- Rating
- 2
- Confidence
- 3
- Informativeness
- 2
- Presentation
- 2
- Soundness
- 1
Read review
Paper summary
This work proposes CiRe, which treats the dot product interaction as matrix factorization and improves it using a probabilistic circuit with an NLL loss. This circuited NLL serves as a replacement of BPR by avoiding its costly negative sampling. Authors then apply this circuited NLL to a broad class of classic-matrix-factorization-based, graph-based and content-based recommender models and empirically show that this approach can improve their top- performance.
Critical strengths
-
This paper introduces a novel approach for matrix-factorization-based recommender systems. Many latent factor recommender models are based on matrix factorization (Section 3.6, Aggarwal, 2016) and are typically trained using SGD or ALS. By viewing these models as circuits, we introduce a neural-network-like structure for their training (Loconte et al, 2025), which can potentially improve their performance.
-
The authors apply this circuit approach to a broad class of classic-matrix-factorization-based, graph-based and content-based recommender models, showing that this approach is general in improving their performance.
-
The authors provide detailed experiments to evaluate their approach, including performance comparison and ablation study.
References:
Lorenzo Loconte, Antonio Mari, et al. What is the relationship between tensor factorizations and circuits (and how can we exploit it)? TMLR, 2025.
Charu C. Aggarwal. Recommender Systems: The Textbook. Springer, 2016.
Critical weaknesses
-
The advantages of the circuit view of matrix factorization are unclear. The authors have discussed the benefits of the combination of the circuit and the probabilistic loss (Section 4), but have not discussed the individual benefits of the circuit. I consider a situation when the circuit is applied to a non-probabilistic loss. For example, classic latent factor models may not use BPR or NLL loss. They can be trained using a squared loss (Section 3.6.4, Aggarwal, 2016), which is non-probabilistic and does not require negative sampling. In this case, is it still beneficial to view as a circuit?
-
The math intuition of the formulations of and is not well-explained. In Section 2,
-
Why must adopt complex-valued embeddings rather than real-valued embeddings?
-
Why does need to partition complex dimensions into disjoint blocks? And what does the term mean in Eq. (4)?
-
Why do such formulations of and help achieve better recommendation performance? Are there any specific reasons that such formulations enable more effective information retrieval?
- Empirical results are not significant.
-
Sections 4.3 and 4.4 show that the circuit does not always effective in improving performance: "Addressing RQ3: neither conditional likelihood nor circuit parametrisation is consistently beneficial", "Addressing RQ4: additional circuit structure does not systematically improve ranking". These results are not significant, as they are too vague to illustrate the effectiveness of circuit. Authors should make the results explicit by showing under what cases circuit can bring relatively consistent performance improvement.
-
Section 4.2 shows that and yield similar ranking performance: "Addressing RQ2, joint and conditional maximum likelihood perform similarly for ranking". This result is still not significant, as it does not tell in what aspects we should prefer one loss over the other. Authors need to compare the advantages of and in some other aspects (e.g., convergence rate, computational cost, etc).
Reference:
Charu C. Aggarwal. Recommender Systems: The Textbook. Springer, 2016.
Clarification: My current assessment is based on the reason that I think authors' circuit view of matrix-factorization-based recommender system is interesting, but they do not provide convincing evidence showing how and why this view is effective in improving recommendation performance.
Questions for authors
See Weaknesses.
10 comments
Answer to Reviewer fdLv (2/6)
| Comparator | SQ: better / within-band / worse | μSQ: better / within-band / worse |
|---|---|---|
| Original recommenders | 28 / 6 / 11 | 32 / 6 / 7 |
| Wider re-tuned originals | 24 / 10 / 11 | 29 / 7 / 9 |
SQ still improves over the wider original in 24/45 settings, while SQ improves in 29/45; clear decreases occur in 11 and 9 settings, respectively. Thus, increasing the capacity of the original embeddings narrows the advantage in some cases but does not remove it, indicating that the observed ranking gains cannot be explained solely by the additional capacity. In the table, “within-band” denotes an absolute Recall@20 difference smaller than 0.002.
