Source-linked AI summary
How to Ask for a Favor: A Case Study on the Success of Altruistic Requests
Tim Althoff, Cristian Danescu-Niculescu-Mizil, Dan Jurafsky
TL;DR
The paper asks which textual and social factors are associated with successful requests, a question complicated by differences in what is requested and what givers receive. It studies altruistic requests for the same contribution, extracts social and narrative features, and finds several features associated with success while a model generalizes to unseen requests. The authors caution that the findings may be specific to this scenario and do not establish causality.
Problem
Request success can depend on what is requested, rewards, timing, and group dynamics, making textual and social factors difficult to disentangle.
Source: §1 · p. 1Method
The paper studies altruistic requests for the same contribution and uses textual and social features, including automatically detected narratives, to analyze and predict success.
Source: §1 · p. 1, §3.3 · p. 4Results
Narratives, gratitude, evidentiality, reciprocity, and asker status are associated with greater success, while politeness and positive sentiment are not associated with success in this setting.
Source: §7 · p. 8Takeaways & Limitations
The findings offer directions for understanding prosocial behavior in communities and a basis for further analysis of success in social media systems.
Source: §7 · p. 9As a case study, the findings may be specific to this scenario, and the authors cannot claim that the studied factors cause success.
Source: §7 · p. 9
Abstract
from arXiv · showhide
Requests are at the core of many social media systems such as question & answer sites and online philanthropy communities. While the success of such requests is critical to the success of the community, the factors that lead community members to satisfy a request are largely unknown. Success of a request depends on factors like who is asking, how they are asking, when are they asking, and most critically what is being requested, ranging from small favors to substantial monetary donations. We present a case study of altruistic requests in an online community where all requests ask for the very same contribution and do not offer anything tangible in return, allowing us to disentangle what is requested from textual and social factors. Drawing from social psychology literature, we extract high-level social features from text that operationalize social relations between recipient and donor and demonstrate that these extracted relations are predictive of success. More specifically, we find that clearly communicating need through the narrative is essential and that that linguistic indications of gratitude, evidentiality, and generalized reciprocity, as well as high status of the asker further increase the likelihood of success. Building on this understanding, we develop a model that can predict the success of unseen requests, significantly improving over several baselines. We link these findings to research in psychology on helping behavior, providing a basis for further analysis of success in social media systems.
1 Introduction
The paper studies which textual and social factors are associated with successful altruistic requests, using a setting that controls several common confounds. It finds that narratives and signals of gratitude, reciprocity, evidentiality, and asker status relate to success, and that a model generalizes to unseen requests.
1 Introduction
- Because request success depends on what is asked, rewards, timing, and group dynamics, these factors make it difficult to isolate why some requests are satisfied.The paper frames understanding successful requests as relevant to improving online communities and to questions in social psychology and linguistic pragmatics.Source: §1 · p. 1
- The study focuses on altruistic requests with no tangible reward, in one community where every request seeks the same contribution.It also focuses on requests a single user can fulfill, reducing group-behavior effects and helping isolate presentation factors.Source: §1 · pp. 1–2
- Narratives are significantly correlated with success, while gratitude, future reciprocity, supporting evidence, and high user status further increase the chance of success.The paper reports no support for associations between success and positive sentiment, politeness, or user similarity.Source: §1 · p. 2
- A prediction task shows that the proposed model generalizes to unseen requests and significantly improves over several baselines.The extracted social features operationalize the relation between recipient and donor.Source: §1 · p. 2
2 The Dataset
The dataset contains requests from Random Acts of Pizza and user histories from Reddit, with success labels restricted to requests whose outcomes can be determined.
2 The Dataset
- The analysis uses 5,728 requests from the 21,577-post Random Acts of Pizza history, after excluding users with multiple requests whose successful request could not be identified.The dataset spans December 8, 2010 to September 29, 2013.Source: §2 · p. 2
- The authors split requests into development and test sets at 70% and 30%, respectively, preserving the dataset’s 24.6% average success rate in both.Features were developed on the development set, and the test set was used once for held-out evaluation.Source: §2 · p. 2
- For 379 requests, beneficiary thank-you posts identify the benefactor, enabling analysis of user similarity and giving.User features were computed using 1.87 million submissions across Reddit histories of users involved in the community.Source: §2 · p. 2
3 Success Factors of Requests
The paper examines textual and social factors that may relate to request success, then uses automatically identified narratives to compare success across kinds of requests.
3.1 Textual Factors of Success
- The authors consider politeness, evidence for need, generalized reciprocity, sentiment, and request length as textual factors that could relate to success.They hypothesize that requests showing gratitude or an intention to pay help forward may encourage giving, while urgent evidence and request length may also matter.Source: §3.1 · pp. 2–3
3.2 Social Factors
- The social factors examined are user status and similarity, previously linked in related work to help and success in online settings.The authors predict that users will be more likely to give to someone like them, and that higher-status users may receive help more often.Source: §3.2 · p. 3
3.3 What Narratives Drive Success?
- The authors use non-negative matrix factorization on noun-only, TF-IDF request representations to identify 10 topics, then define five narrative lexicons for automatic detection.They use concise lexicons rather than fully unsupervised topic assignments because topic boundaries were often unclear and harder to interpret; word counts measure each narrative’s presence.Source: §3.3 · pp. 3–4
- Requests mentioning money or job problems had higher success rates than the 24.6% average, while friend-focused requests had a 17.0% success rate.MONEY1 had a 32.3% success rate and JOB had a 31.9% success rate; the authors suggest friend stories may communicate less need.Source: §3.3 · pp. 3–4
- The topic table reports each topic’s highest-scoring terms and success rate, with the dataset average at 24.6%.The displayed term lists include only the 15 highest-scoring terms per topic.Source: Table 1 · p. 4
4 What Factors Are Predictive of Success?
The section measures textual, social, and temporal factors and relates them to request success. Narrative, evidentiality, gratitude, reciprocity, length, and status are associated with success, while temporal patterns and sentiment show more qualified results.
