Recent advances in general-purpose AI underscore the urgent need to align AI systems with human goals and values. Yet, the lack of a clear, shared understanding of what constitutes "alignment" limits meaningful progress and cross-disciplinary collaboration. In this position paper, we argue that the research community should explicitly define and critically reflect on "alignment" to account for the bidirectional and dynamic relationship between humans and AI. Through a systematic review of over 400 papers spanning HCI, NLP, ML, and more, we examine how alignment is currently defined and operationalized. Building on this analysis, we introduce the Bidirectional Human-AI Alignment framework, which not only incorporates traditional efforts to align AI with human values but also introduces the critical, underexplored dimension of aligning humans with AI – supporting cognitive, behavioral, and societal adaptation to rapidly advancing AI technologies. Our findings reveal significant gaps in current literature, especially in long-term interaction design, human value modeling, and mutual understanding. We conclude with three central challenges and actionable recommendations to guide future research toward more nuanced, reciprocal, and human-AI alignment approaches.

YouTube provides features for users to indicate disinterest when presented with unwanted recommendations, such as the “Not interested” and “Don’t recommend channel” buttons. These buttons are intended to allow the user to correct “mistakes” made by the recommendation system. Yet, relatively little is known about the empirical efficacy of these buttons. Neither is much known about users’ awareness of and confidence in them. To address these gaps, we simulated YouTube users with sock puppet agents. Each agent first executed a “stain phase”, where it watched many videos of an assigned topic; it then executed a “scrub phase”, where it tried to remove recommendations from the assigned topic. Each agent repeatedly applied a single scrubbing strategy, either indicating disinterest in one of the videos visited in the stain phase (disliking it or deleting it from the watch history), or indicating disinterest in a video recommended on the homepage (clicking the “not interested” or “don’t recommend channel” button or opening the video and clicking the dislike button). We found that the stain phase significantly increased the fraction of the recommended videos dedicated to the assigned topic on the user’s homepage. For the scrub phase, using the “Not interested” button worked best, significantly reducing such recommendations in all topics tested, on average removing 88% of them. Neither the stain phase nor the scrub phase, however, had much effect on videopage recommendations (those given to users while they watch a video). We also ran a survey (N = 300) asking adult YouTube users in the US whether they were aware of and used these buttons before, as well as how effective they found these buttons to be. We found that 44% of participants were not aware that the “Not interested” button existed. Those who were aware of it often used it to remove unwanted recommendations (82.8%) and found it to be modestly effective (3.42 out of 5).

In computational social science, researchers often use a pre-trained, black box classifier to estimate the frequency of each class in unlabeled datasets. A variety of prevalence estimation techniques have been developed in the literature, each yielding an unbiased estimate if certain stability assumption holds. This work introduces a framework to rethink the prevalence estimation process as calibrating the classifier outputs against ground truth labels to obtain the joint distribution of a base dataset and then extrapolating to the joint distribution of a target dataset. We call this framework "Calibrate-Extrapolate". It clarifies what stability assumptions must hold for a prevalence estimation technique to yield accurate estimates. In the calibration phase, the techniques assume only a stable calibration curve between a calibration dataset and the full base dataset. This allows for the classifier outputs to be used for disproportionate random sampling, thus improving the efficiency of calibration. In the extrapolation phase, some techniques assume a stable calibration curve while some assume stable class-conditional densities. We discuss the stability assumptions from a causal perspective. By specifying base and target joint distributions, we can generate simulated datasets, as a way to build intuitions about the impacts of assumption violations. This also leads to a better understanding of how the classifier's predictive power affects the accuracy of prevalence estimates: the greater the predictive power, the lower the sensitivity to violations of stability assumptions in the extrapolation phase. We illustrate the framework with an application that estimates the prevalence of toxic comments on news topics over time on Reddit, Twitter/X, and YouTube, using Jigsaw's Perspective API as a black box classifier. Finally, we summarize several practical advice for prevalence estimation.

Because high-quality data is like oxygen for AI systems, effectively eliciting information from crowdsourcing workers has become a first-order problem for developing high-performance machine learning algorithms. Two prevalent paradigms, spot-checking and peer prediction, enable the design of mechanisms to evaluate and incentivize high-quality data from human labelers. So far, at least three metrics have been proposed to compare the performances of these techniques \citepzhang2022high,gao2016incentivizing,burrell2021measurement. However, different metrics lead to divergent and even contradictory results in various contexts. In this paper, we harmonize these divergent stories, showing that two of these metrics are actually the same within certain contexts and explain the divergence of the third. Moreover, we unify these different contexts by introducingSpot Check Equivalence, which offers an interpretable metric for the effectiveness of a peer prediction mechanism. Finally, we present two approaches to compute spot check equivalence in various contexts, where simulation results verify the effectiveness of our proposed metric.

