EduPLEx_API
InfoPrototypeAll docs
Recommendation, reporting & analytics
Recommendation, reporting & analytics
  • Experiments report
    • Key concepts
    • Data sources
    • First demonstrator: ESCO ontologies and semantic matching
    • Software design
      • Endpoints Sbert_eduplex
      • Setup Sbert_eduplex
    • AI Applications
    • Conclusions
    • Recommendation
    • Bibliography
  • Recommendation Engine
  • Reporting and predictive analytics
  • LRS User Journey Visualizer
  • AI Tutor - RAG system
    • LLM-augmented Retrieval and Ranking for Course Recommendations
    • Retrieval of course candidates when searching via title.
    • Answer Generation Evaluation
    • Chunk Size and Retrieval Evaluation
    • Chunking Techniques – Splitters
    • Golden Case CLAPNQ
    • Comparative Retrieval Performance: Modules vs Golden Case
    • LLM-based Evaluator for Context Relevance
    • Retrieval Performance Indexing pdf vs xapi, and Keywords vs Questions
Powered by GitBook
On this page
Edit on GitLab
  1. AI Tutor - RAG system

Retrieval of course candidates when searching via title.

Goal To optimize the retrieval of relevant course candidates based on similar course titles. Data WBS data for 2136 modules, including title, product category, education goal, and learning goal. Method/Approach A random sample of 100 titles from the dataset was used as queries. Vector search was done on two different fields: (1) Title and (2) Combined title and learning goal. Three embedding models were tested: SBERT, OpenAI text-embedding-3-small, OpenAI text-embedding-3-large. Vector search replicated locally with FAISS using the same OpenSearch parameters (M=24, ef_search=100, ef_construction=128). Results The OpenAI 3-large model had the best retrieval scores for both the title-only search (0.92 at k=6) and when combining the title and learning goal (0.85 at k=6). The OpenAI 3-small model wasn’t far behind. SBERT struggled in both cases, showing the lowest scores (0.57 and 0.48 at k=6). Evaluation Metrics Retrieval quality: Cumulative Gain (CG at k=1 to k=6) Conclusions Results highlight the better retrieval effectiveness of OpenAI text embedding model 3-large, particularly at higher ranks, when searching on title and learning goal combined. In contrast, including the learning goal slightly decreased SBERT’s performance, possibly due to noise or irrelevant information that the embeddings struggled to handle effectively.

PreviousLLM-augmented Retrieval and Ranking for Course RecommendationsNextAnswer Generation Evaluation

Last updated 4 months ago