr 6 '15 at 1:20
• Yes, I manually assigned 12 category. – Zahra Hnn Apr 9 '15 at 11:24

## 2 Answers

I don't know about RapidMiner, but for a beginner, Orange seems an awesome open-source tool for the job.

Particularly if you want to do unsupervised learning, i.e. clustering. You'd probably want to preprocess your textual data in some bag of words model and save it as CSV, I guess.

I also hear Orange has a text mining add-on, but I haven't tried it yet.

RapidMiner is actually pretty good at multivariate analysis. The difficult part is getting the data into an appropriate format.

In order to perform clustering, the data should be converted to a format so that two jobs can be compared based on skills. For example, each job can be represented as a 12-dimensional boolean vector $$[x_1,\ldots,x_{12}]$$ where each $x_i$ represents a skill which is $true$ if the job covers the skill and $false$ otherwise. A distance measure between jobs is then for example Hamming-distance. RapidMiner offers a wide range of distance measures between different kinds of feature vectors (Hamming distance is called NominalDistance in RapidMiner, but others might be better).