SVOP: Improving document retrieval accuracy through semantic search
Project name: SVOP semantic search.
- Project’s period: approx 10/2024 until 12/2024, 3 months
- Partner: SVOP
The goal of the project was to improve full-text search to return more accurate results. Traditional word-based search approaches were to be replaced by semantic search, a method that compares text based on meaning rather than exact word matching, enabling more relevant results to be returned in response to user queries.
How we approached it
We implemented a semantic search system built around the Milvus vector database. The solution operated as:
- Document embedding: An embedding model from OpenAI was used to transform entire documents into vectors, which were then inserted into the Milvus database. This encoding captures the meaning of each document rather than its literal word content.
- Query processing: User input was transformed into vectors using the same embedding model, then compared against the stored document vectors. Documents with vectors most similar to the user input were returned as results.
- Testing: The solution was tested using multiple approaches to verify the quality and accuracy of the returned results.
“One of the most rewarding aspects of our work as a research institute is seeing research move beyond the research “lab” and into practice. This project is a great example of how expertise in AI models and semantic technologies can be transformed into a solution that measurably improves the performance of an operational system.”
Martin Tamajka
Technology lead, KInIT
“The project was interesting because it was a prototype of semantic search, which is an approach based on the meaning of the text. The technology used, namely semantic search using vector databases, is a very powerful tool for finding duplicates or significantly similar texts, and the quality of this technology will only grow with the improvement of embedding models and vector databases. Since the client was technically savvy and communicative, the project developed smoothly and helped the client improve their product.”
Denis Mitana
LLMs Lead, NanLab
What we delivered
The partner used our solution as a reference to implement their own semantic search system, built on the ElasticSearch database.
“Our collaboration with KInIT within the Hopero project brought us many new insights and showed us new directions that we had not dared to consider before. The consultations, combined with practical demonstrations, significantly accelerated our ongoing development and enabled its expansion in multiple directions.”
Mgr. Ján Grman, PhD.
Director, Solution architect, SVOP
https://kinit.sk/svop-improving-document-retrieval-accuracy-through-semantic-search
Kempelenov inštitút inteligentných technológií
Beispiel der Dun & Bradstreet-Daten
| Firmenname | AUDI Aktiengesellschaft |
|---|---|
| D-U-N-S-Nummer | 315731679 |
| Registrierungsnummer / Beschreibung | HRB1 Register Number (DE) |
| Registrierungsnummer / Beschreibung | 85049B1 Derived Unique Register Number (DE) |
| Registrierungsnummer / Beschreibung | DE811115368 Value Added Tax Number (DE) |
| Telefon | +49 841890 |
| Subjekttyp | Corporation |
| Rechtsform | Deemed Public Limited Company |
| Jahresumsatz | 64 500 000 000 EUR / 71 157 238 500 USD - 01.01.2024 |
| Mitarbeiterzahl | 86 900 - 01.01.2024 |
| Status | Active |
| Adresse | Auto-Union-Str. 1, Ingolstadt, 85057, Germany |
| Webseite | www.audi.com |
| [email protected] | |
| Geschäftsgegenstand | Mfg motor vehicle/car bodies |
| Eigentumstyp | Privately owned |
| Gründungsdatum | 16.04.1884 |
| Gesetzliche Vertreter | Gernot Döllner |
| Eigentümer | Bundesland Niedersachsen, Streubesitz, GOVERNMENT OF QATAR, Ferdinand Porsche Familien-Privatstiftung, Kai Alexander Porsche, Peter Daniell Porsche, Ferdinand Oliver Porsche, Geraldine Porsche, Mark Philipp Porsche, Hubertus Josef Kiesling, Diana Porsche, Andreas Kiesling, Peter Daniell Porsche, Kai Alexander Porsche, Ferdinand Oliver Porsche, Mark Philipp Porsche, Geraldine Porsche, Diana Porsche, Hubertus Josef Kiesling, Andreas Kiesling, Wolfgang Porsche, Stephanie Porsche- Schröder, Felix Alexander Porsche, Christian Porsche, Ferdinand Rudolf Wolfgang Porsche |