The complete per-setting results are:
| Encoder | Model | Clothing | DBbook | Sports | Baby | ML-1M |
|---|---|---|---|---|---|---|
| BPRMF | Original | 0.0620 | 0.2637 | 0.0856 | 0.2997 | |
| Original, wider | 0.0646 | 0.1037 | 0.0888 | 0.3018 | ||
| SQ | 0.2579 | 0.0934 | 0.0740 | |||
| μSQ | 0.0682 | 0.2668 | 0.0995 | 0.3044 | ||
| LightGCN | Original | 0.0629 | 0.2622 | 0.0943 | 0.0980 | 0.2890 |
| Original, wider | 0.0641 | 0.2700 | 0.1005 | 0.0955 | 0.2927 | |
| SQ | ||||||
| μSQ | 0.0691 | 0.2625 | 0.1078 | 0.0956 | 0.3025 | |
| NGCF | Original | 0.0543 | 0.2562 | 0.0924 | 0.0754 | 0.3013 |
| Original, wider | 0.0557 | 0.2552 | 0.0774 | 0.2989 | ||
| SQ | 0.0911 | |||||
| μSQ | 0.0675 | 0.2672 | 0.0994 | 0.0810 | 0.3047 | |
| VBPR | Original | 0.0705 | 0.2686 | 0.1066 | 0.0944 | 0.3014 |
| Original, wider | 0.0743 | 0.1079 | 0.0950 | |||
| SQ | 0.2696 | 0.1168 | 0.2881 | |||
| μSQ | 0.1015 | 0.2684 | 0.0975 | 0.2942 | ||
| BM3 | Original | 0.0711 | 0.0985 | 0.0960 | 0.1748 | |
| Original, wider | 0.2647 | 0.1011 | 0.1793 | |||
| SQ | 0.0675 | 0.2292 | 0.0961 | 0.0803 | ||
| μSQ | 0.0678 | 0.2459 | 0.0823 | 0.2956 | ||
| FREEDOM | Original | 0.1019 | 0.1161 | 0.1049 | 0.2717 | |
| Original, wider | 0.1025 | 0.2956 | 0.1177 | 0.1053 | 0.2761 | |
| SQ | 0.1073 | 0.2825 | ||||
| μSQ | 0.2881 | 0.1250 | 0.1105 | 0.2992 | ||
| PGL | Original | 0.0913 | 0.1142 | 0.1046 | 0.2934 | |
| Original, wider | 0.0939 | 0.2864 | 0.1145 | 0.2927 | ||
| SQ | 0.1050 | 0.2646 | 0.1238 | 0.1016 | 0.3045 | |
| μSQ | 0.2723 | 0.1100 | ||||
| SLMRec | Original | 0.0700 | 0.2233 | 0.0965 | 0.0742 | 0.1844 |
| Original, wider | 0.0717 | 0.2270 | 0.0760 | 0.1827 | ||
| SQ | 0.2365 | 0.0955 | 0.2924 | |||
| μSQ | 0.0732 | 0.1017 | 0.0916 | |||
| SGL | Original | 0.0652 | 0.2652 | 0.0770 | 0.2810 | |
| Original, wider | 0.0642 | 0.2612 | 0.1060 | 0.0789 | 0.2835 | |
| SQ | 0.2730 | 0.1074 | 0.2970 | |||
| μSQ | 0.0701 | 0.1051 | 0.0852 |
Bold marks the best and underline the second-best value among the four rows of each encoder, for each dataset. These results show that increasing the embedding capacity of the original recommenders does not eliminate the observed circuit gains.
Answer to Reviewer fdLv (3/6)
Why the blocks and monotone factor in Eq. (4)? These are structural components of the more general μSOCS family, and their motivation is the expressiveness analysis of Loconte et al. [1]. A single squared circuit and a monotone circuit (one with only non-negative parameters) are incomparable: each represents some distributions compactly that the other can only represent with exponentially more parameters. Summing several compatible squares, a SOCS, can be exponentially more expressive than both (Theorem 2); a complex modulus square is already a sum of two real squares. Multiplying a monotone circuit by a SOCS gives a μSOCS, which contains both classes, since setting either factor to a constant recovers the other. Eq. (4) instantiates this construction for the user–item interaction: the blocks give a sum of complex squares, and is the monotone factor. These results concern circuits over many variables, whereas our interaction has two, the user and the item; they motivate the family rather than guarantee a gain in our setting, which is why we test and empirically (RQ4).
Concretely, the coordinates are partitioned into disjoint blocks and the resulting modulus squares are added: . The main μSQ model used in our experiments has , so no partition is performed. We vary only to study richer sum-of-squares structures while keeping the total width fixed. Increasing therefore changes the interaction structure rather than increasing its width: for example, contains cross-block terms, whereas does not.
The other factor in Eq. (4), , is a positive separable user--item weighting. Multiplying the squared interaction by this factor preserves nonnegativity and admits the weighted Gram-matrix normalisers derived in Section 2. Our main configuration again uses the simplest case, .
Why should these formulations improve recommendation? The complex squared construction is theoretically motivated by expressiveness. A complex interaction satisfies
so it naturally realises a sum of real squares rather than a single real square. Prior work [1] shows that such sum-of-squares constructions are provably more expressive than single-square circuits, providing a principled reason to use the complex formulation.
This increased expressiveness does not guarantee better top- recommendation in every setting, but it gives the model a richer interaction family while preserving nonnegativity and exact tractable normalisation. Our experiments then test whether this theoretical advantage translates into recommendation performance. We find that the strongest results arise when the circuit interaction is combined with likelihood-based training, whereas circuit parametrisation under BPR alone and likelihood training with a conventional dot product are not consistently beneficial.
