Improving the Effectiveness of Content Popularity Prediction Methods using Time Series Trends

摘要

We here present a simple and effective model to predict the popularity of webcontent. Our solution, which is the winner of two of the three tasks of theECML/PKDD 2014 Predictive Analytics Challenge, aims at predicting userengagement metrics, such as number of visits and social network engagement,that a web page will achieve 48 hours after its upload, using only informationavailable in the first hour after upload. Our model is based on two steps. Wefirst use time series clustering techniques to extract common temporal trendsof content popularity. Next, we use linear regression models, exploiting aspredictors both content features (e.g., numbers of visits and mentions ononline social networks) and metrics that capture the distance between thepopularity time series to the trends extracted in the first step. We discusswhy this model is effective and show its gains over state of the artalternatives.