Enhancing Retrieval-Augmented Generation with Document Link Structure for Multi-Hop Web Question Answering
Bibliographic record
Abstract
The Web is a vast and intricate network of interconnected information sources. While traditional Web search engines have long enabled users to search through billions of linked pages using keyword matching, the emergence of Large Language Models (LLMs) marks a fundamental shift in how we search for and process online information. Organizations are increasingly adopting LLM-based solutions, particularly retrieval-augmented generation (RAG), to automate information extraction from Web documents, going beyond simple keyword matching to understand semantic relationships between content. However, while RAG excels at reasoning within a single context window, it faces significant challenges in multi-hop inference, particularly in Web search scenarios where information must be gathered and processed from different sources. Graph-augmented RAGs address these limitations by leveraging graph-structured data to capture relationships between documents and enable more context-aware responses. This work investigates whether LLMs, enhanced with Web document-level graph-augmented RAG, can better automate the process of navigating and synthesizing information from complex Web documents. Our approach exploits both the textual content of Web documents and their hyperlink structure to create a comprehensive framework for Web-based information navigation and synthesis. We evaluate our approach on benchmark a multi-hop question-answering datasets, and demonstrate measurable improvements in the retrieval performance of information synthesis when compared to baseline approaches. The code and prompt templates are available at https://github.com/mehrshaad/Multi-hop-WebRAG.
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 | 0.001 | 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 | 0.001 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| 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".