As content moderation becomes a central aspect of all social media platforms and online communities, interest has grown in how to make moderation decisions contestable. On social media platforms where individual communities moderate their own activities, the responsibility to address user appeals falls on volunteers from within the community. While there is a growing body of work devoted to understanding and supporting the volunteer moderators' workload, little is known about their practice of handling user appeals. Through a collaborative and iterative design process with Reddit moderators, we found that moderators spend considerable effort in investigating user ban appeals and desired to directly engage with users and retain their agency over each decision. To fulfill their needs, we designed and built AppealMod, a system that induces friction in the appeals process by asking users to provide additional information before their appeals are reviewed by human moderators. In addition to giving moderators more information, we expected the friction in the appeal process would lead to a selection effect among users, with many insincere and toxic appeals being abandoned before getting any attention from human moderators. To evaluate our system, we conducted a randomized field experiment in a Reddit community of over 29 million users that lasted for four months. As a result of the selection effect, moderators viewed only 30% of initial appeals and less than 10% of the toxically worded appeals; yet they granted roughly the same number of appeals when compared with the control group. Overall, our system is effective at reducing moderator workload and minimizing their exposure to toxic content while honoring their preference for direct engagement and agency in appeals.

To reduce the spread of misinformation, social media platforms may take enforcement actions against offending content, such as adding informational warning labels, reducing distribution, or removing content entirely. However, both their actions and their inactions have been controversial and plagued by allegations of partisan bias. When it comes to specific content items, surprisingly little is known about what ordinary people want the platforms to do. We provide empirical evidence about a politically balanced panel of lay raters' preferences for three potential platform actions on 368 news articles. Our results confirm that on many articles there is a lack of consensus on which actions to take. We find a clear hierarchy of perceived severity of actions with a majority of raters wanting informational labels on the most articles and removal on the fewest. There was no partisan difference in terms of how many articles deserve platform actions but conservatives did prefer somewhat more action on content from liberal sources, and vice versa. We also find that judgments about two holistic properties, misleadingness and harm, could serve as an effective proxy to determine what actions would be approved by a majority of raters.

The past decade in the US has been one of the most politically polarizing in recent memory. Ordinary Democrats and Republicans fundamentally dislike and distrust each other, even when they agree on policy issues. This increase in hostility towards opposing party supporters, commonly called affective polarization, has important ramifications that threaten democracy. Political science research suggests that at least part of this polarization stems from Democrats' misperceptions about Republicans' political views and vice-versa. Therefore, in this work, drawing on insights from political science and game studies research, we designed an online casual game that combines the relaxed, playful nonpartisan norms of casual games with corrective information about party supporters' political views that are often misperceived. Through an experiment, we found that playing the game significantly reduces negative feelings toward outparty supporters among Democrats, but not Republicans. It was also effective in improving willingness to talk politics with outparty supporters. Further, we identified psychological reactance as a potential mechanism that affects the effectiveness of depolarization interventions. Finally, our analyses suggest that the game versions with political content were rated to be just as fun to play as a game version without any political content suggesting that, contrary to popular belief, people do like to mix politics and play.

Spacing and procrastination are often thought of as opposites. It is possible, however, for a student to space their studying by doing something every day throughout the semester and still procrastinate by waiting until late in the semester to increase their amount of studying. To analyze the relationship between spacing and procrastination, we examined 674 students’ interactions with a course eBook over four semesters of an introductory programming course. We measured each student’s semester-level spacing as the number of days they interacted with the eBook, and each student’s semester-level procrastination as the average delay from the start of the semester for all their eBook interactions. Surprisingly, there was a small, yet positive, correlation between the two measures. Which, then, matters for course performance: studying over more days or studying earlier in the semester? When controlling for total amount of studying, as well as a number of academic and demographic characteristics in an SEM analysis, we find a strong positive effect of spacing but no significant effect of procrastination on final exam scores.

