UCLA 17S CS 249 Final Project

Dataset

Data found from Stack Exchange Data Dump

Overall Framework

1. Feature Extraction
2. Learning-to-Rank

Implemented Features

• User Features
  • user_age [1 numerical features]
  • user_badge [categorical features]
  • user_reputation [1 numerical features]
  • user_views [1 numerical features]
  • user_votes [1 numerical features]
• User-User Features
  • user-user interactions [1 numerical features]
• Post Features
  • comment_cnt [1 numerical feature]

Instructions

Downloading Data

• Download the data from Stack Exchange Data Dump
• Unzip folders to /raw folder

Preprocessing

cd src/preprocess
./preprocess.py [name of dataset]

• Convert the format from XML to JSON
• Convert HTML-like contents into plaintext
• Link each question to the corresponding answers
  • See data/[name of dataset]/question_answer_mapping.json after preprocessing
• Split the whole set into training and testing sets
  • See data/[name of dataset]/train.* and data/[name of dataset]/test.*
  • Questions without the best answer (ground truth) and with less than two answers are removed

Example of Feature Extraction

Extract user_age features

cd src/feature_extraction
./user_age.py [name of dataset]

Directory Descriptions

Here some descriptions briefly show the purpose of each directory.

• raw/: The directory for the raw data (i.e., Posts.xml, Users.xml, etc)
  • raw/[name of dataset]/: the corresponding raw data for a certain dataset (e.g., StackOverflow)
  • Note that the file names should not be modified.
• data/: The directory for the preprocessed data
  • data/[name of dataset]/: the corresponding preprocessed data for a certain dataset (e.g., StackOverflow)
• src/: The directory for all source codes
  • src/preprocess/: Codes for preprocessing raw data
• model/: The directory for some trained model on large English corpus