StackOverflow-Tag-Prediction

We have to predict the tags from the tile and description associated to the questions so that we can better create an ecosystem to send the questions to the right set of people to answer it.

It is a multi-label classification problem

Data Description

Source - https://www.kaggle.com/c/facebook-recruiting-iii-keyword-extraction/data

Attribute Information

Field Name	Description
Id	Unique idenfier for each question
Title	Tile of the question as the name suggest
Body	description of the question
Tags	Tags assigned by the person who posted that question

Analysis

• Frequency of the tags

• Predict what's the most frequent number of tags found in most of the questions

• WordCloud to represent which tag is more frequent

Model Comparision

Model	Featurization	Loss	Micro F1 Score
OneVsRest+SGD Classifier	Tf-idf	log	0.3738
OneVsRest+Log Reg. Classifier	Tf-idf	log	0.395
OneVsRest+SGD Classifier	Bag-of-words	log	0.292
OneVsRest+SGD Classifier	Bag-of-words	Hinge	0.3026

References

Applied AI Course