Generating multiple-choice questions is known to improve students' critical thinking and deep learning. Visualizing relationships between concepts enhances meaningful learning, students' ability to relate new concepts to previously learned concepts. We designed and deployed a collaborative learning process through which students generate multiple-choice questions and represent the prerequisite knowledge structure between questions as visual links in a shared map, using a variation of Concept Maps that we call "QMap." We conducted a four-month study with 19 undergraduate students. Students sustained voluntary contributions, creating 992 good questions, and drawing 1,255 meaningful links between the questions. Through analyzing self-reports, observations, and usage data, we report on the technical and social design features that led students to sustain their motivation.

E-commerce sites have an incentive to encourage impulse buying, even when not in the consumer's best interest. This study investigates what features e-commerce sites use to encourage impulse buying and what tools consumers desire to curb their online spending. We present two studies: (1) a systematic content analysis of 200 top e-commerce websites in the U.S. and (2) a survey of online impulse buyers (N=151). From Study 1, we find that e-commerce sites contain multiple features that encourage impulsive buying, including those that lower perceived risks, leverage social influence, and enhance perceived proximity to the product. Conversely, from Study 2 we find that online impulse buyers want tools that (a) encourage deliberation and avoidance, (b) enforce spending limits and postponement, (c) increase checkout effort, (d) make costs more salient, and (e) reduce product desire. These findings inform the design of "friction'' technologies that help users make more deliberative consumer choices.

Bayesian statistical analysis has gained attention in recent years, including in HCI. The Bayesian approach has several advantages over traditional statistics, including producing results with more intuitive interpretations. Despite growing interest, few papers in CHI use Bayesian analysis. Existing tools to learn Bayesian statistics require significant time investment, making it difficult to casually explore Bayesian methods. Here, we present a tool that lowers the barrier to exploration: a set of R code templates that guide Bayesian novices through their first analysis. The templates are tailored to CHI, supporting analyses found to be most common in recent CHI papers. In a user study, we found that the templates were easy to understand and use. However, we found that participants without a statistical background were not confident in their use. Together our contributions provide a concise analysis tool and empirical results for understanding and addressing barriers to using Bayesian analysis in HCI.

Retrieval practice, spacing, and interleaving are known to enhance long-term learning and transfer, but reduce short-term performance. It can be difficult to get both students and instructors to use these techniques since they perceive them as impeding initial student learning. We leveraged user experience design and research techniques, including survey and participant observation, to improve the design of a practice tool during a semester of use in a large introductory Python programming course. In this paper, we describe the design features that made the tool effective for learning as well as motivating. These include requiring spacing by giving credit for each day that a student answered a minimum number of questions, adapting a spaced repetition algorithm to schedule topics rather than specific questions, providing a visual representation of the evolving schedule in order to support meta-cognition, and providing several gameful design elements. To assess effectiveness, we estimated a regression model: each hour spent using the practice tool over the course of a semester was associated with an increase in final exam grades of 1.04%, even after controlling for many potential confounds. To assess motivation, we report on the amount of practice tool use: 62 of the 193 students (32%) voluntarily used the tool more than the required 45 days. This provides evidence that the design of the tool successfully overcame the typically negative perceptions of retrieval practice, spacing, and interleaving.

In an introductory Python programming course intended for non-majors with little prior CS experience, with 85 male and 108 female students, we were able to capture electronic traces of students' studying and problem-solving. There was no significant difference in final exam scores by gender but we found that female students spent 12.1 more hours studying over the semester while male students on average earned 2.7 more points per hour of solving problem set questions over the first half of the semester. We were able to capture their learning behavior because students studied using the Runestone interactive textbook and completed weekly problem sets in the same platform for the first half of the semester. We analyzed these logs to determine three quantities for each student. One is study time, as measured by total use of Runestone outside of weekly assignments. The second is speed, as measured by the number of points students earned per hour working on problem sets. The third is earliness, as measured by how far before the deadlines they worked on weekly assignments. We conclude that male students were faster at completing problem sets early in the semester but that female students found an alternative pathway to success.

We introduce an adversarial method for producing high-recall explanations of neural text classifier decisions.Building on an existing architecture for extractive explanations via hard attention, we add an adversarial layer which scans the residual of the attention for remaining predictive signal.Motivated by the important domain of detecting personal attacks in social media comments, we additionally demonstrate the importance of manually setting a semantically appropriate "default" behavior for the model by explicitly manipulating its bias term.We develop a validation set of humanannotated personal attacks to evaluate the impact of these changes.