We will clarify these aspects in the revised version of our manuscript.
[1] Loconte, L., et al. (2025). Sum of Squares Circuits. Proceedings of the AAAI Conference on Artificial Intelligence, 39(18), 19077–19085.
3. “Empirical results are not significant”
We thank the reviewer for this comment. We agree that the conclusions of RQ3 and RQ4 should state more explicitly when circuitisation is effective, and that RQ2 should clarify the practical distinction between joint and conditional maximum likelihood beyond ranking accuracy.
3a. On the Validity of RQ3 and RQ4
RQ3 identifies which part of the recipe is associated with these gains. In our controlled comparison, each circuit head is trained under BPR and conditional maximum likelihood with the same initialisation within that head: Conditional likelihood improves SQ in 9/9 comparisons and μSQ in 8/9. Conversely, circuit parametrisation under BPR alone is not consistently beneficial. Aiming to address their concern on the effectiveness of CiRe recipe, we extended the analysis of RQ3 to the full set of recommender and dataset:
Answer to Reviewer fdLv (1/6)
We thank the reviewer for the thoughtful feedback, recognizing the potential of our approach. We respond to each concern below.
1. “The advantages of the circuit view of matrix factorization are unclear”
Our goal is to revisit the interaction layer of factorisation-based recommenders and turn it into a tractable generative model, without changing the underlying encoder. With maximum-likelihood training, the circuit defines a normalised distribution over user--item interactions and supports probabilistic queries beyond ranking, such as p(u,i), p(u), p(i), p(u∣i), and p(I∈C∣u). The circuit structure makes the computation of the partition function efficient without enumerating the full user--item score table. Joint and conditional normalisation cost O((U+I)d2) and O((I+B)d2), respectively, compared with O(UId) and O(BId) for exact dot-product softmax, where U and I are the numbers of users and items, B is the batch size and d is the embedding width. Our new scalability experiments show that this difference becomes substantial as the catalogue grows (please see “Empirical results are not significant”). We therefore view the main advantage of the circuit formulation as enabling tractable generative modelling and maximum-likelihood learning, rather than as an intrinsic benefit of the parametrisation under a non-probabilistic loss.
Furthermore, we emphasise that our evaluation is not restricted only to recommenders originally trained with BPR. For example, SLMRec uses in-batch InfoNCE and feature-alignment contrastive losses, while BM3 uses a negative-free bootstrap objective. This supports the broader point that the effect of circuitisation cannot be reduced to replacing BPR or removing negative sampling. Consistently, our ablations show that neither circuit parametrisation under BPR nor likelihood training with a conventional dot product is sufficient to explain the gains; the strongest results arise from their combination. A whole-data squared-loss recommender is therefore a relevant non-sampling baseline, but it addresses a different question: it can avoid negative sampling without yielding the same tractable generative model and probabilistic queries. When negative sampling is present, maximum-likelihood training can additionally reduce training cost for lightweight encoders, whereas the trade-off can reverse when propagation or auxiliary objectives dominate computation.
2. “The math intuition of the formulations is not well-explained”
We thank the reviewer for asking for clarifications of the math formulation. The constructions are built on prior works connecting tensor factorisations and tractable circuits and from squared and sum-of-squares probabilistic circuits [Loconte et al., 2023; Loconte et al., 2025a,b]. Our goal is to preserve the factorised user--item computation while turning its output into a nonnegative mass that can be normalised exactly. Complex coordinates play a structural role in this construction. They allow the latent factors to remain unrestricted while, through squaring, producing a real nonnegative interaction, particularly μSQ, where the key operation is the squared modulus. Starting from the complex interaction c(u,i)=∑rzurzir, we define ∣c(u,i)∣2=c(u,i)c(u,i)=(ℜc(u,i))2+(ℑc(u,i))2≥0.
This is important because the embeddings zu and zi remain completely unconstrained: their entries can take any complex values, while the final interaction is guaranteed to be real and nonnegative. SQ uses a related construction,
qSQ(u,i)=[ℜ∑rzurzir]2 where nonnegativity follows from squaring the real-valued score.
Thus, the complex parametrisation naturally realises a sum of two real squares while preserving the factorised structure required for tractable normalisation.
To test whether this additional representation capacity drives CiRe’s gains, instead of the whole proposed recipe, we perform a matched-width analysis: since the circuit heads use complex-valued embeddings, an embedding of width d contains 2d real-valued coordinates, compared with d for the original real-valued embeddings. We therefore independently re-tune all nine original recommenders on all five datasets using real embedding widths {128,512}, corresponding to the real-valued capacity of circuit widths {64,256}. Increasing the width improves the original recommenders in 34/45 settings, but the average gain is only 0.0016 Recall@20. We then compare SQ and μSQ against both the original and the wider re-tuned baselines: