Constitutional AI (CAI) guides LLM behavior using constitutions, but identifying which principles are most effective for model alignment remains an open challenge.We introduce the C3AI framework (Crafting Constitutions for CAI models), which serves two key functions: (1) selecting and structuring principles to form effective constitutions before fine-tuning; and (2) evaluating whether finetuned CAI models follow these principles in practice.By analyzing principles from AI and psychology, we found that positively framed, behavior-based principles align more closely with human preferences than negatively framed or trait-based principles.In a safety alignment use case, we applied a graph-based principle selection method to refine an existing CAI constitution, improving safety measures while maintaining strong general reasoning capabilities.Interestingly, fine-tuned CAI models performed well on negatively framed principles but struggled with positively framed ones, in contrast to our human alignment results.This highlights a potential gap between principle design and model adherence.Overall, C3AI provides a structured and scalable approach to both crafting and evaluating CAI constitutions. CCS Concepts•

Misinformation proliferates in the online sphere, with evident impacts on the political and social realms, influencing democratic discourse and posing risks to public health and safety. The corporate world is also a prime target for fake news dissemination. While recent studies have attempted to characterize corporate misinformation and its effects on companies, their findings often suffer from limitations due to qualitative or narrative approaches and a narrow focus on specific industries. To address this gap, we conducted an analysis utilizing social media quantitative methods and crowd-sourcing studies to investigate corporate misinformation across a diverse array of industries within the S&P 500 companies. Our study reveals that corporate misinformation encompasses topics such as products, politics, and societal issues. We discovered companies affected by fake news also get reputable news coverage but less social media attention, leading to heightened negativity in social media comments, diminished stock growth, and increased stress mentions among employee reviews. Additionally, we observe that a company is not targeted by fake news all the time, but there are particular times when a critical mass of fake news emerges. These findings hold significant implications for regulators, business leaders, and investors, emphasizing the necessity to vigilantly monitor the escalating phenomenon of corporate misinformation.

Self-supervised learning (SSL) has become the de facto training paradigm of large models, where pre-training is followed by supervised fine-tuning using domain-specific data and labels. Despite demonstrating comparable performance with supervised methods, comprehensive efforts to assess SSL's impact on machine learning fairness (i.e., performing equally on different demographic breakdowns) are lacking. Hypothesizing that SSL models would learn more generic, hence less biased representations, this study explores the impact of pre-training and fine-tuning strategies on fairness. We introduce a fairness assessment framework for SSL, comprising five stages: defining dataset requirements, pre-training, fine-tuning with gradual unfreezing, assessing representation similarity conditioned on demographics, and establishing domain-specific evaluation processes. We evaluate our method's generalizability on three real-world human-centric datasets (i.e., MIMIC, MESA, and GLOBEM) by systematically comparing hundreds of SSL and fine-tuned models on various dimensions spanning from the intermediate representations to appropriate evaluation metrics. Our findings demonstrate that SSL can significantly improve model fairness, while maintaining performance on par with supervised methods-exhibiting up to a 30% increase in fairness with minimal loss in performance through self-supervision. We posit that such differences can be attributed to representation dissimilarities found between the best- and the worst-performing demographics across models-up to x13 greater for protected attributes with larger performance discrepancies between segments. Code: https://github.com/Nokia-Bell-Labs/SSLfairness

Parks are essential spaces for promoting urban health, and recommender systems could assist individuals in discovering parks for leisure and health-promoting activities. This is particularly important in large cities like London, which has over 1,500 named parks, making it challenging to understand what each park offers. Due to the lack of datasets and the diverse health-promoting activities parks can support (e.g., physical, social, nature-appreciation), it is unclear which recommendation algorithms are best suited for this task. To explore the dynamics of recommending parks for specific activities, we created two datasets: one from a survey of over 250 London residents, and another by inferring visits from over 1 million geotagged Flickr images taken in London parks. Analyzing the geographic patterns of these visits revealed that recommending nearby parks is ineffective, suggesting that this recommendation task is distinct from Point of Interest recommendation. We then tested various recommendation models, identifying a significant popularity bias in the results. Additionally, we found that personalized models have advantages in recommending parks beyond the most popular ones. The data and findings from this study provide a foundation for future research on park recommendations.

