Differential Adaptive DiffusionUnderstanding Diversity and Learning Whom to Trust in Viral Marketing<br />Hossam Sharara, ...
Introduction<br />Viral marketing techniques rely on the social network among customers to spread product recommendations<...
Motivation<br />Ann<br />Janet<br />John<br />Bob and Mary will definitely be interested. However, I think Ann is not much...
Motivation<br />Peer-influence is dynamic<br />Dependent on previous user interactions<br />Viral marketing strategies hav...
Motivation<br />Peer-influence is dependent on the type of product being spread<br />Users have varying preferences for di...
Background<br />Popular Diffusion Models<br />Threshold Models (e.g. Linear threshold model)<br />Cascade Models     (e.g....
Objectives<br />Capture the diversity in user preferences for different products<br />Model the change in influence probab...
Outline<br />Case Study: Digg.com<br />Differential Adaptive Diffusion Model<br />Adaptive Viral Marketing<br />Conclusion...
Case Study: Digg.com<br />Social news website<br />Users “submit” stories in differenttopics, which can then be “digged”by...
Case Study: Digg.com<br />Following links define the social network<br />User submissions serve as  proxy of user preferen...
Dataset<br />Social Network (user-user following links)<br />11,942 users<br />1.3M follow edges<br />Digg Network (user-s...
Observation 1User Submissions vs. Diggs<br />Smaller percentage of users who digg stories in topics that vary significantl...
Characterizing User Submissions<br />Focused UsersHighly skewed preferences toward one or two topics<br />Biased UsersLess...
Observation 2Effect of Homophily on Adoption<br />Peers with different topic preferences lose confidence in each other’s r...
Outline<br />Case Study: Digg.com<br />Differential Adaptive Diffusion Model<br />Adaptive Viral Marketing<br />Conclusion...
Differential Adaptive Diffusion<br />The influence probability between two peers (u,v) for product category c can be re-wr...
Experimental Evaluation<br />Evaluate the model performance in predicting future adoptions<br />We use the first four mont...
Baselines<br />We compare our approach with two baselines* that incorporate temporal dynamics in learning the influence pr...
Results<br />The Adaptive model, taking both the  diffusion dynamics and the  users heterogeneity into  account, yields be...
Outline<br />Case Study: Digg.com<br />Differential Adaptive Diffusion Model<br />Adaptive Viral Marketing<br />Conclusion...
Adaptive Viral Marketing<br />User recommendations are most effective when recommended to the right subset of friends<br /...
Adaptive Rewards<br />Successful recommendations are awarded (αxr)units, while failed ones are penalized ((1-α) xr) units<...
Experimental Setup<br />Agent-based models to simulate the behavior of customers in different settings<br />When an agent ...
Experimental Setup<br />Two sets of experiments<br />Fully observable: The agents are allowed to directly observe the pref...
Fully Observable<br />Intermediate values for α (e.g. α= 0.5) consistently maintains high adoption rates and high overall ...
Learning Preferences<br />Allowing agents to learn the preferences accounts for both the product preference as well as the...
Effect of Spammers<br />To test the robustness of our proposed method, we inserted spamming agents in the network<br />A s...
Effect of Spammers<br />The network adapts to the presence of spammers (dropping their confidence levels), and continues t...
Outline<br />Case Study: Digg.com<br />Differential Adaptive Diffusion Model<br />Adaptive Viral Marketing<br />Conclusion...
Conclusion<br />Network dynamics and users heterogeneity have a considerable impact on user interactions<br />The proposed...
Future Work<br />Potential Applications<br />Social Recommendation<br />Collaborative Filtering<br />Analyzing the impact ...
Thank You<br />Questions?<br />
Differential Adaptive Diffusion: Understanding Diversity and Learning whom to Trust in Viral Marketing
Upcoming SlideShare
Loading in …5
×

Differential Adaptive Diffusion: Understanding Diversity and Learning whom to Trust in Viral Marketing

469 views

Published on

Published in: Business, Technology
0 Comments
0 Likes
Statistics
Notes
  • Be the first to comment

  • Be the first to like this

No Downloads
Views
Total views
469
On SlideShare
0
From Embeds
0
Number of Embeds
1
Actions
Shares
0
Downloads
10
Comments
0
Likes
0
Embeds 0
No embeds

