Krishna Yeswanth Kamath

Spatio-temporal dynamics of online memes: a study of geo-tagged tweets

We conduct a study of the spatio-temporal dynamics of Twitter hashtags through a sample of 2 billion geo-tagged tweets. In our analysis, we (i) examine the impact of location, time, and distance on the adoption of hashtags, which is important for understanding meme diffusion and information propagation; (ii) examine the spatial propagation of hashtags through their focus, entropy, and spread; and (iii) present two methods that leverage the spatio-temporal propagation of hashtags to characterize locations. Based on this study, we find that although hashtags are a global phenomenon, the physical distance between locations is a strong constraint on the adoption of hashtags, both in terms of the hashtags shared between locations and in the timing of when these hashtags are adopted. We find both spatial and temporal locality as most hashtags spread over small geographical areas but at high speeds. We also find that hashtags are mostly a local phenomenon with long-tailed life spans. These (and other) findings have important implications for a variety of systems and applications, including targeted advertising, location-based services, social media search, and content delivery networks.

Detecting collective attention spam

We examine the problem of collective attention spam, in which spammers target social media where user attention quickly coalesces and then collectively focuses around a phenomenon. Compared to many existing spam types, collective attention spam relies on the users themselves to seek out the content -- like breaking news, viral videos, and popular memes -- where the spam will be encountered, potentially increasing its effectiveness and reach. We study the presence of collective attention spam in one popular service, Twitter, and we develop spam classifiers to detect spam messages generated by collective attention spammers. Since many instances of collective attention are bursty and unexpected, it is difficult to build spam detectors to pre-screen them before they arise; hence, we examine the effectiveness of quickly learning a classifier based on the first moments of a bursting phenomenon. Through initial experiments over a small set of trending topics on Twitter, we find encouraging results, suggesting that collective attention spam may be identified early in its life cycle and shielded from the view of unsuspecting social media users.

Spatial influence vs. community influence: modeling the global spread of social media

In this paper we seek to understand and model the global spread of social media. How does social media spread from location to location across the globe? Can we model this spread and predict where social media will be popular in the future? Toward answering these questions, we develop a probabilistic model that synthesizes two conflicting hypotheses about the nature of online information spread: (i) the spatial influence model, which asserts that social media spreads to locations that are close by; and (ii) the community affinity influence model, which asserts that social media spreads between locations that are culturally connected, even if they are distant. Based on the geospatial footprint of 755 million geo-tagged hashtags spread through Twitter, we evaluate these models at predicting locations that will adopt hashtags in the future. We find that distance is the single most important explanation of future hashtag adoption since hashtags are fundamentally local. We also find that community affinities (like culture, language, and common interests) enhance the quality of purely spatial models, indicating the necessity of incorporating non-spatial features into models of global social media spread.

Toward traffic-driven location-based web search

The emergence of location sharing services is rapidly accelerating the convergence of our online and offline activities. In one direction, Foursquare, Google Latitude, Facebook Places, and related services are enriching real-world venues with the social and semantic connections among online users. In analogy to how clickstreams have been successfully incorporated into traditional web ranking based on content and link analysis, we propose to mine traffic patterns revealed through location sharing services to augment traditional location-based search. Concretely, we study location-based traffic patterns revealed through location sharing services and find that these traffic patterns can identify semantically related locations. Based on this observation, we propose and evaluate a traffic-driven location clustering algorithm that can group semantically related locations with high confidence. Through experimental study of 12 million locations from Foursquare, we extend this result through supervised location categorization, wherein traffic patterns can be used to accurately predict the semantic category of uncategorized locations. Based on these results, we show how traffic-driven semantic organization of locations may be naturally incorporated into location-based web search.

