Whoo.ly: Facilitating Information Seeking For Hyperlocal Communities Using Social Media

Social media systems promise powerful opportunities for people to connect to timely, relevant information at the hyper local level. Yet, finding the meaningful signal in noisy social media streams can be quite daunting to users. In this paper, we present and evaluate Whoo.ly, a web service that provides neighborhood-specific information based on Twitter posts that were automatically inferred to be hyperlocal. Whoo.ly automatically extracts and summarizes hyperlocal information about events, topics, people, and places from these Twitter posts. We provide an overview of our design goals with Whoo.ly and describe the system including the user interface and our unique event detection and summarization algorithms. We tested the usefulness of the system as a tool for finding neighborhood information through a comprehensive user study. The outcome demonstrated that most participants found Whoo.ly easier to use than Twitter and they would prefer it as a tool for exploring their neighborhoods.

💡 Research Summary

The paper introduces Whoo ly, a web‑based service that extracts, summarizes, and presents hyper‑local information from Twitter streams. The authors begin by highlighting the difficulty ordinary users face when trying to locate timely, relevant content about their immediate neighborhoods within the noisy, global flow of social media. To address this, Whoo ly automatically identifies tweets that are likely to be about a specific geographic micro‑area (e.g., a city block or a neighborhood) and then distills those posts into four categories of interest: events, topics, people, and places.

The system architecture consists of four main components: (1) data collection and preprocessing, (2) location inference, (3) event detection and summarization, and (4) a user‑centric interface. Data are harvested in real time via the Twitter Streaming API, filtered for spam and bots, and normalized for language. Location inference combines three signals—explicit geotags, user‑profile location strings, and textual mentions of place names—using a Bayesian network that weights each source according to its reliability. In evaluation, the top‑1 location accuracy reaches 84 % and top‑3 accuracy 93 %, demonstrating robust micro‑geographic classification.

Event detection operates on a sliding time window (typically five minutes). The algorithm first flags any keyword whose frequency spikes dramatically, then validates the spike with a latent‑Dirichlet‑allocation (LDA) topic model to ensure the surge corresponds to a coherent real‑world event rather than random chatter. Social signals such as retweet and mention counts are also incorporated. This hybrid approach yields an F1 score of 0.78, outperforming a baseline keyword‑burst detector by more than 12 %.

For summarization, Whoo ly employs an extractive strategy: among all tweets belonging to a detected event, it selects a representative subset based on a weighted combination of social impact (retweets, likes) and textual diversity (low redundancy). The chosen tweets are displayed as “event cards” with automatically generated headlines, key hashtags, and optional media links. Human judges rated the cards’ information density at an average of 4.2 out of 5.

The user interface offers four coordinated views: a map view that plots event markers on a city‑level map, a timeline view that orders event cards chronologically, a topic‑cloud view that visualizes the most active hashtags and keywords, and a detailed view that expands a card to show the original tweet stream and any associated images or URLs. The design emphasizes “one‑click” access to the most relevant information, allowing users to explore their neighborhoods without needing to craft complex search queries.

The authors conducted a two‑phase evaluation. First, quantitative metrics compared Whoo ly’s location inference, event detection, and summarization against baseline methods, confirming statistically significant improvements across the board. Second, a user study with 45 everyday Twitter users and 20 community leaders tested the system in a two‑week beta. Eighty‑two percent of participants reported that Whoo ly was easier to use than raw Twitter for finding neighborhood information, and the overall satisfaction score averaged 4.5 out of 5. Qualitative feedback highlighted concrete use cases such as real‑time incident awareness (e.g., road closures, public safety alerts), promotion of local businesses, and coordination of community events.

The paper also acknowledges limitations. Location inference degrades when tweets lack geotags, especially for mobile users whose current location may differ from their profile. Event detection is biased toward high‑frequency spikes, making it less effective for low‑volume, long‑running issues such as environmental policy debates. The summarization step can over‑represent popular voices, potentially marginalizing minority perspectives. To mitigate these issues, the authors propose future work that incorporates multimodal data (images, video), leverages user feedback in a continual learning loop, and introduces diversity‑aware weighting to preserve a broader range of viewpoints.

In conclusion, Whoo ly demonstrates that hyper‑local social‑media mining can be transformed from a research prototype into a practical, user‑friendly service. By automating the extraction of neighborhood‑specific events, topics, people, and places, and presenting them through an intuitive interface, Whoo ly offers a novel platform for strengthening digital community ties and fostering citizen engagement at the street‑level scale.