13
ReBoRK: A REVIEW-BASED BOOK RECOMMENDER FOR K-12 READERS Sole Pera CS 652-2012

ReBoRK : A Review-based Book Recommender for K-12 Readers

  • Upload
    rufina

  • View
    59

  • Download
    0

Embed Size (px)

DESCRIPTION

ReBoRK : A Review-based Book Recommender for K-12 Readers. Sole Pera CS 652-2012. Introduction. Existing book recommenders Make suggestions that match readers’ interests Problems:. One-size-fits-all. Required personal historical data may not always be available. - PowerPoint PPT Presentation

Citation preview

Page 1: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

ReBoRK:A REVIEW-BASED BOOK

RECOMMENDER FOR K-12 READERS

Sole PeraCS 652-2012

Page 2: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

2

Introduction Existing book recommenders

Make suggestions that match readers’ interests

Problems:

Required personal historical data may not always be available

Suggest books without considering the reading ability of

its users

Existing Recommenders

Not personalized enough

One-size-fits-all

Page 3: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

3

Proposed Solution ReBoRK

Review-based book recommender that generates personalized suggestions recommendations tailored towards K-12 students

Feature/ Opinion

Similarity

Metadata Similarity

Grade Level

Similarity

? ? ?

Item-Item Similarity

ReBoRK

Page 4: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

4Information Extraction Anaphora Resolution

Use Guitar system to replace all discourse referents by their corresponding entities the reviews

Review Summarization Use [Pera et al, WISE `11] and SentiWordNet to

retain the portions of the reviews that express sentiment

Linguistic and Syntactic Analysis Use Stanford's NLP tools

Part-Of-Speech tagging Dependency parsing

Extraction Module

Page 5: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

5Information Extraction Information Extraction Rules

Based on improvements upon the IE rules proposed in [Kamal et al., WIMS `12] Identify features on reviews Identify opinions on reviews Identify relationships between features and

opinions Example

Extraction Module

“It is written in simple rhyming patterns and illustrated with adorable pictures!”

Review for “Bear's New Friend” by Karma Wilson, extracted from Amazon.com

Page 6: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

6Readability

User profile

Potential recommendations

Averaged Grade Level

Grade Level

?MatchedReading

Level

Recommendation Module

Page 7: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

7Feature-Opinion Similarity

User profile

Potential recommendation

Feature-Opinion Distribution

Feature-Opinion Distribution

?Matched Preferenc

es

Recommendation Module

Page 8: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

8Item-Item Similarity

User profile

Potential recommendation

<…, , , , , , …>

< …, , , , , , …> ?Matched

Bookmarking Patterns

Recommendation Module

Page 9: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

9Content Similarity

User profile

Potential recommendation

Metadata: titles + descriptions

Metadata: title + description

?Matched Content

Recommendation Module

Page 10: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

10

Item-Item

Readability

Feature/ Opinion

Metadata

Ranking

Top-10 Recommendati

ons

Fusion Strategies

Recommendation Module

Page 11: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

11Proposed Validation Dataset

Metrics Precision Recall F-Measure

Extraction Module

Page 12: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

12Proposed Validation Datasets

Validation Strategy N-fold cross-validation

Metrics Precision@K Mean Reciprocal Rank Normalized Discounted Cumulative Gain

(nDCG)

Recommendation Module

Page 13: ReBoRK : A Review-based  Book  Recommender  for K-12 Readers

13Questions