Popularity systems, like Twitter retweets, Reddit upvotes, and Pinterest pins have the potential to guide people toward posts that others liked. That, however, creates a feedback loop that reduces their informativeness: items marked as more popular get more attention, so that additional upvotes and retweets may simply reflect the increased attention and not independent information about the fraction of people that like the items. How much information remains? For example, how confident can we be that more people prefer item A to item B if item A had hundreds of upvotes on Reddit and item B had only a few? We investigate using an Internet game called GuessTheKarma that collects independent preference judgments (N=20,674) for 400 pairs of images, approximately 50 per pair. Unlike the rating systems that dominate social media services, GuessTheKarma is devoid of social and ranking effects that influence ratings. Overall, Reddit scores were not very good predictors of the true population preferences for items as measured by GuessTheKarma: the image with higher score was preferred by a majority of independent raters only 68% of the time. However, when one image had a low score and the other was one of the highest scoring in its subreddit, the higher scoring image was preferred nearly 90% of the time by the majority of independent raters. Similarly, Imgur view counts for the images were poor predictors except when there were orders of magnitude differences between the pairs. We conclude that popularity systems marked by feedback loops may convey a strong signal about population preferences, but only when comparing items that received vastly different popularity scores.

Effective communication is crucial for instructors and students in programming courses. However, communicating about code can be difficult --- particularly in asynchronous settings where an instructor authors an explanation meant to be read and understood by a student later on. Communicating about code is uniquely difficult for two reasons. First, because of the dichotomous nature of the explanation, which consists of fragments of code and natural language descriptions. Second, instructors' explanations of code often involve modifying code throughout their explanation. This paper introduces chat.codes, a new tool for creating guided explanations about code. chat.codes introduces two features that make it easier to communicate about code. First, it adds deictic code references that allows instructors to write messages that reference specific regions of code. Second, it tracks and summarizes code edits in-line with messages, allowing instructors to create explanations in stages. An evaluation showed that these features were beneficial for both instructors and students.

E-commerce designers must decide how many products to display at one time. Choice overload research has demonstrated the surprising finding that more choice is not necessarily better?selecting from larger choice sets can be more cognitively demanding and can result in lower levels of choice satisfaction. This research tests the choice overload effect in an e-commerce context and explores how the choice overload effect is influenced by an individual's tendency to maximize or satisfice decisions. We conducted an online experiment with 611 participants randomly assigned to select a gourmet chocolate bar from either 12, 24, 40, 50, 60, or 72 different options. Consistent with prior work, we find that maximizers are less satisfied with their product choice than satisficers. However, using Bayesian analysis, we find that it's unlikely that choice set size affects choice satisfaction by much, if at all. We discuss why the decision-making process may be different in e-commerce contexts than the physical settings used in previous choice overload experiments.

Consumers are turning to Facebook Groups to buy and sell with strangers in their local communities. This trend is counter-intuitive given Facebook's lack of conventional e-commerce features, such as sophisticated search engines and reputation systems. We interviewed 18 members of two Mom-to-Mom Facebook sale groups. Despite a lack of commerce tools, members perceived sale groups as an easy-to-use way to quickly and conveniently buy and sell. Most important to members was that the groups felt safe and trustworthy. Drawing on these insights, we contribute a novel framing, community commerce, which explains the trust mechanisms that enable transactions between strangers in some groups. Community commerce fosters trust through (a) exclusive membership to a closed group, (b) regulation and sanctioning of behavior at the admin, member, and group level, and (c) a shared group identity or perceived similarity (though, surprisingly, not through social bonding). We discuss how community commerce affords unique and sometimes superior trust assurances and propose design implications for platforms hoping to foster trust between members who buy, sell, or share amongst themselves.

Runestone is an open-source ebook platform designed to create and publish interactive computer science textbooks (See http://runestoneinteractive.org/). Runestone textbooks support programming within the browser, code visualizations, and a wide variety of practice activities, from multiple choice and fill-in-the-blank questions to Parsons Problems (drag-and-drop mixed-up code). The presenters have several years of experience developing and using ebooks for CS1, AP CS A, AP CSP, data structures, and web programming. Several studies have demonstrated good usability and positive learning and attitude impacts on students using these ebooks. Runestone ebooks can be customized to meet the needs of individual courses and teachers. The goal of this workshop is to help computer science teachers use and modify Runestone ebooks. The hands-on session will start by leading participants through use of the ebooks as if they were students. Participants will next create their own custom course of any existing ebook in the library and will use the instructor's dashboard to review student activity, modify the course, and grade students. Finally, participants will create their own assignments using Runestone's active learning components, which serves as a starting point for authoring their own content in Runestone. Laptop Required.