How big is the crowd?: event and location based population modeling in social media

In this paper, we address the challenge of modeling the size, duration, and temporal dynamics of short-lived crowds that manifest in social media. Successful population modeling for crowds is critical for many services including location recommendation, traffic prediction, and advertising. However, crowd modeling is challenging since 1) user-contributed data in social media is noisy and oftentimes incomplete, in the sense that users only reveal when they join a crowd through posts but not when they depart; and 2) the size of short-lived crowds typically changes rapidly, growing and shrinking in sharp bursts. Toward robust population modeling, we first propose a duration model to predict the time users spend in a particular crowd. We propose a time-evolving population model for estimating the number of people departing a crowd, which enables the prediction of the total population remaining in a crowd. Based on these population models, we further describe an approach that allows us to predict the number of posts generated from a crowd. We validate the crowd models through extensive experiments over 22 million geo-location based check-ins and 120,000 event-related tweets.

Spatio-temporal meme prediction: learning what hashtags will be popular where

In this paper, we tackle the problem of predicting what online memes will be popular in what locations. Specifically, we develop data-driven approaches building on the global footprint of 755 million geo-tagged hashtags spread via Twitter. Our proposed methods model the geo-spatial propagation of online information spread to identify which hashtags will become popular in specific locations. Concretely, we develop a novel reinforcement learning approach that incrementally updates the best geo-spatial model. In experiments, we find that the proposed method outperforms alternative linear regression based methods.

Predicting semantic annotations on the real-time web

The explosion of the real-time web has spurred a growing need for new methods to organize, monitor, and distill relevant information from these large-scale social streams. One especially encouraging development is the self-curation of the real-time web via user-driven linking, in which users annotate their own status updates with lightweight semantic annotations -- or hashtags. Unfortunately, there is evidence that hashtag growth is not keeping pace with the growth of the overall real-time web. In a random sample of 3 million tweets, we find that only 10.2% contain at least one hashtag. Hence, in this paper we explore the possibility of predicting hashtags for un-annotated status updates. Toward this end, we propose and evaluate a graph-based prediction framework. Three of the unique features of the approach are: (i) a path aggregation technique for scoring the closeness of terms and hashtags in the graph; (ii) pivot term selection, for identifying high value terms in status updates; and (iii) a dynamic sliding window for recommending hashtags reflecting the current status of the real-time web. Experimentally we find encouraging results in comparison with Bayesian and data mining-based approaches.

Community-based ranking of the social web

The rise of social interactions on the Web requires developing new methods of information organization and discovery. To that end, we propose a generative community-based probabilistic tagging model that can automatically uncover communities of users and their associated tags. We experimentally validate the quality of the discovered communities over the social bookmarking system Delicious. In comparison to an alternative generative model (Latent Dirichlet Allocation (LDA), we find that the proposed community-based model improves the empirical likelihood of held-out test data and discovers more coherent interest-based communities. Based on the community-based probabilistic tagging model, we develop a novel community-based ranking model for effective community-based exploration of socially-tagged Web resources. We compare community-based ranking to three state-of-the-art retrieval models: (i) BM25; (ii) Cluster-based retrieval using K-means clustering; and (iii) LDA-based retrieval. We find that the proposed ranking model results in a significant improvement over these alternatives (from 7% to 22%) in the quality of retrieved pages.

Discovering trending phrases on information streams

We study the problem of efficient discovery of trending phrases from high-volume text streams -- be they sequences of Twitter messages, email messages, news articles, or other time-stamped text documents. Most existing approaches return top-k trending phrases. But, this approach neither guarantees that the top-k phrases returned are all trending, nor that all trending phrases are returned. In addition, the value of k is difficult to set and is indifferent to stream dynamics. Hence, we propose an approach that identifies all the trending phrases in a stream and is flexible to the changing stream properties.

Board coherence in Pinterest: non-visual aspects of a visual site

Pinterest is a fast-growing interest network with significant user engagement and monetization potential. This paper explores quality signals for Pinterest boards, in particular the notion of board coherence. We find that coherence can be assessed with promising results and we explore its relation to quality signals based on social interaction.