Export
Research Overview
Surendra Sedhai focuses on social media analytics, information retrieval, and spam detection within microblogging platforms. His research specifically examines how hashtags function in hyperlinked tweets and how malicious actors exploit these systems through hashtag hijacking. The most recent work in his trajectory addresses the impact of spam on recommendation algorithms, highlighting the necessity of robust dataset evaluation.
Research Trajectory
Sedhai’s research trajectory demonstrates a clear progression from developing information retrieval systems for social media to addressing the adversarial behaviors that compromise those systems. His early work focused on enriching the context of microblog posts. As the first author of Hashtag recommendation for hyperlinked tweets (2014), Sedhai investigated methods for suggesting relevant hashtags for tweets containing URLs. This research argued that hashtags do more than categorize short text; they provide critical context that can be extended to the linked web documents themselves. The proposed system utilized candidate selection schemes based on similar tweets, document content, and named entities to improve recommendation accuracy.
Recognizing that the utility of hashtags makes them a prime target for manipulation, Sedhai’s focus subsequently shifted toward data integrity and spam detection. In 2015, he led the development of HSpam14: A Collection of 14 Million Tweets for Hashtag-Oriented Spam Research. This work addressed a significant gap in the field: while email and web spam were well-studied, hashtag-oriented spam—such as hashtag hijacking—lacked large-scale, systematically annotated datasets. By collecting and annotating 14 million tweets matching trending hashtags, Sedhai provided a foundational resource for understanding how spam diffuses through dynamic Twitter communities. This dataset creation aligns with the broader area of Adversarial Linguistic Evasion in Content Moderation, providing the raw material necessary to train systems to recognize malicious content disguised within popular trends.
In 2016, Sedhai synthesized these two research streams—recommendation systems and spam detection—by examining how the latter affects the former. In Effect of Spam on Hashtag Recommendation for Tweets, he utilized the HSpam14 dataset to demonstrate that the presence of spam fundamentally alters algorithm evaluation. The study revealed a critical methodological pitfall: because spam tweets often contain near-duplicate content, recommendation algorithms can easily suggest the "correct" hashtags for them. Consequently, failing to filter out spam can artificially inflate the perceived performance of a recommendation system, masking its actual effectiveness on legitimate, non-spam data.
Current Directions
Based on the most recent papers in the provided evidence, dating to 2016, Sedhai’s work investigates the intersection of recommendation algorithms and data quality. The core problem addressed in this recent research is the vulnerability of system evaluation metrics to noisy or adversarial data. By demonstrating that spam can artificially boost recommendation accuracy due to near-duplicate text, this work emphasizes the need for rigorous, spam-aware evaluation frameworks in information retrieval. The overarching theme is the development of social media analytics that are not only algorithmically sound but also resilient to the realities of platform manipulation and hashtag hijacking.
Profile updated
Loading new papers…
Loading research neighbors…
See something wrong? Report a mistake