The Guardian ---the fifth most widely read online newspaper in the world as of 2014---changed conversations on its commenting platform by altering its design from non-threaded to single-level threaded in 2012. We studied this naturally occurring experiment to investigate the impact of conversation threading on user retention as mediated by several potential changes in conversation structure and style. Our analysis shows that the design change made new users significantly more likely to comment a second time, and that this increased stickiness is due in part to a higher fraction of comments receiving responses after the design change. In mediation analysis, other anticipated mechanisms such as reciprocal exchanges and comment civility did not help to explain users' decision to return to the commenting system; indeed, civility did not increase after the design change and reciprocity declined. These analyses show that even simple design choices can have a significant impact on news forums' stickiness. Further, they suggest that this influence is more powerfully shaped by affordances---the new system made responding easier---than by changes in users' attention to social norms of reciprocity or civility. This has an array of implications for designers.

Undergraduates interviewed about privacy concerns related to online data collection made apparently contradictory statements. The same issue could evoke concern or not in the span of an interview, sometimes even a single sentence. Drawing on dual-process theories from psychology, we argue that some of the apparent contradictions can be resolved if privacy concern is divided into two components we call intuitive concern, a "gut feeling," and considered concern, produced by a weighing of risks and benefits. Consistent with previous explanations of the so-called privacy paradox, we argue that people may express high considered concern when prompted, but in practice act on low intuitive concern without a considered assessment. We also suggest a new explanation: a considered assessment can override an intuitive assessment of high concern without eliminating it. Here, people may choose rationally to accept a privacy risk but still express intuitive concern when prompted.

Existing retrieval systems rely on a single active query to pull documents from the index. Relevance feedback may be used to iteratively refine the query, but only one query is active at a time. If the user's information need has multiple aspects, the query must represent the union of these aspects. We consider a new paradigm of retrieval where multiple queries are kept ``active'' simultaneously. In the presence of rate limits, the active queries take turns accessing the index to retrieve another ``page'' of results. Turns are assigned by a multi-armed bandit based on user feedback. This allows the system to explore which queries return more relevant results and to exploit the best ones. In empirical tests, query pools outperform solo, combined queries. Significant improvement is observed both when the subtopic queries are known in advance and when the queries are generated in a user-interactive process.

Walking and other forms of physical activity have many health benefits, but people often fail to follow through on their own goals of being more active. To address gaps in current understanding of how to design technology-supported physical activity interventions, we conducted a randomized field experiment of a commitment device: making public announcements. In a control condition, weekly commitments were kept private. In two treatment conditions, they were announced on Facebook and by email. In one of the two, the announcements also included results: whether the previous week's commitment was kept. We find that, with or without public results, these posts can elicit supportive replies from the poster's social networks. People in both public announcements conditions were less likely to make commitments. We conclude that the prospect of public accountability may suppress the making of commitments in a way that counteracts the benefits of that accountability. Designers will need to address this limitation in order to make effective use of public accountability as a commitment device.

Many previous techniques identify trending topics in social media, even topics that are not pre-defined. We present a technique to identify trending rumors, which we define as topics that include disputed factual claims. Putting aside any attempt to assess whether the rumors are true or false, it is valuable to identify trending rumors as early as possible. It is extremely difficult to accurately classify whether every individual post is or is not making a disputed factual claim. We are able to identify trending rumors by recasting the problem as finding entire clusters of posts whose topic is a disputed factual claim.

With a growing awareness of the importance of Human Computer Interaction (HCI) and User Experience (UX) Design and Research in industry, many departments are examining ways to integrate this topic into their curriculum. What are the best ways to do this? Should HCI and UX be integrated into existing courses such as event-driven programming or mobile computing? Or should it be a separate course that integrates visual design, psychology and ergonomics? How important is this topic to students looking for internships and entry-level jobs? At what levels should these topics be taught in the curriculum? This is important as there is a growing demand in industry for User Experience specialists due to an increased focus on software quality and product success. A major complaint of employers is that this skill is lacking in many college graduates. Teaching computing students how to incorporate UX into the design and development process helps to address this critical need. This BOF will let participants share their ideas and learn from their peers. Some of the discussion leaders led a similar lively and enthusiastic discussion at SIGCSE 2013. Discussion spanned the role of HCI in the participants? various curricula; modules versus dedicated courses; projects versus standalone assignments; and textbooks, tools, and other pedagogical materials. Bringing together new and previous participants will bring new insights and expansion of these, including discussions of our recent experiences and lessons learned, as well as the broaching of new topics such as the roles of non-computing courses in HCI education.