No notes for slide
  • With the increased customer resistance
  • ----- Meeting Notes (7/18/11 15:48) -----encouraging her
  • Add a slide for describing digg.com / add key words (submissions vs. diggs)
  • One interpretation for the high divergence users that they are imitatorsWe use the topic distribution of the user postings as an influence-independent source for measuring preferencesFig1: KL-divergence between the topics of user submissions and diggsFig2: KL-divergence between user submissions (prefs) and uniform distribution of topics
  • * Reduce text- In order to capture both the heterogeneity in user preferences as well as the temporal dynamics, we split the influence probability into two components- The product preferences are either given or can be inferred from an influence-independent source (such as user submission in the case of Digg)
  • Mention that these baselines use the independent cascade model
  • Color the series if possible We compared against two models that take the changing dynamics of the influence probability into account, but doesn’t address the user preference / heterogeneity aspect
  • If a user is very selective and makes each recommendation to only a few friends, then the chances of success are slim due to limited network exposure. On the other hand, recommending a product to everyone may have limited returns as well, due to the effect of irrelevant recommendations on the confidence levels between peersThe natural question to ask now is
  • Typical viral marketing strategies reward users for successful recommendation, but don’t penalize them for failed ones 0 representing fully conservative behavior and 1 representing fully nonconservative behavior.
  • No text for alpha=1,0 – just arrowarrows for this and next
  • In more realistic settings, the user friends learn them from her responses to different recommendation*The settings where users are allowed to learn preferences are better than the settings where users directly observe preference (prev slide) as this one
  • Varying the percentage of spammers (individuals recommending any product to all their peers), and analyzing the over all effect (set alpha = 0.5)As percentage of spammers increases, the overall adoption and confidence level decreases. However, the network is able to adapt to their presence by assigning lower confidence levels to
  • *Add future work / implicationsUtilized in social recommendations (incorporate both “trust” in a friend and their preference combined with your preference to make better automated recommendations – Facebook, x like this page)
  • Differential Adaptive Diffusion: Understanding Diversity and Learning whom to Trust in Viral Marketing

