Loading presentation...

Present Remotely

Send the link below via email or IM

Copy

Present to your audience

Start remote presentation

  • Invited audience members will follow you as you navigate and present
  • People invited to a presentation do not need a Prezi account
  • This link expires 10 minutes after you close the presentation
  • A maximum of 30 users can follow your presentation
  • Learn more about this feature in our knowledge base article

Do you really want to delete this prezi?

Neither you, nor the coeditors you shared it with will be able to recover it again.

DeleteCancel

Make your likes visible on Facebook?

Connect your Facebook account to Prezi and let your likes appear on your timeline.
You can change this under Settings & Account at any time.

No, thanks

Exploiting Social Context for Review Quality Prediction

Presenting this paper at TAMU Infolab weekly meeting
by

Amir Fayazi

on 10 October 2014

Comments (0)

Please log in to add your comment.

Report abuse

Transcript of Exploiting Social Context for Review Quality Prediction

Reviews in Their Social Context
Solution Ideas
Problem
Reviews on Their Own
This Paper
Exploiting Social Context for Review Quality Prediction
By Yue Lu (UIUC) et al. from MS Research
appeared in WWW 2010
Data Set
Validating Hypotheses
Experiments
Conclusion
Prediction Performance
Consider
Yelp
Epinions
Amazon
There are High quality and Low quality reviews
i.e. High/Low helpfulness
Problem
Predict the quality of an online review
What do we have?
Items (products, restaurants, etc.)
Users (authors of reviews)
Reviews (written by a user about an item)
The social network of Users
VS
Reviews are treated as standalone text documents
Representation
error
training set size
weights (model)
review
true helpfulness
regularization param
Textual Features
Premises
User's own quality affects his/her reviews quality
User's own quality depends on the quality of his/her peers in the social network
Ways to Incorporate Social Aspect
Augment the feature space
Enforce certain constraints
Constraints
Based on authors' hypotheses
Author Consistency
Reviews from the same author have similar quality
Let's Exploit these
Link Consistency
Quality of users across a trust relationship is closer than two random user
Co-Citation Consistency
If two users are trusted by a third user, their qualities are similar
Trust Consistency
Social link means trust. Trust means higher quality on the side of trustee
Exploit Hypothesized Constraints
Example
Author Consistency
Reviews from the same author have similar quality
Realized as a new regularization term
A is a matrix on Reviews
A(i,j) = 1
If r_i and r_j are written by the same author
A(i,j) = 0
Otherwise
Graph A laplacian = D(A) - A
Modeling
Linear Regression problem
Least squares solution
Reviews are f-dim vectors in
Quality function
Likewise derivations for the rest of constraints
Reviews matrix = [r1, r2, ..., rn]
"w" has closed form solution
Reviews from Ciao UK
Three categories
Cell phones
Beauty
Digital Cameras
Ground Truth
Avg helpfulness
Cellphone
Beauty
Digital Camera
Author Consistency Validation
Consider all review pairs (r_i, r_j)
Partition them into
Same Reviewer
Different Reviewer
pdfs are different based on some test with low p-value
Density Estimates of Review Quality Difference
Social Net. Consistency Validation
Reviewer Quality: Avg quality of his/her reviews
Each graph below is a subset of user pairs
Density Estimates of Reviewer Quality Difference
Observations
Trust has a negative skew
Linked users have less variance
50% Train 50% Test
Polynomial Kernel (order?)
alpha of regularization learned from text-base baseline
Text-free vs Text-only
Effect of adding Social Context to Text only baseline
Automatic determination of review quality
Incorporating social context
Added features
Regularization constraint
Full transcript