44
A simple survey of Diversity A simple survey of Diversity and novelty metrics for and novelty metrics for recommender systems recommender systems Reporter: 孙孙孙 2012.07.11

Diversity and novelty for recommendation system

  • Upload
    zhenv5

  • View
    1.793

  • Download
    2

Embed Size (px)

DESCRIPTION

A simple survey of Diversity and novelty metrics for recommender systems

Citation preview

Page 1: Diversity and novelty for recommendation system

A simple survey of Diversity and A simple survey of Diversity and novelty metrics for recommender novelty metrics for recommender systemssystems

Reporter:孙建凯

2012.07.11

Page 2: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Move beyond accuracy metricsMove beyond accuracy metrics

while the majority of algorithms proposed in recommender systems literature have focused on improving recommendation accuracy

other important aspects of recommendation quality, such as the diversity of recommendations, have often been overlooked.

The recommendations that are most accurate according to the standard metrics are sometimes not the recommendations that are most useful to users[1]

2

Page 3: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Diversity and Novelty

Accurate is not always good: How Accuracy Metrics have hurt Recommender Systems

GroupLensResearch,CHI'06 

Page 4: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Accuracy does not tell the whole story

Page 5: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Diversity

Individual Diversity Aggregate Diversity

Page 6: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Individual Diversity

Diversity Difficulty[3]

Average dissimilarity between all pairs of items recommended to a given user(intra-list similarity) [2,4]

Page 7: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Diversity Difficulty

What We Talk About When We Talk About Diversity [DDR’12 Northeastern University USA]

Like query difficulty in IR

For a specific query and corpus, query difficulty is a measure of how successful the average search engine should be at ad-hoc retrieval.

Page 8: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Diversity Difficulty

Diversity Difficulty is defined with respect to a query and a corpus.

Describes diversity-the number of subtopics which are covered by a list;

Describes novelty-which is inversely proportional to the number of times a list repeats a subtopic

Page 9: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Finding needles in the haystack

Imagine a query with 10 subtopics ,1000 documents relevant to only the first subtopic, and each of the remaining subtopics covered by a single, unique document.

On the other hand ,if there are large numbers of documents relevant to multiple subtopics, it would be easy to produce a diversity list.

Page 10: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Diversity Difficulty function

The maximum amount of diversity achievable by any ranked list-dmax

The ease with a system can produce a diverse ranked list.-dmean

Harmonic function

Page 11: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Examples

Page 12: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Improving Recommendation Lists Through Topic Diversification

Introduce the intra-list similarity metric to access the topic diversification of recommendation lists and the topic diversification approach for decreasing the intra-list similarity

Average dissimilarity between all pairs of items recommended to a given user

Page 13: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Intra-list Similarity

Page 14: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Taxonomy-based similarity Metrics

Instantiate c with their metric for taxonomy-driven filtering.[5]

Page 15: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Topic Diversification Algorithm

Algorithm A brief textual sketch

Page 16: Diversity and novelty for recommendation system

Experiments

precision diversity

Page 17: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Aggregate Diversity

improving recommendation Diversity using ranking-based techniques[IEEE transaction’12]

Use the total number of distinct items recommended across all users as an aggregate diversity measure, define as follows:

Page 18: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

General overview of ranking-based approaches for improving diversity

Page 19: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Re-Ranking Approach

Page 20: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Other Re-ranking Approach

Page 21: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Other Re-ranking Approach

Page 22: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Other Re-ranking Approach

Page 23: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Other Re-ranking Approach

Page 24: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Other Re-ranking Approach

Page 25: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Other Re-ranking Approach

Page 26: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Combining Ranking Approaches

Many possible ways to combine several ranking functions

In this paper , linear combination

Open issue: letor ? Neural network?

Page 27: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Entropy

A study of Heterogeneity in Recommendations for a social Music Service[6]

Page 28: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Open issue:probability

Page 29: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Entropy

Aggregate Entropy:

Item popularity between lists?

Individual Entropy:

subtopic popularity?

Page 30: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Bipartite network

Bipartite network projection and personal recommendation[Tao Zhou, Physical Review]

Solving the apparent diversity-accuracy dilemma of recommender systems[Tao Zhou]

Page 31: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Illustration of resource-allocation process in bipartite network

Page 32: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Solving the apparent diversity-accuracy dilemma

heats probs

Page 33: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Hybrid Methods

weight hybrid

Page 34: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Diversity Measure

Page 35: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Surprisal/novelty

Page 36: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Results-why better?

Page 37: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Surprise me

Tangent: A novel, ‘surprise me’, recommendation algorithm [kdd’09]

Page 38: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Framework of Tangent Algorithm

Suggest items which are not only relevant to user preference but also have a large connectivity to other groups.

Consisting three parts as follows:

1 Calculate relevance score(RS) for each node

2 Calculate bridging score(BRS) for each node

3 Compute the Tangent score by somehow merging two criteria above

Page 39: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Case study

Page 40: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Case study

Page 41: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Call for papers

September 20, 2012

Page 42: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Reference

1. Accurate is not always good: How Accuracy Metrics have hurt Recommender Systems

2.improving recommendation Diversity using ranking-based techniques

3. What We Talk About When We Talk About Diversity

4. Improving Recommendation Lists Through Topic Diversification

5. Taxonomy-driven computation of product recommendations

Page 43: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

Reference

6. A study of Heterogeneity in Recommendations for a social Music Service

7. Bipartite network projection and personal recommendation

8.Solving the apparent diversity-accuracy dilemma of recommender systems

9. Tangent: A novel, ‘surprise me’, recommendation algorithm

Page 44: Diversity and novelty for recommendation system

Copyright 2012 by IRLAB@SDU

thanks