- Rating
- 3(+1)
- Confidence
- 4
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 3(+1)
Read review
Paper summary
The paper introduces CiteProbe, a method for identifying source sentences that support an LLM-generated answer using attention patterns across individual attention heads. The authors first investigate which attention heads contain useful citation-related information and then train a lightweight linear probe on head-specific attention features to score and rank candidate source sentences based on their likelihood of supporting the generated answer.
Critical strengths
-
The problem addressed in this paper is relevant and important, particularly given the growing need for reliable citation attribution in LLM-generated responses.
-
The idea of using attention patterns from different layers and heads to identify supporting evidence is interesting. It also avoids the repeated inference required by several existing attribution methods.
-
CiteProbe has low additional inference cost compared with the evaluated baselines, making it a potentially practical approach for citation attribution in RAG systems.
Critical weaknesses
-
The main results focus on mAP and Top-1 accuracy, which measure how well CiteProbe ranks sources but do not show how much supporting evidence it actually identifies. Appendix L reports low recall in some cases, but this is not discussed in the main text.
-
Recall is particularly important because the paper evaluates CiteProbe on multi-hop datasets, where multiple pieces of evidence are needed to support an answer. Finding just one relevant sentence is not enough if other required evidence is missing, but the main results do not show how much of this evidence CiteProbe actually identifies.
-
Even with the current evaluation setup used in the main text, CiteProbe shows only small improvements over the Top-16-head baseline and sometimes performs worse. It is unclear how much the learned linear probe actually helps.
-
The claim that CiteProbe extends to proprietary models is overstated. The authors pass their answers through an open-source model and use its attention weights to find supporting evidence. They could do the same with ground-truth answers. This only shows that CiteProbe works on externally generated text.
Questions for authors
-
Why does the main evaluation emphasize Top-1 accuracy without reporting recall, especially for questions that require multiple supporting sentences, when mAP already measures ranking performance?
-
In the proprietary-model experiments, CiteProbe simply takes the generated answers and uses an open-source model to identify supporting evidence. The same could be done with ground-truth answers. What do these experiments demonstrate about CiteProbe's extension to proprietary models beyond its ability to process any given answer?
-
How does CiteProbe turn its source predictions into actual citations? Are citations assigned to individual claims or the whole answer? Why is this comparison with citation generation methods only discussed in Appendix K and not in main text?
7 comments
W2. Recall is particularly important because the paper evaluates CiteProbe on multi-hop datasets, where multiple pieces of evidence are needed to support an answer. Finding just one relevant sentence is not enough if other required evidence is missing, but the main results do not show how much of this evidence CiteProbe actually identifies.
Q1. Why does the main evaluation emphasize Top-1 accuracy without reporting recall, especially for questions that require multiple supporting sentences, when mAP already measures ranking performance?
Reply: We agree that recall is important for identifying multiple supporting sentences, particularly for multi-hop questions. Top-1 measures the single-citation use case, and recall is shown through R/P/F1 and mAP. Since recall can be obtained by trading off precision, we address this concern directly by using reported data (Table 23, Appendix M.2) showing CiteProbe's Recall and Precision, alongside F1 and also the added AUPRC on multi-hop HotpotQA and MuSiQue, alongside the strongest competing baseline:
| Backbone | Method | HotpotQA Recall | HotpotQA Precision | HotpotQA F1 | HotpotQA AUPRC | MuSiQue Recall | MuSiQue Precision | MuSiQue F1 | MuSiQue AUPRC |
|---|---|---|---|---|---|---|---|---|---|
| Phi-4-mini | Runner-up | 69.1% | 72.7% | 70.9% | .750$^{\dagger}$ | 53.2% | 69.3% | 60.2% | .536$^{*}$ |
| CiteProbe | 76.4% | 66.6% | 71.2% | .762 | 62.6% | 69.0% | 65.6% | .690 | |
| Ministral-3-8B | Runner-up | 66.5% | 71.2% | 68.8% | .744$^{\dagger}$ | 60.5% | 64.7% | 62.5% | .584$^{*}$ |
| CiteProbe | 70.6% | 68.4% | 69.5% | .750 | 71.2% | 66.7% | 68.9% | .724 | |
| Qwen3.5-9B | Runner-up | 69.9% | 54.8% | 61.4% | .512$^{*}$ | 83.6% | 29.6% | 43.8% | .447$^{*}$ |
| CiteProbe | 74.6% | 65.7% | 69.9% | .743 | 74.3% | 65.8% | 69.8% | .740 | |
| Gemma-4-12B | Runner-up | 60.8% | 73.3% | 66.4% | .708$^{\dagger}$ | 53.6% | 45.9% | 49.4% | .385$^{*}$ |
| CiteProbe | 74.1% | 71.9% | 73.0% | .787 | 69.1% | 68.8% | 68.9% | .732 |
$^{\dagger}$ denotes AT2, and denotes ARC-JSD.