We consider a scenario where a searcher requires both high precision and high recall from an interactive retrieval process. Such scenarios are very common in real life, exemplified by medical search, legal search, market research, and literature review. When access to the entire data set is available, an active learning loop could be used to ask for additional relevance feedback labels in order to refine a classifier. When data is accessed via search services, however, only limited subsets of the corpus can be considered, subsets defined by queries. In that setting, relevance feedback has been used in a query enhancement loop that updates a query.

The CubeThat browser extension for Chrome displays recommended additional news stories related to the same topic as the current news story. The recommended stories are organized into clusters, and clusters that the user has already sampled from are grayed out, in order to encourage users to explore multiple aspects of a story. Users can also provide feedback to improve the clustering, by dragging stories from one cluster to another.

Two public display systems, with different methods of posting, were deployed over several years. One, the Thank You Board, was designed to give people an outlet specifically for publicly thanking and acknowledging others in the community. The other, SI Display, showed any Twitter post directed to the display and did not have explicit usage guidelines. People preferred the flexibility of the latter, but ambiguity about its purpose and norms of usage persisted even six months after deployment and made some people hesitant to post. Also, using Twitter as the posting mechanism facilitated participation for some but also created barriers for those not using Twitter and for Twitter users who were wary of mixing their professional and non-professional contexts.

To understand why and how people share health information online, we interviewed fourteen people with significant health concerns who participate in both online health communities and Facebook. Qualitative analysis of these interviews highlighted the ways that people think about with whom and how to share different types of information as they pursue social goals related to their personal health, including emotional support, motivation, accountability, and advice. Our study suggests that success in these goals depends on how well they develop their social networks and how effectively they communicate within those networks. Effective communication is made more challenging by the need to strike a balance between sharing information related to specific needs and the desire to manage self-presentation. Based on these observations, we outline a set of design opportunities for future systems to support health-oriented social interactions online, including tools to help users shape their social networks and communicate effectively within those.

Is a polarized society inevitable, where people choose to be exposed to only political news and commentary that reinforces their existing viewpoints? We examine the relationship between the numbers of supporting and challenging items in a collection of political opinion items and readers' satisfaction, and then evaluate whether simple presentation techniques such as highlighting agreeable items or showing them first can increase satisfaction when fewer agreeable items are present. We find individual differences: some people are diversity-seeking while others are challenge-averse. For challenge-averse readers, highlighting appears to make satisfaction with sets of mostly agreeable items more extreme, but does not increase satisfaction overall, and sorting agreeable content first appears to decrease satisfaction rather than increasing it. These findings have important implications for builders of websites that aggregate content reflecting different positions.

Contests and challenges have energized researchers and focused attention in many fields recently, including recommender systems. At the 2008 RecSys conference, winners were announced for a contest proposing new startup companies. The 2009 conference featured a panel reflecting on the then recently completed Netflix challenge.

Conversation double pivots recommend target items related to a source item, based on co-mentions of source and target items in online forums. We deployed several variants on the drupal.org site that supports the Drupal open source community, and assessed them through clickthrough rates. A similarity metric based on correlation of mentions rather than mere co-occurrence reduced the problem of over-recommending the most popular modules, but additional corrections for recency and uniqueness of mentions were not helpful. Detection of more module mentions in conversations dramatically improved the quality of recommendations, even though the detection algorithm then had more false positives. Recommendations based on conversation co-mention were more effective than those based on co-installation, because co-installation data only led to recommendations of complementary modules and not substitutes. Recommendations based on co-mention were more effective than those based on text similarity matching for navigating from the most popular modules, but less effective than text matching for less popular modules.

We study protocols to enable one user (the principal) to make potentially profitable but risky interactions with another user (the agent), in the absence of direct trust between the two parties. In such situations, it is possible to enable the interaction indirectly through a chain of credit or "trust" links. We introduce a model that provides insight into many disparate applications, including open currency systems, network trust aggregation systems, and manipulation-resistant recommender systems. Each party maintains a trust account for each other party. When a principal's trust balance for an agent is high enough to cover potential losses from a bad interaction, direct trust is sufficient to enable the interaction. Allowing indirect trust opens up more interaction opportunities, but also expands the strategy space of an attacker seeking to exploit the community for its own ends. We show that with indirect trust exchange protocols, some friction is unavoidable: any protocol that satisfies a natural strategic safety property that we call sum-sybilproofness can sometimes lead to a reduction in expected overall trust balances even on interactions that are profitable in expectation. Thus, for long-term growth of trust accounts, which are assets enabling risky but valuable interactions, it may be necessary to limit the use of indirect trust. We present the hedged-transitive protocol and show that it achieves the optimal rate of expected growth in trust accounts, among all protocols satisfying the sum-sybilproofness condition.

