paper reading position bias RecSys

2010-Personalized Click Prediction in Sponsored Search

Posted by Xiaoye's Blog on March 10, 2023

Table of Contents

  1. Problem Definition
  2. Previous Methods & new Methods
  3. Impacts
  4. Further Improvement
  5. TODO & Questions

Problem Definition

  1. Main objective: the design of personalized click prediction
  2. Essential part: development of new user-related features.
    1. No user-related features in baseline model

Previous Methods & new Methods

  1. New method: add user features
    1. demographic group features
    2. user-specific features
      1. activity normalized COEC -> worth to explore

Impacts

  1. Another interesting point is that adding demographic features to the model that already included user-specific features helped improve accuracy, especially for users without enough user-specific click feedback.

Further Improvement

  1. Another direction for future research is to take into account session-based information.
  2. group users in different ways -> clustering maybe

TODO & Questions

  1. Authors introduced Click Feedback feature (CF)
    1. COEC in different segmentation
    2. expected clicks
CATALOG
    FEATURED TAGS
    machine learning paper reading 学习笔记-course note Industry RecSys data structure & algorithms deep learning 公开课-MOOC Airbnb pricing Python causal inference deep reinforcement learning leetcode candidate generation item2vector position bias representation learning travel