RecSys ’17: Modeling the Assimilation-Contrast Effects in Online Product Rating Systems: Debiasing and Recommendations

Arthur Lee
3 min readJun 3, 2020

--

Recommendation system paper challenge (13/50)

paper link

What problem do they solve?

  1. Predicted item ration by considering the fact historical rating affect user rating bahavior

What model do they propose?

HIALF: Historical Influence Aware Latent Factor Model (HIALF)

b_u and q_up is traditional latent variable model

a_u is meaning how easy the user u to be affected.

h_pi represents the distortion from historical rating

|H_pi| represents the size of historical ratings, which is helpful for how affect the users. Imagine that if you saw 2 item with same rating 5/5 but 1 item with 10,000 ratings and the other one with 10 ratings. People prefer the first one. In this case, they take f(*) function to represent the effect.

β(x) is a categorical function to represent the induced bias when the difference between e_pi and q_up, they applied kernel function for it

Modeling prior expectation e_pi:

they applied 2 strategies:

  1. average of historical mean
  2. weighed more on more recent data

For learning:

With the objective functions with regularization terms, they applied SGD (stochastic gradient descent) to update the parameters.

Data

Comparing models

latent factor (LF) model: traditional model

HEARD (Quantifying Herding Effects in Crowd Wisdom): it is a model with considering social influence of prior collective opinions.

HIALF: their proposal model

HIALF-AVG: prior expectation is taken as the average of historical ratings without emphasis on recent ratings.

Result

Other related blogs:

Beyond Clicks: Dwell Time for Personalization

RecSys’15: Context-Aware Event Recommendation in Event-based Social Networks

RecSys16: Adaptive, Personalized Diversity for Visual Discovery

RecSys ’16: Local Item-Item Models for Top-N Recommendation

COLING’14: Deep Convolutional Neural Networks for Sentiment Analysis of Short Texts

NAACL’19: Utilizing BERT for Aspect-Based Sentiment Analysis via Constructing Auxiliary Sentence

RecSys ’17: Translation-based Recommendation

Best paper in RecSys:

https://recsys.acm.org/best-papers/

My Website:

https://light0617.github.io/#/

--

--

Arthur Lee
Arthur Lee

Written by Arthur Lee

An machine learning engineer in Bay Area in the United States

No responses yet