SIOP 2026 Recommender

SIOP 2026 Conference Recommender

Planning a SIOP schedule can be overwhelming: many promising sessions, limited time, and plenty of overlapping options. This app helps narrow the program into an inspiration-first shortlist, a suggested non-overlapping agenda, and a network map of similar sessions.

What It Does

• Ranks sessions by how well they match your conference goals.
• Lets you steer the search with intent, keywords, avoid terms, and optional CV text.
• Marks schedule conflicts in the recommendation table.
• Builds a suggested personal agenda that avoids overlapping sessions.
• Shows an ONA-style similarity map with your current top recommendations highlighted.

How To Use It

1. Start in Profile. Write what you want from the conference: topics you want to explore, skills you want to develop, questions you want answered, or problems you want to bring back to your work.
2. Treat the intent field as the strongest signal. A focused three-to-five sentence statement is usually more useful than a long, vague description.
3. Add keywords for specific topics or methods, and add avoid terms for themes you do not want to prioritize.
4. Optionally upload or paste CV text. CV text is a supporting signal, not the main driver.
5. Open Recommendations, adjust the weights if needed, and scan the top sessions. In most cases, keep intent as the strongest weight.
6. Open My Agenda to see a suggested non-overlapping plan, then use Explore Map to inspect related sessions around promising choices.

How The Agenda Is Built

The agenda is generated automatically from the current recommendations. It is not simply the first Top N rows. The app favors higher-scoring sessions, removes sessions that overlap in time, and applies the selected travel buffer. A lower-ranked session may appear in the agenda when a higher-ranked session conflicts with another stronger choice.

How The Map Works

The Explore Map shows sessions as a network. Each node is a session. A link means two sessions are semantically similar enough to pass the selected similarity threshold. Raising the threshold creates a stricter, sparser map; lowering it creates a denser map. Click a node or search for a title to see nearby sessions connected by the current threshold.

Limitations

• The app is not an official SIOP source and should not be treated as one.
• Program details can change. Times, rooms, session titles, and availability should be verified.
• The app uses the session information that could be extracted from public program data. Some sessions may have limited or missing descriptive text.
• Information extraction can introduce errors, duplicates, missing fields, or imperfect session IDs.
• Embeddings capture broad semantic similarity, but they do not understand every nuance of your goals, the field, or the conference context.
• Poster sessions and very terse descriptions may be scored less reliably.

Light Technical Notes

• Session text is represented with precomputed local embeddings from Alibaba-NLP/gte-multilingual-base.
• Matching uses cosine similarity between session embeddings and separate profile signals: intent, keywords, CV text, and avoid terms.
• The agenda optimizer uses weighted interval scheduling with a travel buffer to avoid time conflicts.
• The Explore Map is an ONA-style network using browser-side vis-network physics.
• The public app uses a static frontend and a FastAPI backend on Cloud Run.

Contact

Mail: ludek.stehlik@gmail.com LinkedIn: ludekstehlik