Discovery & Augmentation

Helping People Discover What They'd Miss

How do we help people find and create what they'd miss? Find ideas, connections, and collaborators they wouldn't encounter on their own?

This has been my longest-running research thread, and it's taken many shapes from building better topic hierarchies so marketers could see content gaps they were missing, to helping scientists recombine concepts across disciplines, to surfacing analogies that retrieval systems overlook.

The through-line is the same: human attention has natural biases and boundaries. I build systems that push past them. What started as understanding user needs has evolved into designing tools for human-augmented discovery.

Research Papers

Scideator: Human-LLM Scientific Idea Generation Grounded in Research-Paper Facet Recombination.

Marissa Radensky*, Simra Shahid*, Raymond Fok, Pao Siangliulue, Tom Hope^, Daniel S. Weld^

(*=Equal Contribution, ^=Equal Advisors)

In AI & Scientific Discovery Workshop, NAACL 2025.
[Paper]

Contextual Alchemy: A Framework for Enhanced Readability through Cross-Domain Entity Alignment

Simra Shahid, Surgan Jandial, Nikitha Srikanth, Balaji Krishnamurthy

In Machine Learning for Creativity and Design Workshop in Neural Information Processing Systems (NeurIPS), 2023.
[Paper] [Webpage]

HyHTM: Hyperbolic Geometry based Hierarchical Topic Models

Simra Shahid*, Tanay Anand*, Nikitha Srikanth*, Sumit Bhatia, Balaji Krishnamurthy, Nikaash Puri

(equal contribution)

In Findings of the Association for Computational Linguistics: ACL 2023
[Paper] [Webpage]

Graph-Guided Unsupervised Knowledge Identification for Dialogue Agents

Shrinivas Khiste, Tathagata Raha, Milan Aggarwal, Sumit Bhatia, Simra Shahid

In The 3rd Workshop on Document-Grounded Dialog and Conversational Question Answering (Doc2Dial), ACL 2023
[Paper]

Interpretable and Hierarchical Topic Model using Hyperbolic Geometry

Simra Shahid, Tanay Anand, Nikaash Puri, Balaji Krishnamurthy, Sumit Bhatia

In Advances in Neural Information Processing Systems (NeurIPS), 2021.
[Oral Presentation]

Identification of Emergency Blood Donation Request on Twitter

Puneet Mathur, Meghna Ayyar, Sahil Chopra, Simra Shahid, Laiba Mehnaz, Rajiv Ratn Shah

In Proceedings of the 2018 EMNLP Workshop SMM4H: The 3rd Social Media Mining for Health Applications Workshop & Shared Task, Brussels, Belgium. October, 2018.
[Paper]

Identifying Adverse Drug Reactions and Personal Health Experience Mentions from Twitter

Debanjan Mahata, Sarthak Anand, Haimin Zhang, Simra Shahid, Laiba Mehnaz, Yaman Kumar, Rajiv Ratn Shah

In Proceedings of the Fourth Social Media Mining for Health Applications Workshop & Shared Task in Association for Computational Linguistics (ACL), Florence, Italy. August 2019.
[Paper]

Identifying Offensive Posts and Targeted Offense from Twitter

Debanjan Mahata, Haimin Zhang, Karan Aggarwal, Yaman Kumar, Rajiv Ratn Shah, Simra Shahid, Laiba Mehnaz, Sarthak Anand

In Proceedings of the 13th International Workshop on Semantic Evaluation in Association for Computational Linguistics (ACL), Minneapolis, Minnesota, USA. June, 2019.
[Paper]

Innovative Proposals

Contextual Alchemy

Reading, Reimagined

This proposal proposes to keep articles evergreen such that they resonate and remain relevant to readers through personalized examples. Using LLMs, it changes an example according to user topic interest, geographic region, or even generation.

Outcomes:

Accepted at NeurIPS Workshop, 2023 [Poster] [Paper] [Website]
Presented at Adobe GenAI Internal Hackathon [Video Link]
US Patent filed.

Large Langauge Model Fact-Checking Information Extraction

Content Tensor

What to write about?

Problem: Content marketers often rely on manual research to understand which content strategies work, leading to repetitive approaches like "how-to articles" that fail to engage customers effectively.
Solution: To address this, I proposed the Content Tensor, an AI-powered tool that analyzes existing content and visualizes it by strategy, topic, and engagement. The tool is designed to help markters quickly identify gaps in content (blog ideas) by showing successful or under-utilized strategies across topics. I led this initiative from identifying the market opportunity, prototyping ML-backend and XD demos, and finally presenting it to Senior VPs at Adobe.

Outcomes:

One of Seven projects to be selected from 117 submissions for Semi-Finals of Adobe Sneaks 2022 [Video Link]

Adobe Sneaks is a highly-selective projects shown in Adobe Summit as next-generation customer experiences in Adobe.

Few-Shot Classification

WorkInFlow

More Creating... Less Catching Up...

This proposal enables creatives to assign ad-hoc tasks to others within their network, based on profiles, availability, task preferences, and past interactions. It streamlines the tracking and communication process, alleviating the burden from creatives.

Outcomes:

Presented in Adobe Hackathon [Video Link]

Topic Models Profile Matching Load-Balancing

Projects

Content Gap with Session History

Session Transformer

In the rapidly evolving landscape of Generative AI, identifying which content to generate remains a challenge for marketers. Traditional analytics and SEO methods fall short in capturing the complete user journey and context. We introduce a novel framework that leverages users’ session his- tory to pinpoint gaps in existing content. We train a transformer-based model on data about what users have read and how long they spent reading it. We utilise this model to identify cluster of content attributes that users find relevant but are missing in the current repository, signifying a con- tent gap. This gap can be addressed by generating more such content, enhancing user engagement and retention.

Link to Proposal: [LINK]
Work done in collaboration with Shripad Deshmukh and Tanay Anand.

Routing user queries to experts.

Topic Models Patent Adobe Product

Adobe Support Community is a trusted online platform for Adobe customers to discuss queries about products, similar to Stack Overflow. These forums have millions of queries and it becomes difficult for the experts to find posts that they can answer. This calls for efficient and quick routing of user queries to interested experts. We proposed a routing framework, where we develop expert profiles using topic models and match it with the post topics.

Document Recommendation

Language Model Patent Adobe Product

Product 1: Identify similar user queries on the Adobe Support Community platform (similar to Stack Exchange), with the goal of reducing the number of new queries posted for an existing one. We observed a 22% increase in user’s engagement and a 20% decrease in new posts.

Product 2: We designed a workflow to recommend blogs using search signals and different types of user interactions with posts on the Adobe Marketo Nation Community Platform. In the first 8 weeks of launch, we observed a 40% increase in engagement.

Product 3: With this new cutting-edge approach to surfacing content recommendations on Marketing Nation Community, users receive recommended content based on their interactions with Marketo's email programs with the ultimate goal of increasing adoption and Community engagement. We observed a 30% increase in new users’ engagement with the Marketo offerings.

Page updated

Google Sites

Report abuse