2. Job recommendation with Hawkes processes
Alibaba group ,Wenming Xiao
Winner of Recsys challenge 2016
2
3. Introduction
• [Recsys Challenge 2016]
•
•
3
PROBLEM STATEMENT
Given the job role information of the users, the
content of job posting, and the historical log of
users activities, the key task of this contest is to
recommend a list of job posts, which the users
might interact with in the next week
Job post | impressions | interaction | users
www.xing.com
4. Framework
• Xiao et al[2]
4
Feature Engineering
Job user interact
Xi Xu Bui
Semantic fx Education Behavior Mat
[title, industry] [0,0,1,0] [0,1]
Master
5. Related work
• Xiao et al[2]
5
Weak learners
LR GBDT XGBOOST
score 1 Score2 Score3
Pairwise scores
The temporally intensity measure the
interest of the user to an item in time
6. Evaluation measure
6
R: Set of predicted user item pair
T: set of ground truth user item pair
Pk: Precision@k
7. Review
• Few feature engineering process and ensemble learning
• Temporal intensity [For the recurrent user activity toward type
of item, Provide intensity the user will have toward new item]
• Originality: 4
• Novelty : 3
7
8. References
• [2] W. Xiao, X. Xu.,K. Liang,J Mao,J Wang 2016. Job Recommendation with Hawkes Process. 10th
ACM Conference on Recommender Systems - RecSys ’16 (in Press).
For Recsys challenge paper and presentation slides http://2016.recsyschallenge.com/
8