Manasi Patwardhan

Research profile

Research summary

Manasi Patwardhan is a researcher at the Indian Institute of Technology Bombay whose work spans two equally weighted directions — AI evaluation and scientometrics with NLP on one side, semantic web and data integration through knowledge representation with NLP on the other — each accounting for half of the field distribution in the underlying OpenAlex record. The topic profile is led by topic modeling (35 items), followed by multimodal machine learning applications (15), natural language processing techniques (12), software engineering research (10), and information retrieval and search behavior (9). Bibliometric aggregates include an h-index of 11 and an i10-index of 13, though impact scoring for this profile has not been finalized. Activity is the standout signal: 43 works fall in the most recent three-year window, with annual counts rising from 1 (2010) and 5 (2016) to 18 (2025) and 17 (2026). Two 2026 arXiv papers are cataloged and reviewed by Evolve: BackTrend, a validated retrospective benchmark for scientific weak-signal prediction whose low system scores (F1 ≤ 10.1%) may partly reflect strict matching and construction bias, and a study on self-demonstrations for LLM-based schema-ontology mapping reporting 0.56 average F1 on RODI, with strong gains but only 3 scenarios and no significance testing. Both papers carry medium reproducibility readiness, and the reproducibility dimension averages 55 across them.

Research keywords

Highlights

• BackTrend: a retrospective benchmark for scientific weak-signal prediction

  Evolve judges this 2026 arXiv paper a validated retrospective benchmark for evaluating scientific weak-signal prediction with LLM-as-judge, deep research agents, and RAG. Low system scores (F1 ≤ 10.1%) may partly reflect strict matching and benchmark-construction bias; reproducibility readiness is rated medium.

• Self-demonstrations lift LLM schema–ontology mapping

  Evolve's verdict: neuro-symbolic decomposition plus pattern-guided self-demonstrations lift RODI mapping to 0.56 average F1 with LLMs. Gains are strong but rest on only 3 scenarios with no significance testing; reproducibility readiness is medium.

• A 2026 evaluation-methodology thread across both strands

  Both cataloged papers, released as 2026 arXiv preprints, share an LLM-capability evaluation core: one benchmarks deep research agents and RAG systems on scientific weak signals, the other probes in-context self-demonstrations for mapping tasks. This mirrors the 50/50 field split and the 17 works recorded for 2026.

Timeline

1. 2010First year with a recorded publication in the OpenAlex-sourced counts (1 work).
2. 2016Output reaches 5 works, the highest annual total of the mid-2010s.
3. 20206 works published, matching the 2018 level.
4. 20237 works — output edges past the 2018/2020 plateau.
5. 2025Annual output jumps to 18 works, the highest year on record.
6. 202617 works so far, including the two arXiv papers reviewed in the Evolve catalog (BackTrend and the self-demonstrations study).

Papers in the Evolve catalog

• 2026-09-21

  BackTrend: Evaluating Scientific Weak-Signal Prediction via Backward Reconstruction

• 2026-09-12

  Surprising Effectiveness of Self-Demonstrations in Enhancing Schema-Ontology Mapping with LLMs

Frequent co-authors