arXiv subjects

Ankit Peshin

Publications and source records attributed to Ankit Peshin.

5 recordsLinked to original sources

arXiv2026

The Case Against Generation for Retrieval: Discriminative Language Models as Effective Retrievers

Large Language Models (LLMs) have emerged as powerful assets for recommender systems. However, deploying them as generative recommenders or zero-shot rankers at web-scale remains bottlenecked by prohibitive computational overhead and grounding challenges. In this paper, we revitalize the classic, highly efficient two-tower retrieval architecture by adapting LLMs as semantic representation backbones rather than generative engines. We introduce an LLM-native two-tower framework engineered for high-throughput, large-scale retrieval. Our architecture introduces several key innovations: a shared LLM encoder for joint user-item modeling, End-Of-Sentence (EOS) token pooling for compact sequence embedding, cross-dataset transfer learning, knowledge distillation from powerful cross-encoder teachers, and latent reasoning within the user tower. Extensive evaluation across three public benchmarks demonstrates that cross-encoder architecture outperforms current state-of-the-art (SoTA) models, while the efficient two-tower student achieves SoTA-comparable retrieval performance. Furthermore, experiments on internal large-scale production systems yield substantial topline retrieval improvements along with high resilience to model staleness and superior data scaling. Our findings demonstrate that when augmented with modern representation learning, the traditional two-tower paradigm remains an exceptionally competitive and practical solution for industrial retrieval systems.

cs.IR↗

arXiv2026

Tokenizing Numerical and Embedding Features for LLM RecSys

Large language models (LLMs) are increasingly used as backbone architectures for recommender systems because of their strong sequence modeling and representation learning capabilities. However, most LLM-based recommenders operate primarily on discrete textual tokens, whereas practical recommendation pipelines also rely on continuous numerical features and dense embedding features produced by upstream feature engineering or pretrained encoders. This mismatch limits the ability of LLM-based models to exploit fine-grained non-textual signals. We propose a soft-token fusion framework that maps numerical and embedding features into the LLM embedding space, allowing heterogeneous recommendation signals to be consumed through the standard token interface. We instantiate the framework in a shared-parameter LLM-based two-tower retrieval model and introduce an interaction-based fusion module that refines embedding and numerical soft tokens before they are inserted into the final LLM input. Experiments on three Amazon recommendation benchmarks show that soft-token fusion improves retrieval performance over LLM-based baselines, and that interaction-based fusion is more effective than direct concatenation of heterogeneous soft tokens.

cs.IR↗

arXiv2022

Explaining Website Reliability by Visualizing Hyperlink Connectivity

As the information on the Internet continues growing exponentially, understanding and assessing the reliability of a website is becoming increasingly important. Misinformation has far-ranging repercussions, from sowing mistrust in media to undermining democratic elections. While some research investigates how to alert people to misinformation on the web, much less research has been conducted on explaining how websites engage in spreading false information. To fill the research gap, we present MisVis, a web-based interactive visualization tool that helps users assess a website's reliability by understanding how it engages in spreading false information on the World Wide Web. MisVis visualizes the hyperlink connectivity of the website and summarizes key characteristics of the Twitter accounts that mention the site. A large-scale user study with 139 participants demonstrates that MisVis facilitates users to assess and understand false information on the web and node-link diagrams can be used to communicate with non-experts. MisVis is available at the public demo link: https://poloclub.github.io/MisVis.

cs.SI↗

arXiv2022

Domain-Level Detection and Disruption of Disinformation

How, in 20 short years, did we go from the promise of the internet to democratize access to knowledge and make the world more understanding and enlightened, to the litany of daily horrors that is today's internet? We are awash in disinformation consisting of lies, conspiracies, and general nonsense, all with real-world implications ranging from horrific humans rights violations to threats to our democracy and global public health. Although the internet is vast, the peddlers of disinformation appear to be more localized. To this end, we describe a domain-level analysis for predicting if a domain is complicit in distributing or amplifying disinformation. This process analyzes the underlying domain content and the hyperlinking connectivity between domains to predict if a domain is peddling in disinformation. These basic insights extend to an analysis of disinformation on Telegram and Twitter. From these insights, we propose that search engines and social-media recommendation algorithms can systematically discover and demote the worst disinformation offenders, returning some trust and sanity to our online communities.

cs.CY↗

arXiv2021

Mutual Hyperlinking Among Misinformation Peddlers

The internet promised to democratize access to knowledge and make the world more open and understanding. The reality of today's internet, however, is far from this ideal. Misinformation, lies, and conspiracies dominate many social media platforms. This toxic online world has had real-world implications ranging from genocide to, election interference, and threats to global public health. A frustrated public and impatient government regulators are calling for a more vigorous response to mis- and disinformation campaigns designed to sow civil unrest and inspire violence against individuals, societies, and democracies. We describe a large-scale, domain-level analysis that reveals seemingly coordinated efforts between multiple domains to spread and amplify misinformation. We also describe how the hyperlinks shared by certain Twitter users can be used to surface problematic domains. These analyses can be used by search engines and social media recommendation algorithms to systematically discover and demote misinformation peddlers.

cs.SI↗