Circadian rhythm has been linked to both physical and mental health at an individual level in prior research. Such a link at population level has been long hypothesized but has never been tested, largely because of lack of data. To partly fix this literature gap, we need: a dataset on population-level circadian rhythms, a dataset on population-level health conditions, and strong associations between these two partly independent sets. Recent work has shown that affect on social media data relates to population-level circadian rhythms. Building upon that work, we extracted five circadian rhythm metrics from 6M Reddit posts across 18 major cities (for which the number of residents is highly correlated with the number of users), and paired them with three ground-truth health metrics (daily number of steps, sleep quantity, and sleep quality) extracted from 233K wearable users in these cities. We found that rhythms of online activity approximated sleeping patterns rather than, what the literature previously hypothesized, alertness levels. Despite that, we found that these rhythms, when computed in two specific times of the day (i.e., late at night and early morning), were still predictive of the three ground-truth health metrics: in general, healthier cities had morning spikes on social media, night dips, and expressions of positive affect. These results suggest that circadian rhythms on social media, if taken at two specific times of the day and operationalized with literature-driven metrics, can approximate the temporal evolution of people's shared underlying biological rhythm as it relates to physical activity (R2=0.492), sleep quantity (R2=0.765), and sleep quality (R2=0.624).

Decades of social science research identified ten fundamental dimensions that provide the conceptual building blocks to describe the nature of human relationships. Yet, it is not clear to what extent these concepts are expressed in everyday language and what role they have in shaping observable dynamics of social interactions. After annotating conversational text through crowdsourcing, we trained NLP tools to detect the presence of these types of interaction from conversations, and applied them to 160M messages written by geo-referenced Reddit users, 290k emails from the Enron corpus and 300k lines of dialogue from movie scripts. We show that social dimensions can be predicted purely from conversations with an AUC up to 0.98, and that the combination of the predicted dimensions suggests both the types of relationships people entertain (conflict vs. support) and the types of real-world communities (wealthy vs. deprived) they shape.

Airbnb is one of the most successful examples of sharing economy marketplaces. With rapid and global market penetration, understanding its attractiveness and evolving growth opportunities is key to plan business decision making. There is an ongoing debate, for example, about whether Airbnb is a hospitality service that fosters social exchanges between hosts and guests, as the sharing economy manifesto originally stated, or whether it is (or is evolving into being) a purely business transaction platform, the way hotels have traditionally operated. To answer these questions, we propose a novel market analysis approach that exploits customers’ reviews. Key to the approach is a method that combines thematic analysis and machine learning to inductively develop a custom dictionary for guests’ reviews. Based on this dictionary, we then use quantitative linguistic analysis on a corpus of 3.2 million reviews collected in 6 different cities, and illustrate how to answer a variety of market research questions, at fine levels of temporal, thematic, user and spatial granularity, such as (i) how the business vs social dichotomy is evolving over the years, (ii) what exact words within such top-level categories are evolving, (iii) whether such trends vary across different user segments and (iv) in different neighbourhoods.

To walk around the city, individuals use mobile mapping services, and such services mostly suggest shortest routes. To go beyond recommending such walkable routes, we propose a new framework for automatic wayfinding for pedestrians. This framework tackles two main drawbacks from which past work suffers, namely coarse-grained representation of space and absence of contextual dynamics. We model the human tendency to regularize space by borrowing a spatial representation, Space Syntax, from the discipline of Architecture. Moreover, the proposed framework accounts for contextual dynamics of individual streets by predicting the popularity of each street under different contexts (e.g., at a given time, with a certain weather condition). Using Foursquare check-ins (i.e., whereabouts of the users of the popular location-based service) and publicly available weather data, we validate our framework in the entire city of Barcelona. We find that, with paths slightly longer than the shortest ones, our framework is able to accommodate our mental topography and effectively capture contextual changes.