Many sites on the web offer collaborative databases that catalog items such as bands, events, products, or software modules. Conversation pivots allow readers to navigate from pages about these items to conversations about them on the same site or elsewhere on the Internet. Double pivots allow readers to navigate from item pages to pages about other items mentioned in the same conversations. Using text mining techniques specific to the collection it is possible to find references to collected items in online conversations. We implemented conversation pivots for the CPAN archive of Perl modules, and for Drupal.org, the reference site for the Drupal content management system.

Attackers may seek to manipulate recommender systems in order to promote or suppress certain items. Existing defenses based on analysis of ratings also discard useful information from honest raters. In this paper, we show that this is unavoidable and provide a lower bound on how much information must be discarded. We use an information-theoretic framework to exhibit a fundamental tradeoff between manipulation-resistance and optimal use of genuine ratings in recommender systems. We define a recommender system to be (n, c)-robust if an attacker with n sybil identities cannot cause more than a limited amount c units of damage to predictions. We prove that any robust recommender system must also discard Ω(log (n/c)) units of useful information from each genuine rater.

Large-scale online communities need to manage the tension between critical mass and information overload. Slashdot is a news and discussion site that has used comment rating to allow massive participation while providing a mechanism for users to filter content. By default, comments with low ratings are hidden. Of users who changed the defaults, more than three times as many chose to use ratings for filtering or sorting as chose to suppress the use of comment ratings. Nearly half of registered users, however, never strayed from the default filtering settings, suggesting that the costs of exploring and selecting custom filter settings exceeds the expected benefit for many users. We recommend leveraging the efforts of the users that actively choose filter settings to reduce the cost of changing settings for all other users. One strategy is to create static schemas that capture the filtering preferences of different groups of readers. Another strategy is to dynamically set filtering thresholds for each conversation thread, based in part on the choices of previous readers. For predicting later readers' choices, the choices of previous readers are far more useful than content features such as the number of comments or the ratings of those comments.

An attacker can draw attention to items that don't deserve that attention by manipulating recommender systems. We describe an influence-limiting algorithm that can turn existing recommender systems into manipulation-resistant systems. Honest reporting is the optimal strategy for raters who wish to maximize their influence. If an attacker can create only a bounded number of shills, the attacker can mislead only a small amount. However, the system eventually makes full use of information from honest, informative raters. We describe both the influence limits and the information loss incurred due to those limits in terms of information-theoretic concepts of loss functions and entropies.

One of the important challenges faced by designers of online communities is eliciting sufficent contributions from community members. Users in online communities may have difficulty either in finding opportunities to add value, or in understanding the value of their contributions to the community. Various social science theories suggest that showing users different perspectives on the value they add to the community will lead to differing amounts of contribution. The present study investigates a design augmentation for an existing community Web site that could benefit from additional contribution. The augmented interface includes individualized opportunities for contribution and an estimate of the value of each contribution to the community. The value is computed in one of four different ways: (1) value to self; (2) value to a small group the user has affinity with; (3) value to a small group the user does not have affinity with; and (4) value to the entire user community. The study compares the effectiveness of the different notions of value to 160 community members.

The RideNow Project is designed to help individuals within a group or organization coordinate ad hoc shared rides. This paper describes three design decisions the RideNow team made in order to allow incremental adoption and evolution and to capitalize on local conditions. (1) The system allows users to interact with the system through email or Web, because we anticipate that email will be most convenient when there are few users but the Web interface will be more useful as the number of users increase. (2) The system does not force structure on user-entered data such as dates, times, and locations, instead allowing conventions to emerge. (3) We use the group's shared physical spaces to provide additional information about ride sharing activity.

Analysis of usage history for a large panel of eBay sellers suggests that both seller and buyer behavior change in response to changes in a seller's feedback profile. Sellers are more likely to stop listing items right after receiving a negative feedback. Sellers who continue listing do not seem to improve their performance in order to salvage their reputations. Instead, sellers get more negative feedback after receiving a negative feedback. One reason is that observed negative feedback appears to be symptomatic of a temporary decline in the seller's quality, which is also reflected in other transactions around the same time. Receipt of negative feedback might also cause a decline in seller quality, but we find only weak evidence of that. Empirical evidence does support a second hypothesis, that buyers appear to be more willing to give negative feedback to sellers who have recently received other negative feedback.