Overall, CiteProbe demonstrates consistently strong multi-hop evidence recovery, achieving higher recall in 7/8 settings, while maintaining higher F1 and AUPRC in all 8 settings against runner-up baselines. More specifically, on Qwen3.5-9B/MuSiQue, ARC-JSD reaches 83.6% recall by trading precision down to 29.6%. The resulting F1 of 43.8% and AUPRC of .447 show the cost of that trade-off.
Recall should therefore be interpreted together with precision and AUPRC. The evidence for multi-sentence recovery comes directly from the citation-set evaluation using Recall and F1, while mAP and Top-1 provide complementary information about ranking quality. AUPRC further evaluates performance across operating thresholds and avoids relying on a single threshold choice.
Recognizing your inputs, and as discussed in W1, we added AUPRC to comprehensively address recall concerns. On this standard metric, CiteProbe managed the highest mean AUPRC on five of six backbones.
W3. Even with the current evaluation setup used in the main text, CiteProbe shows only small improvements over the Top-16-head baseline and sometimes performs worse. It is unclear how much the learned linear probe actually helps.
Reply: Thank you for this observation. We agree that the gains over Top-16 are modest in some settings. Top-16 is a strong diagnostic baseline and essentially a restricted linear probe, derived from our preliminary analysis showing that a small subset of informative heads can already support effective citation attribution. We included Top-16 as a transparent diagnostic baseline, and CiteProbe generalizes this idea by learning head-specific weights jointly in a single linear fit, rather than relying on a separate head-ranking and selection procedure.
To quantify the difference more carefully, we report 95% paired bootstrap confidence intervals (CI) for CiteProbe − Top-16. Across the 48 settings, CiteProbe achieves 18 wins, 25 ties, and 5 losses, indicating modest gains over the strong Top-16 baseline.
CiteProbe vs Top-16
| Backbone | HotpotQA | MuSiQue | TyDiQA | WiCE |
|---|---|---|---|---|
| Phi-4-mini () | ○ -.007 [-.016, .001] | ▲ .013 [.005, .021] | ○ .011 [-.001, .024] | ○ .002 [-.006, .010] |
| Phi-4-mini () | ○ .007 [-.013, .027] | ○ .013 [-.009, .035] | ○ .015 [-.005, .038] | ▲ .025 [.009, .042] |
| Ministral-3-8B () | ○ -.003 [-.014, .007] | ▲ .030 [.021, .039] | ○ .005 [-.021, .031] | ▲ .013 [.002, .024] |
| Ministral-3-8B () | ○ .021 [-.003, .044] | ▲ .047 [.022, .074] | ○ .007 [-.040, .055] | ▲ .054 [.030, .078] |
| Qwen3.5-9B () | ▼ -.014 [-.023, -.006] | ▲ .015 [.008, .021] | ○ -.012 [-.027, .003] | ○ .001 [-.008, .011] |
| Qwen3.5-9B () | ○ .002 [-.017, .021] | ▲ .032 [.013, .050] | ○ -.023 [-.050, .004] | ▲ .031 [.010, .051] |
| Gemma-4-12B () | ○ -.006 [-.013, .000] | ▲ .028 [.022, .034] | ○ .007 [-.005, .019] | ○ .006 [-.002, .013] |
| Gemma-4-12B () | ▼ -.018 [-.034, -.003] | ○ .012 [-.000, .025] | ▲ .023 [.003, .046] | ○ .015 [-.000, .031] |
| Claude Opus 5 (surr.) () | ▼ -.026 [-.035, -.016] | ▲ .014 [.005, .022] | ▲ .054 [.031, .078] | ▲ .050 [.037, .061] |
| Claude Opus 5 (surr.) () | ○ -.002 [-.022, .019] | ▲ .030 [.009, .051] | ▲ .102 [.058, .147] | ▲ .108 [.081, .135] |
| GPT-5.6 Sol (surr.) () | ▼ -.030 [-.039, -.020] | ▲ .022 [.014, .031] | ○ -.006 [-.021, .010] | ▼ -.016 [-.024, -.009] |