The Death and Life of Great American Cities was written in 1961 and is now one of the most influential book in city planning. In it, Jane Jacobs proposed four conditions that promote life in a city. However, these conditions have not been empirically tested until recently. This is mainly because it is hard to collect data about "city life". The city of Seoul recently collected pedestrian activity through surveys at an unprecedented scale, with an effort spanning more than a decade, allowing researchers to conduct the first study successfully testing Jacobs's conditions.

Sharing economy platforms have become extremely popular in the last few years, and they have changed the way in which we commute, travel, and borrow among many other activities. Despite their popularity among consumers, such companies are poorly regulated. For example, Airbnb, one of the most successful examples of sharing economy platform, is often criticized by regulators and policy makers. While, in theory, municipalities should regulate the emergence of Airbnb through evidence-based policy making, in practice, they engage in a false dichotomy: some municipalities allow the business without imposing any regulation, while others ban it altogether. That is because there is no evidence upon which to draft policies. Here we propose to gather evidence from the Web. After crawling Airbnb data for the entire city of London, we find out where and when Airbnb listings are offered and, by matching such listing information with census and hotel data, we determine the socio-economic conditions of the areas that actually benefit from the hospitality platform. The reality is more nuanced than one would expect, and it has changed over the years. Airbnb demand and offering have changed over time, and traditional regulations have not been able to respond to those changes. That is why, finally, we rely on our data analysis to envision regulations that are responsive to real-time demands, contributing to the emerging idea of ``algorithmic regulation''.

Community-based question answering platforms can be rich sources of information on a variety of specialized topics, from finance to cooking. The usefulness of such platforms depends heavily on user contributions (questions and answers), but also on respecting the community rules. As a crowd-sourced service, such platforms rely on their users for monitoring and flagging content that violates community rules. Common wisdom is to eliminate the users who receive many flags. Our analysis of a year of traces from a mature Q&A site shows that the number of flags does not tell the full story: on one hand, users with many flags may still contribute positively to the community. On the other hand, users who never get flagged are found to violate community rules and get their accounts suspended. This analysis, however, also shows that abusive users are betrayed by their network properties: we find strong evidence of homophilous behavior and use this finding to detect abusive users who go under the community radar. Based on our empirical observations, we build a classifier that is able to detect abusive users with an accuracy as high as 83%.

Walkability has many health, environmental, and economic benefits. That is why web and mobile services have been offering ways of computing walkability scores of individual street segments. Those scores are generally computed from survey data and manual counting (of even trees). However, that is costly, owing to the high time, effort, and financial costs. To partly automate the computation of those scores, we explore the possibility of using the social media data of Flickr and Foursquare to automatically identify safe and walkable streets. We find that unsafe streets tend to be photographed during the day, while walkable streets are tagged with walkability-related keywords. These results open up practical opportunities (for, e.g., room booking services, urban route recommenders, and real-estate sites) and have theoretical implications for researchers who might resort to the use social media data to tackle previously unanswered questions in the area of walkability.

Urban crowd-sourcing has become a popular paradigm to harvest spatial information about our evolving cities directly from citizens. OpenStreetMap is a successful example of such paradigm, with an accuracy of its user-generated content comparable to that of curated databases (e.g., Ordnance Survey). Coverage is however low and most importantly non-uniformly distributed across the city. Being able to model the spontaneous growth of digital information in these domains is required, so to be able to plan interventions aimed at gathering content about areas that would otherwise be neglected. Inspired by models of physical urban growth developed by urban planners, we build a model of digital growth of crowd-sourced spatial information that is both easy to interpret and dynamic, so to be able to determine what factors impact growth and how these change over time. We build and test the model against five years of OpenStreetMap data for the city of London, UK. We then run the model against two other cities, chosen for their different physical and digital growth's characteristics, so to stress-test the model. We conclude with a discussion of the implications of this work on both developers and users of urban crowd-sourcing applications.

To bring their innovative ideas to market, those embarking in new ventures have to raise money, and, to do so, they have often resorted to banks and venture capitalists. Nowadays, they have an additional option: that of crowdfunding. The name refers to the idea that funds come from a network of people on the Internet who are passionate about supporting others' projects. One of the most popular crowdfunding sites is Kickstarter. In it, creators post descriptions of their projects and advertise them on social media sites (mainly Twitter), while investors look for projects to support. The most common reason for project failure is the inability of founders to connect with a sufficient number of investors, and that is mainly because hitherto there has not been any automatic way of matching creators and investors. We thus set out to propose different ways of recommending investors found on Twitter for specific Kickstarter projects. We do so by conducting hypothesis-driven analyses of pledging behavior and translate the corresponding findings into different recommendation strategies. The best strategy achieves, on average, 84% of accuracy in predicting a list of potential investors' Twitter accounts for any given project. Our findings also produced key insights about the whys and wherefores of investors deciding to support innovative efforts.

Planners and social psychologists have suggested that the recognizability of the urban environment is linked to people's socio-economic well-being. We build a web game that puts the recognizability of London's streets to the test. It follows as closely as possible one experiment done by Stanley Milgram in 1972. The game picks up random locations from Google Street View and tests users to see if they can judge the location in terms of closest subway station, borough, or region. Each participant dedicates only few minutes to the task (as opposed to 90 minutes in Milgram's). We collect data from 2,255 participants (one order of magnitude a larger sample) and build a recognizability map of London based on their responses. We find that some boroughs have little cognitive representation; that recognizability of an area is explained partly by its exposure to Flickr and Foursquare users and mostly by its exposure to subway passengers; and that areas with low recognizability do not fare any worse on the economic indicators of income, education, and employment, but they do significantly suffer from social problems of housing deprivation, poor living conditions, and crime. These results could not have been produced without analyzing life off- and online: that is, without considering the interactions between urban places in the physical world and their virtual presence on platforms such as Flickr and Foursquare. This line of work is at the crossroad of two emerging themes in computing research - a crossroad where "web science" meets the "smart city" agenda.

Mobile, local recommendations are on the rise. Surprisingly however, research addressing user perceptions of local recommendations and local differences when interacting with such recommendation services is yet scarce. Location-based recommendation services are mostly evaluated from a `recommendation systems' standpoint, with limited experiential insights from users and limited focus on local differences that may apply. This workshop focuses on the local, personal user experience, and provides a forum to exchange experiences, insights and strategies in personalizing local mobile applications and generating local recommendations that fit local user needs.

Social-networking sites have started to offer tools that suggest "guests" who should be invited to user-defined social events (e.g., birthday parties, networking events). The problem of how to recommend people to events is similar to the more traditional (recommender system) problem of how to recommend events (items) to people (users). Yet, upon Foursquare data of "who visits what" in the city of London, we show that a state-of-the-art recommender system does not perform well -mainly because of data sparsity. To fix this problem, we add domain knowledge to the recommendation process. From the complex system literature in human mobility, we learn two insights: 1) there are special individuals (often called power users) who visit many places; and 2) individuals go to a venue not only because they like it but also because they are close-by. We model these insights into two simple models and learn that: 1) simply recommending power users works better than random but is far from producing the best recommendations; 2) an item-based recommender system produces accurate recommendations; and 3) recommending places that are closest to a user's geographic center of interest produces recommendations that are as accurate as, if not more accurate than, item-based recommender's. This last result has practical implications as it offers guidelines for designing location-based recommender systems and for partly addressing cold-start situations.

