Bibliographic record
Abstract
Digital content has grown exponentially, requiring efficient and context-aware information retrieval systems. Search methods which utilize key word based functions, such as BM25, rely on lexical matching but often fail to capture semantic relationships. Using dense vector embeddings models, Small Language Models (SLMs) can improve retrieval accuracy, however Retrieval-Augmented Generation (RAG) incorporates the concept of contextual ranking through generative artificial intelligence to further enhance search relevance. An experiment on a single document corpus and a cryptography-related query evaluates BM25, SLM embeddings, and SLM + RAG for document retrieval. Based on experimental results, BM25 achieves a moderate relevance score of 0.500, retrieving documents based on exact matches but lacking contextual understanding. As SLM embeddings identify semantically similar concepts, they increase recall for queries with conceptual variations and a relevance score 0.7668. SLM + RAG outperforms both approaches' relevance score 0.9202, retrieving the most relevant document with accuracy and contextually enriched responses. A hybrid retrieval model that combines dense embeddings and generative ranking improves search quality substantially. This exploratory study has multiple implications for information retrieval and including the need for scalable multi-document retrieval, FAISS/DPR integration for efficient vector searches, and domain-specific fine-tuning of SLMs. Using this approach, enterprises can achieve faster, more accurate, and contextually aware document retrieval by combining SLMs and RAG methods. Hybrid approaches could be explored in large-scale retrieval settings, with stable workflow integration processes across diverse industries like legal research, healthcare, government, and finance.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.000 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.000 | 0.001 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".