Date post: | 02-Aug-2015 |
Category: |
Data & Analytics |
Upload: | ramaninder-singh-jhajj |
View: | 78 times |
Download: | 6 times |
How to Ask for a Favor
A Case Study on the Success of Altruistic Requests
Presented By: Ramaninder Singh Jhajj
Original Authors of the Research Paper: Tim Althoff, Christian, Dan Jurafsky
Introduction• Requests : Act of asking formally for
something.
• Core of many social media systems.
• Factors that lead members to satisfy a request remain largely unknown.
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 2
Questions• Does the Language of the Request Matters?
• If Yes, How it matters?
• Is it possible to predict the success or failure of Altruistic Requests.
No Incentive in return for the favor
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 3
Outcome• Goal was to understand what motivates people to
give when they do not receive anything tangible in return.
• Developed a framework for controlling potential confounds while studying the role of two aspects that characterize compelling requests:
– Social Factors– Linguistic Factors
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 4
DOES THE LANGUAGE OF THE REQUEST MATTERS?
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 5
Complexities in Requests• What is being requested?
• What the giver receives in return?
• Group Dynamics
• In peer-to-peer, people are more likely to give to
projects what others are already giving to.
How to distangle the effect of
these factors and focus on
language?
(Mitra and Gilbert 2014; Mollick 2014; Etter, Grossglauser, and Thiran 2013; Ceyhan, Shi, and Leskovec 2011; Teevan, Morris, and Panovich 2011; Burke et al. 2007; Wash 2013; Cialdini 2001) 6
Random Acts of Pizza• Online community facilitating sending and
receiving free pizzas between strangers.
• As they say „Together we aim to restore faith in humanity, one slice at a time“
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 7
• What? – All requests ask for same thing, a pizza.
• Incentive? – No incentives or rewards
• Group Dynamics? – Social Networking, Requests are largely Textual
• Request Satisfiers? – Single User
What about complexities?
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 8
Dataset• 21,577 posts (8th Dec. 2010 – 29th Sept.
2012)• Only users with a single request, 5728
requests.• Average success rate 24.6%• Split Dataset mirrioring the success rate:
– Development (70%)– Test Set (30%)
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 9
HOW DOES LANGUAGE OF THE REQUEST MATTERS?
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 10
Success Factors1. Textual Factors– Politeness– Evidentiality– Reciprocity– Sentiment– Length
2. Social Factors– Status– Similarity
3. Narratives– Money– Job– Student– Family– Craving
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 11
• “My gf and I have hit some hard times with her losing her job and then unemployment as well for being physically unable to perform her job due to various hand injuries as a server in a restaurant. She is currently petitioning to have unemployment reinstated due to medical reasons for being unable to perform her job, but until then things are really tight and ANYTHING would help us out right now.
I [...] would certainly return the favor again when I am able to reciprocate.”
Example
Length : Favorable for success
Evidentiality : Urgents requests are met fast.
Reciprocity : Promise to return the favor
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 12
Measuring the Factors (1/2)• Temporal Factors : Temporal or Seasonal effects are
controlled
• Politeness : Measure politeness by extracting 19 individual features deom computational politenes model (Danescu-Niculescu-Mizil et al. 2013).
• Evidentiality : Presence of an image link, providing evidence for their claim (86% of images in random sample included some kind of evidence)
• Reciprocity : If the request includes phrases like „pay it forward“, „pay it back“ or „return the favor“.
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 13
Measuring the Factors (2/2)• Sentiment : Extracted sentiment annotations for each
sentence of the request using the Stanford CoreNLP Package, count features based on lexicons of positive and negative words from LIWC, detecting emoticons.
• Length: Total number of words in request.
• Status : Karma points, measure whether or not uses has posted on RAOP before, user account age.
• Narrative : measure usage of all 5 narratives by word count features that measures how often a given requests mention words from the previously defined nattative lexicons.
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 14
Regression Results (1/2)Coefficient Estimat
eSE
Community Age -0.13*** 0.01
First Half of Month 0.22** 0.08
Gratitude 0.27** 0.08
Including Image 0.81*** 0.17
Reciprocity 0.32** 0.10
Strong Positive Sentiment 0.14 0.08
Strong Negative Sentiment
-0.07 0.08
Length in 100 words 0.30*** 0.05***p < 0.001, **p < 0.01, *p < 0.05
Temporal Control
PolitenessEvidentiality
Reciprocity
Length
Sentiment
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 15
Regression Results (2/2)Coefficient Estimat
e SE
Karma 0.13*** 0.02
Posted in RAOP before 1.34*** 0.16
Narrative Craving -0.34*** 0.09
Narrative Family 0.22* 0.09
Narrative Job 0.26** 0.09
Narrative Money 0.19** 0.08
Narrative Student 0.09 0.09***p < 0.001, **p < 0.01, *p < 0.05
Status
Narratives
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 16
Interpretation
Length : 50 wordsNarrative : Craving
Length: 50 wordsNarrative : Job and Money
Length : 150 wordsNarrative : Job and MoneyIncludes Picture, Gratitute and Reciprocity
9.8%
19.4%
56.8%
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 17
IS SUCCESS PREDICTABLE?
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 18
Test Setup• Predicting held-out requests (1.6k)
• Model: Area under receiver operating characteristic curve (ROC AUC) – Illustrate performance of binary classifier systems.– Graphical plot.– True Positive Rate VS False Positive Rate
(Friedman, Hastie, and Tibshirani 2010; DeLong, DeLong, and Clarke-Pearson 1988) 19
Test Results
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests
Feature ROC AUC (***p < 0.001)
Random Baseline 0.500
Unigram Baseline 0.621***
Bigram Baseline 0.618***
Trigram Baseline 0.618***
Text Features 0.625***
Social Features 0.576***
Temporal Features 0.579***
Temporal + Social 0.638***
Temporal + Social + Text 0.669***
Temporal + Social + Text + Unigram 0.672***20
User Similarity?• User similarity (in terms of Interest and
Activity)had NO significant effect on giving.
• Did not included user similarity as a feature in the logistic regression model since authors were able to observe givers for a small subset.
21
Conclusion (1/2)• Language of requests matter a lot.
• Narratives of request also play a role in success.
• Reciprocity matters : Promise to pay it forward
• Pro-social behavior towards requestors who are of high status.
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 22
Conclusion (2/2)• Expressing gratitude is also vital.
• And finaly we can say:
Success is Predictable!
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests 23
24
THANK YOU!
How to Ask for a Favor : A Case Study on the Success of Altruistic Requests
25