Social media sites have used recommender systems to suggest items users might like but are not already familiar with. These items are typically movies, books, pictures, or songs. Here we consider an alternative class of items - pictures posted by design-conscious individuals. We do so in the context of a mobile application in which users find "cool" items in the real world, take pictures of them, and share those pictures online. In this context, temporal dynamics matter, and users would greatly profit from ways of identifying the latest design trends. We propose a new way of recommending trending pictures to users, which unfolds in three steps. First, two types of users are identified - those who are good at uploading trends (trend makers) and those who are experienced in discovering trends (trend spotters). Second, based on what those "special few" have uploaded and rated, trends are identified early on. Third, trends are recommended using existing algorithms. Upon the complete longitudinal dataset of the mobile application, we compare our approach's performance to a traditional recommender system's.

Recommendation systems exist to help users discover content in a large body of items. An ideal recommendation system should mimic the actions of a trusted friend or expert, producing a personalised collection of recommendations that balance between the desired goals of accuracy, diversity, novelty and serendipity. We introduce the Auralist recommendation framework, a system that - in contrast to previous work - attempts to balance and improve all four factors simultaneously. Using a collection of novel algorithms inspired by principles of "serendipitous discovery", we demonstrate a method of successfully injecting serendipity, novelty and diversity into recommendations whilst limiting the impact on accuracy. We evaluate Auralist quantitatively over a broad set of metrics and, with a user study on music recommendation, show that Auralist's emphasis on serendipity indeed improves user satisfaction.

The rise of location-enabled mobile phones and location based services offers a great opportunity to apply personalization and recommender system technology to people's everyday lives. A variety of digital traces can now be used to infer how people move about their city and extract their context and habits. Personalization and recommender systems, potentially merged with the data that people store online (e.g., social networks, web ratings), can then not only be used to recommend new places and events that they may find interesting to attend, but, more broadly, personalize and enhance any service that people find themselves using.

A city offers thousands of social events a day, and it is difficult for dwellers to make choices. The combination of mobile phones and recommender systems can change the way one deals with such abundance. Mobile phones with positioning technology are now widely available, making it easy for people to broadcast their whereabouts, recommender systems can now identify patterns in people's movements in order to, for example, recommend events. To do so, the system relies on having mobile users who share their attendance at a large number of social events: cold-start users, who have no location history, cannot receive recommendations. We set out to address the mobile cold-start problem by answering the following research question: how can social events be recommended to a cold-start user based only on his home location? To answer this question, we carry out a study of the relationship between preferences for social events and geography, the first of its kind in a large metropolitan area. We sample location estimations of one million mobile phone users in Greater Boston, combine the sample with social events in the same area, and infer the social events attended by 2,519 residents. Upon this data, we test a variety of algorithms for recommending social events. We find that the most effective algorithm recommends events that are popular among residents of an area. The least effective, instead, recommends events that are geographically close to the area. This last result has interesting implications for location-based services that emphasize recommending nearby events.

We propose FriendSensing, a framework that automatically suggests friends to mobile social-networking users. Using short-range technologies (e.g., Bluetooth) on her mobile phone, a social-networking user "senses" and keeps track of other phones in her proximity. FriendSensing processes proximity records using a variety of algorithms that are based on social network theories of geographical proximity and of link prediction. It then returns a personalized and automatically generated list of people the user may know. We evaluate the extent to which FriendSensing helps users find people they know against real mobility and social network data.

Using mobile devices, such as smart phones, people may create and distribute different types of digital content (e.g., photos, videos). One of the problems is that digital content, being easy to create and replicate, may likely swamp users rather than informing them. To avoid that, users may organize content producers that they know and trust in a web of trust. Users may then reason about this web of trust to form opinions about content producers with whom they have never interacted before. These opinions will then determine whether content is accepted. The process of forming opinions is called trust propagation. We design a mechanism for mobile devices that effectively propagates trust and that is lightweight and distributed (as opposed to previous work that focuses on centralized propagation). This mechanism uses a graph-based learning technique. We evaluate the effectiveness (predictive accuracy) of this mechanism against a large real-world data set. We also evaluate the computational cost of a J2ME implementation on a mobile phone.