Research Group Finder Built on Open Publication Data

The problem students have finding a supervisor

Students applying for a research master's or PhD are told to "find a supervisor whose work matches your interests", and then left alone with a search engine. Most start from university rankings, which say nothing about who is actively publishing in a narrow topic such as federated learning on edge devices or low-resource Urdu speech recognition. They email famous names who stopped taking students years ago, miss the assistant professor who published four relevant papers last year, and spend weeks on a list that could be built in an afternoon with the right data.

The pain is sharper for students outside the big academic networks, including many in Pakistan, who have no senior lab member to ask. Scholarship applications such as DAAD, Chevening or Fulbright often expect a named potential supervisor or a clear research fit, so the quality of this list matters.

What already exists and the gap

• Google Scholar has author profiles and keyword search, but no way to ask "who leads active groups in this topic, by country" and no export.
• Semantic Scholar offers paper search, author pages and a public API with citation data, but ranking people by topic is left to you.
• CSRankings ranks computer science institutions by faculty papers at selective conferences using DBLP data. It is useful, but it covers CS only, works at institution level and counts a fixed conference list.
• FindAPhD and similar portals list advertised positions, which is a fraction of the supervisors who would take a strong student.

The gap is a topic-first tool, for any field, that ranks researchers by recent relevant output, explains every ranking, and hands the student a reading list rather than a mailing list. OpenAlex, a free and open index of works, authors and institutions, makes it possible without scraping.

What you build

1. Data layer: query OpenAlex for works matching a topic over the last few years, then collect authors, their institutions and co-authors. Cache every response locally so reruns cost nothing.
2. Ranking: start simple and explainable. Count relevant papers per author in the last three to five years, weight recent papers higher, and reward last-author or corresponding-author positions, which often signal a group leader. Make weights adjustable in a config file.
3. Search interface: type a topic, filter by country and years, see a ranked list with a one-line reason per person ("6 relevant papers since 2022, last author on 4"), and open a profile with recent papers and co-authors.
4. Shortlist export: a CSV with names, institutions and paper links, so the student reads before writing to anyone.

The technically interesting part

Two problems deserve your supervisor's attention. The first is author disambiguation. OpenAlex assigns author IDs, but merges and splits still happen, especially for common names. Measure how often your top results contain a merged person, and try a simple fix: flag authors whose institutions jump between continents in the same year.

The second is relevance, not popularity. Keyword search misses papers that use different words for the same idea, and raw citation counts push famous senior researchers to the top. Compare keyword matching with OpenAlex topic tags, and optionally with embedding similarity on abstracts, and report which gives better top-20 lists. This is a small information retrieval study inside a practical tool.

How to evaluate it

• Recall against a reference: for 5 topics, build a list of 20 active researchers by hand from recent papers at the main venues. Report how many your tool puts in its top 20, before and after each ranking change.
• Error analysis: list every failure you find (merged names, moved researchers, missed synonyms) with a count.
• User test: give 5 students applying abroad a topic each and time how long they take to build a shortlist of 10 with and without the tool; ask which names they would actually write to.
• Cost and speed: API calls per search with a cold cache and a warm cache.

Suggested plan

1. Weeks 1 to 2: explore the OpenAlex API, design the schema, build the cached client.
2. Weeks 3 to 5: ingestion for a topic, first ranking, command-line output.
3. Weeks 6 to 8: web interface, profiles, filters and export.
4. Weeks 9 to 10: reference lists and evaluation, ranking improvements.
5. Weeks 11 to 12: user test, error analysis, write-up.

Path to a product

The first users are your classmates applying for research degrees this year. Beyond them, university career offices and education consultants who help students apply abroad do this search by hand for every client, and a per-seat licence or a white-label version for a consultancy is a realistic model. Keep the core open source so students trust it, and offer a hosted version with saved searches and team shortlists. Scaling means caching popular topics and respecting OpenAlex's usage terms.

Scope warning

Do not scrape email addresses and do not build a "contact all" feature; the tool helps people read, not spam. Do not build a general academic search engine or a citation manager. Say plainly in the interface that publication counts are a rough signal and that a student must read someone's recent work before writing to them.

Learn what you need on Sefism

• Data analytics roadmap : SQL, pandas and the habits of checking data before trusting a ranking.
• Backend cheat sheet : API design and caching for the FastAPI layer.
• PostgreSQL Documentation : indexes and full-text search for the local store.
• Google Scholar : the tool your users use today, and the baseline to beat.
• Master's abroad or work experience first : the decision your users are in the middle of.

Get help with this idea

This project is strongest when the ranking rules are simple, written down and tested against a real reference list. Tauseef can help you narrow the scope, choose a stack you can finish, and design an evaluation your examiners will respect. If you are new here, create a free account first.

• Discuss this idea with a mentor
• Send a query to Tauseef
• Book a 1:1 session