4.1 Measuring the Factors
- The study measures temporal factors by calendar timing and community age, and operationalizes evidentiality and generalized reciprocity using image links and pay-forward phrases.Narratives are measured with lexicon word counts normalized by request length; status includes karma deciles and prior community participation.Source: §4.1 · pp. 4–5
- The analysis uses logistic regression to estimate success from textual, social, and temporal features, testing each factor while controlling for the others.The model uses a Likelihood Ratio test for significance; politeness is measured with 19 features, and sentiment with sentence-level and lexicon-based measures.Source: §4.1 · p. 4, §4.2 · p. 5
4.2 Results
- Daily and yearly seasonality do not differentiate outcomes, but success declines with community age; sentiment loses significance when other variables are controlled.Success was 36.6% in the first 10% of requests and 16.9% in the last 10%. Pay-it-forward claimants later reciprocated at 9.9%, versus a 5.9% baseline (p < 0.01).Source: §4.2 · pp. 5–6
- Gratitude, images, and linguistic willingness to give back are associated with success; images are attributed to communicating need, urgency, and trust.Only gratitude among the 19 politeness features is significantly correlated with success.Source: §4.2 · p. 6
- Longer requests and higher Reddit karma or prior community participation correlate with success, with prior participation having a particularly strong positive effect.The authors suggest longer requests allow more evidence about the requester’s situation.Source: §4.2 · p. 6
- Job, money, and family narratives increase predicted success, while craving has a strongly negative association; student does not significantly improve model fit.The authors interpret this pattern as support for narratives that clearly communicate need.Source: §4.2 · pp. 6–7, Table 3 · p. 6
4.3 Interpretation
- All logistic-regression feature-set models significantly outperform the random baseline (p < 0.001).The interpretation section also illustrates how estimated success probabilities differ across request formulations.Source: Table 4 · p. 7
- A 50-word craving-only request, assuming median karma and community age, has 9.8% estimated success; a 150-word request with picture evidence, gratitude, and a pay-forward offer reaches 56.8%.Source: §4.3 · p. 7
5 Is Success Predictable?
On held-out requests, combining textual, social, and temporal features gives the strongest reported model, while the concise text features capture nearly all the variance explained by simple text models. Social and temporal features also predict success, but the best model remains imperfect.
5 Is Success Predictable?
- The authors evaluate held-out predictions with ROC AUC because the dataset is unbalanced and prediction involves a trade-off between true- and false-positive rates.They use L1-penalized logistic regression across feature sets and compare against n-gram and random baselines.Source: §5 · p. 7
- 0.669 AUC for all three feature sets significantly improves over the temporal + social model (0.638; p < 0.01) and the textual model (p < 0.001).The model combines textual, social, and temporal features.Source: §5 · p. 8
- 0.638 AUC for temporal + social features significantly improves on temporal features alone (0.579; p < 0.001).Source: §5 · p. 8
- The 9-feature textual model is not significantly different from unigram, bigram, and trigram baselines, despite using orders of magnitude fewer features (DeLong’s test, p ≥ 0.612).Source: §5 · p. 8
- Adding a unigram model to the three feature sets does not significantly improve predictive accuracy (AUC 0.672; DeLong’s test, p = 0.348).Source: §5 · p. 8
- The best model reaches 0.669 AUC, and all models significantly outperform the random baseline (Mann–Whitney U test, p < 0.001).The authors note that the model is imperfect and omit factors observable only after a request, including responses, updates, and comments.Source: §5 · p. 8
6 Does User Similarity Increase Giving?
The authors test whether givers and requesters with more similar interests are more likely to be paired. Using subreddit activity as the measure of interests, they find no significant difference between actual and randomized pairs.
6 Does User Similarity Increase Giving?
- User similarity is measured from the sets of subreddits where each giver and requester had posted before the pizza request.The study uses intersection size and Jaccard similarity to compare the sets.Source: §6 · p. 8
- Random giver–receiver pairs serve as a null model, representing a random rewiring of the bipartite graph between givers and receivers.Source: §6 · p. 8
- Actual giver–receiver pairs do not have significantly different measured interest similarity from random pairs, using either intersection size or Jaccard similarity.The result is limited to similarity measured through users’ subreddit interests and activity.Source: §6 · p. 8
- The authors exclude user similarity from the logistic regression because givers are observed for only a small subset of requests.Source: §6 · p. 8
7 Conclusion
This case study holds the requested contribution constant at a pizza to examine textual and social factors associated with request success. The authors report associations with narratives, gratitude, evidentiality, reciprocity, and asker status, while noting limits on generalization and causal conclusions.
7 Conclusion
- Holding the requested contribution constant at a pizza lets the case study disentangle what is requested from textual and social factors.Source: §7 · p. 8
- Narratives and linguistic indications of gratitude, evidentiality, and reciprocity, as well as high asker status, are associated with greater request success; politeness and positive sentiment are not.The study extracts high-level social features from text to operationalize relations between recipients and donors.Source: §7 · p. 8
- The authors connect their findings to psychology research on helping behavior.Source: §7 · p. 8
- The findings may be specific to the case-study scenario, do not establish causality, and may omit factors such as group behavior dynamics.Source: §7 · p. 9
- The authors present the findings as a basis for further research on success factors and helping behavior in other online communities.Source: §7 · p. 9