Can a system of distributed moderation quickly and consistently separate high and low quality comments in an online conversation? Analysis of the site Slashdot.org suggests that the answer is a qualified yes, but that important challenges remain for designers of such systems. Thousands of users act as moderators. Final scores for comments are reasonably dispersed and the community generally agrees that moderations are fair. On the other hand, much of a conversation can pass before the best and worst comments are identified. Of those moderations that were judged unfair, only about half were subsequently counterbalanced by a moderation in the other direction. And comments with low scores, not at top-level, or posted late in a conversation were more likely to be overlooked by moderators.

CSCW as a field has been driven primarily by researchers' desire to solve real world problems of groups and organizations, and to use new technology to solve these problems. The field has accumulated a set of empirically-based interdisciplinary studies and many interesting new applications. The question to be addressed in this panel is whether CSCW as a field is ready for theory--whether theory is needed to move the field along, or on the contrary, whether the problems and the technology are still too new or are changing too fast to accommodate theory. The panelists will describe some of the organization theories that could be applied to CSCW, and debate their usefulness, taking both sides of the question.

Under-contribution is a problem for many online communities. Social psychology theories of social loafing and goal-setting can provide mid-level design principles to address this problem. We tested the design principles in two field experiments. In one, members of an online movie recommender community were reminded of the uniqueness of their contributions and the benefits that follow from them. In the second, they were given a range of individual or group goals for contribution. As predicted by theory, individuals contributed when they were reminded of their uniqueness and when they were given specific and challenging goals, but other predictions were not borne out. The paper ends with suggestions and challenges for mining social science theories as well as implications for design.

No abstract available.

Menus, lists, and forms are the workhorse dialogue structures in telephone-based interactive voice response applications. Despite diversity in applications, there is a surprising homogeneity in the menu, list, and form styles commonly employed. There are, however, many alternatives, and no single style fits every prospective application and user population. A design space for each dialogue structure organizes the alternatives and provides a framework for analyzing their benefits and drawbacks. In addition to phone-based interactions, the design spaces apply to any limited-bandwidth, temporally constrained display devices, including small-screen devices such as personal digital assistants (PDAs) and screen phones.

Collaborative filters help people make choices based on the opinions of other people. GroupLens is a system for collaborative filtering of netnews, to help people find articles they will like in the huge stream of available articles. News reader clients display predicted scores and make it easy for users to rate articles after they read them. Rating servers, called Better Bit Bureaus, gather and disseminate the ratings. The rating servers predict scores based on the heuristic that people who agreed in the past will probably agree again. Users can protect their privacy by entering ratings under pseudonyms, without reducing the effectiveness of the score prediction. The entire architecture is open: alternative software for news clients and Better Bit Bureaus can be developed independently and can interoperate with the components we have developed.

Telephones are the most ubiquitous, best-networked, and simplest computer terminals available today. They have been used for voice mail but largely overlooked as a platform for asynchronous cooperative-work applications such as event calendars, issue discussions, and question-and-answer gathering. HyperVoice is a software toolkit for constructing such applications. Its building blocks are high-level presentation formats for collections of structured voice messages. The presentation formats can themselves be presented and manipulated, enabling significant customization of applications by phone. Results of two field trials suggest social-context factors that will influence the success or failure of phone-based cooperative work applications in particular settings.

The current generation of telephone interfaces is frustrating to use, in part because callers have to wait through the recitation of long prompts in order to find the options that interest them. In a visual medium, users would shift their gaze in order to skip uninteresting prompts and scan through large pieces of text. We present skip and scan, a new telephone interface style in which callers issue explicit commands to accomplish these same skipping and scanning activities. In a laboratory experiment, subjects made selections using skip and scan menus more quickly than using traditional, numbered menus, and preferred the skip and scan menus in subjective ratings. In a field test of a skip and scan interface, the general public successfully added and retrieved information without using any written instructions.

A major shift is underway in how we think about telephones.For decades, they were used solely for oneto-one, synchronous communication.The increasing use of answering machines and voice messaging, however, is shifting the public perception of telephones, thus opening a space for more innovative applications.Five years from now, some of the most interesting and popular cooperative work applications will probably use telephones as the primary means of access.This paper presents evidence that there are practical phone-based cooperative work applications and describes a set of software tools that facilitate the development of such applications..,. ,.