| GPT-5.6 Sol (surr.) () | ○ -.007 [-.023, .008] | ○ .012 [-.001, .025] | ○ .001 [-.026, .029] | ○ -.002 [-.017, .014] |
| Win (▲) / tie (○) / loss (▼) | 0 / 8 / 4 | 9 / 3 / 0 | 3 / 9 / 0 | 6 / 5 / 1 |
W4. The claim that CiteProbe extends to proprietary models is overstated. The authors pass their answers through an open-source model and use its attention weights to find supporting evidence. They could do the same with ground-truth answers. This only shows that CiteProbe works on externally generated text.
Q2. In the proprietary-model experiments, CiteProbe simply takes the generated answers and uses an open-source model to identify supporting evidence. The same could be done with ground-truth answers. What do these experiments demonstrate about CiteProbe's extension to proprietary models beyond its ability to process any given answer?
Reply: We thank the reviewer for raising this distinction. Indeed, the same approach could be applied to ground-truth or other externally generated answers, which highlights the broader applicability of CiteProbe. However, for proprietary LLMs, external text is precisely all that is available. The question is therefore whether CiteProbe transfers well. For instance, BM25 drops sharply on GPT-5.6 Sol, possibly because GPT-5.6 paraphrases more extensively and reduces lexical overlap. In contrast, CiteProbe continues to maintain strong performance. Therefore, our experiments establish three practical findings.
First, CiteProbe's performance transfers to closed-source models without retraining, despite differences in answer lengths, styles, and paraphrasing.
Second, the results are strong across different surrogates. CiteProbe achieves the highest mean performance on Claude Opus 5, and the highest mean Top-1 on GPT-5.6 Sol.
Third, this makes citation of proprietary-model outputs practical, since CiteProbe requires only a single forward pass.
Nonetheless, we agree that our original wording could be more precise and have revised "extending to proprietary models" to "providing citations for proprietary-model outputs through model surrogacy".
Dear Reviewer YwMe,
Thank you for your detailed review. We are glad that you found the problem relevant and important. Below, we meticulously respond to each comment and outline the corresponding revisions.
Reply: Thank you for highlighting this concern. We agree that recall is important, and can directly inform multi-hop citation performance. The issue of low sporadic low recall occurs because of validation-selected operating points that maximizes F1. Given that an operating point can trade off recall for precision and vice versa, we added Area Under the Precision-Recall Curve (AUPRC) to
Table 1in the main text and precision-recall curves toFigure 15inAppendix M.2. AUPRC evaluates the precision-recall trade-off across all thresholds, alleviating issues with single operating points selected on validation F1. CiteProbe achieves the highest mean AUPRC among all methods on five of six backbones. CiteProbe's stronger overall AUPRC is reflected in its favorable precision-recall trade-offs across most settings, with the ideal curve approaching the upper-right corner inFigure 15. We summarize the AUPRC figures as follows:M denotes Ministral, G denotes Gemma, and Q denotes Qwen surrogates.