    1. 1. Differential Adaptive DiffusionUnderstanding Diversity and Learning Whom to Trust in Viral Marketing<br />Hossam Sharara, William Rand, LiseGetoorICWSM, Jul 18th 2011<br />
    2. 2. Introduction<br />Viral marketing techniques rely on the social network among customers to spread product recommendations<br />Key Assumption<br />Peer-influence is both static, and independent of the product type<br />
    3. 3. Motivation<br />Ann<br />Janet<br />John<br />Bob and Mary will definitely be interested. However, I think Ann is not much into movies<br />Mary<br />WOW… I’ll send it over to everyone<br />Women’s fashion Store(Invite a friend and get 10% off your next purchase)<br />MovieRental.com(Refer a friend and get $10 off your next rental)<br />Bob<br />
    4. 4. Motivation<br />Peer-influence is dynamic<br />Dependent on previous user interactions<br />Viral marketing strategies have an implicit effect on the underlying social network<br /> Sometimes changing the structure of the underlying network altogether <br />
    5. 5. Motivation<br />Peer-influence is dependent on the type of product being spread<br />Users have varying preferences for different products<br />Both factors play a role in product adoption<br />
    6. 6. Background<br />Popular Diffusion Models<br />Threshold Models (e.g. Linear threshold model)<br />Cascade Models (e.g. Independent cascade model)<br />Influence probabilities are assumed to be static, insensitive to the product type, and known in advance<br />
    7. 7. Objectives<br />Capture the diversity in user preferences for different products<br />Model the change in influence probabilities across multiple campaigns<br />Design a viral marketing strategy that takes these factors into account<br />
    8. 8. Outline<br />Case Study: Digg.com<br />Differential Adaptive Diffusion Model<br />Adaptive Viral Marketing<br />Conclusion and Future Work<br />
    9. 9. Case Study: Digg.com<br />Social news website<br />Users “submit” stories in differenttopics, which can then be “digged”by other users<br />Users can “follow” other users to get their submissions and diggs on their homepage<br />
    10. 10. Case Study: Digg.com<br />Following links define the social network<br />User submissions serve as proxy of user preferences for different topics<br />User diggs are analogous to product adoptions<br />
    11. 11. Dataset<br />Social Network (user-user following links)<br />11,942 users<br />1.3M follow edges<br />Digg Network (user-story digging links)<br />48,554 news stories<br />1.9M digg edges<br />6 months (Jul 2010 – Dec 2010)<br />
    12. 12. Observation 1User Submissions vs. Diggs<br />Smaller percentage of users who digg stories in topics that vary significantly from the topics they post in <br />Most users adopt only stories of interest to them <br />
    13. 13. Characterizing User Submissions<br />Focused UsersHighly skewed preferences toward one or two topics<br />Biased UsersLess skewed preferences toward a larger set of topics<br />Balanced (Casual) UsersAlmost uniform preference across all topics<br />
    14. 14. Observation 2Effect of Homophily on Adoption<br />Peers with different topic preferences lose confidence in each other’s recommendations over time (followed diggs/adoption )<br />Peers with similar topic preferences gain confidence in each other’s recommendations over time (followed diggs/adoption )<br />
    15. 15. Outline<br />Case Study: Digg.com<br />Differential Adaptive Diffusion Model<br />Adaptive Viral Marketing<br />Conclusion and Future Work<br />
    16. 16. Differential Adaptive Diffusion<br />The influence probability between two peers (u,v) for product category c can be re-written as<br />Confidence of user vin u at campaign i<br />Preference of user vin product type c<br /><ul><li>The confidence weights are updated at the end of each campaign</li></li></ul><li>Kernel Functions<br />Linear Kernel<br />If vadopts the product<br />Each peer uwho recommended the product to vgets a credit proportional to the time elapsed (last recommender  max. credit)<br />If vdoesn’t adopt the product<br />Each peer uwho recommended the product to vgets penalized proportional to the time elapsed (last recommender  max. penalty)<br />
    17. 17. Experimental Evaluation<br />Evaluate the model performance in predicting future adoptions<br />We use the first four months in Digg.com dataset for learning the influence probabilities, and the last two months for testing <br />
    18. 18. Baselines<br />We compare our approach with two baselines* that incorporate temporal dynamics in learning the influence probabilities<br />Bernoulli<br />Each product recommendation  Bernoulli Trial<br />Influence probabilities are estimated using MLE over a given contagion time for each user<br />Bernoulli-PC<br />Same Bernoulli representation<br />Partial credit for each recommending peer within the contagion period<br />*A. Goval, F. Bonchi, and L. Lakshmanan. “Learning influence probabilities in social networks.” In Proceedings of the Third ACM International Conference on Web Search and Data Mining (WSDM’10), 2010<br />
    19. 19. Results<br />The Adaptive model, taking both the diffusion dynamics and the users heterogeneity into account, yields better performance<br />
    20. 20. Outline<br />Case Study: Digg.com<br />Differential Adaptive Diffusion Model<br />Adaptive Viral Marketing<br />Conclusion and Future Work<br />
    21. 21. Adaptive Viral Marketing<br />User recommendations are most effective when recommended to the right subset of friends<br />Highly selective behavior  Limited exposure<br />Spamming  lower confidence levels, limited returns<br />What is the appropriate mechanism for maximizing both the product spread and adoption? <br />
    22. 22. Adaptive Rewards<br />Successful recommendations are awarded (αxr)units, while failed ones are penalized ((1-α) xr) units<br />α  conservation parameter<br />Most existing viral marketing strategies assume α=1 (no reason for the user to be selective)<br />The penalty term helps maintain the average overall confidence level between different peers<br />
    23. 23. Experimental Setup<br />Agent-based models to simulate the behavior of customers in different settings<br />When an agent adopts the product, it makes a probabilistic decision to send a recommendation based on its knowledge about the peers’ preferences<br />The objective of each agent is to maximize its expected reward according to the existing strategy<br />
    24. 24. Experimental Setup<br />Two sets of experiments<br />Fully observable: The agents are allowed to directly observe the preferences of their peers <br />Learning preferences: The agents have to learn the peer’s preferences based on their response to previous recommendations<br />Simulate the diffusion of 500 campaigns for products from 5 different categories <br />We use a linear kernel for adjusting the confidence levels between peers after each campaign<br />
    25. 25. Fully Observable<br />Intermediate values for α (e.g. α= 0.5) consistently maintains high adoption rates and high overall confidence over large number of marketing campaigns<br />
    26. 26. Learning Preferences<br />Allowing agents to learn the preferences accounts for both the product preference as well as the confidence level<br />
    27. 27. Effect of Spammers<br />To test the robustness of our proposed method, we inserted spamming agents in the network<br />A spamming agent forwards all product recommendation for all its peers, regardless of their preferences<br />We set (α = 0.5) for all the other agents, and vary the number of seeded spammers<br />
    28. 28. Effect of Spammers<br />The network adapts to the presence of spammers (dropping their confidence levels), and continues to maintain adoption levels through trusted links<br />
    29. 29. Outline<br />Case Study: Digg.com<br />Differential Adaptive Diffusion Model<br />Adaptive Viral Marketing<br />Conclusion and Future Work<br />
    30. 30. Conclusion<br />Network dynamics and users heterogeneity have a considerable impact on user interactions<br />The proposed adaptive diffusion model incorporates both aspects to better model the diffusion process<br />Adaptive rewarding mechanism for viral marketing maintains higher confidence levels over time<br />
    31. 31. Future Work<br />Potential Applications<br />Social Recommendation<br />Collaborative Filtering<br />Analyzing the impact of the proposed model on opinion leader identification<br />Incorporating the time-variability aspect of user-product preferences in the model<br />
    32. 32. Thank You<br />Questions?<br />

    ×