Pith. sign in

REVIEW 5 major objections 5 minor 40 references

Theoretical Guarantees for LT-TTD: A Unified Transformer-based Architecture for Two-Level Ranking Systems

T0 review · 5 major / 5 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read The paper claims a unified retrieval–reranking architecture, LT-TTD, provably cuts the upper limit on irretrievable relevant items.

desk verdict The paper's headline error-propagation bound is algebraically unsound, and the rest of the theory is either trivially true or rests on unstated assumptions. read the letter →

arxiv 2505.04434 v1 pith:ESJBITJH submitted 2025-05-07 cs.IR stat.ML

Ayoub Abraich This is my paper · ORCID
classification cs.IRstat.ML
keywords two-levelrankingcascaderetrievaltwo-towermodelslistwisetransformerknowledgedistillationmulti-objectiveoptimizationerrorpropagationUPQE
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Modern search and recommendation systems usually filter billions of candidates in two isolated stages: a fast retrieval model narrows the set, then a slower reranker orders it. The paper argues that this separation is the root of two failures: relevant items dropped in the first stage can never be recovered, and the two stages' objectives can fight each other. It proposes LT-TTD, an architecture that couples a two-tower encoder with a listwise transformer through a bidirectional knowledge-distillation bridge, and claims formal guarantees that this coupling reduces the upper bound on irretrievable relevant items by a factor depending on distillation strength, reaches at least as good a joint optimum as disjoint training, converges in $O(\log(1/\epsilon))$ steps under smoothness assumptions, and keeps inference cost at $O(d\log N + k^2d)$. A sympathetic reader would care because these would be rare analytic guarantees for a class of systems normally tuned empirically.

What carries the argument

The load-bearing object is the Knowledge Distillation Bridge (KDB), a set of three losses connecting the two towers and the listwise transformer: forward KL distillation from transformer to two-tower, backward mean-squared score alignment, and embedding alignment through a learned linear map. In the theory, the bridge is what makes LT-TTD's miss probability an interpolation $1 - [P_{\mathrm{TTE}} + \alpha(P_{\mathrm{LT}} - P_{\mathrm{TTE}})]$ between the two-tower and transformer recall rates, with $\alpha$ as the distillation strength. The multi-objective loss $L_{\mathrm{total}} = \lambda_1 L_{\mathrm{retrieve}} + \lambda_2 L_{\mathrm{rank}} + \lambda_3 L_{\mathrm{forward}} + \lambda_4 L_{\mathrm{backward}} + \lambda_5 L_{\mathrm{align}}$ carries the optimality argument, and the UPQE metric carries the evaluation argument.

What would settle it

On a public retrieval benchmark, train LT-TTD and a traditional cascade with the same two-tower and reranker backbones, then for held-out queries compute the empirical miss ratio $E_{\mathrm{LT-TTD}}/E_{\mathrm{cascade}}$ and compare it to $(1-\alpha\beta)$ for the best-fitting $\alpha,\beta\in[0,1]$; if no constants fit the data, Theorem 3's assumed form for the miss probability is contradicted. A cheaper check is to measure $P_{\mathrm{LT}}-P_{\mathrm{L2}}$ across queries and test whether it is constant, as equation (43) requires.

Watch

Extended reading notes

Core claim

The paper's central claim is that unifying the retrieval and reranking stages, rather than optimizing them separately, provably raises the ceiling of a two-level ranking system. It states that with LT-TTD the expected number of relevant items missed before reranking satisfies $E_{\mathrm{LT-TTD}} \le (1-\alpha\beta)\,E_{\mathrm{cascade}}$, where $\alpha$ measures how much the listwise transformer's signal improves retrieval through distillation and $\beta$ measures the transformer's ranking advantage over the cascade's reranker. It also states that the multi-objective loss $\lambda_1 L_{\mathrm{retrieve}}+\lambda_2 L_{\mathrm{rank}}$ has a joint optimum no worse than the disjoint optimum, because minimizing the sum cannot be beaten by parameters chosen for the summands separately. On the evaluation side, it introduces UPQE, a metric combining relative NDCG, the fraction of relevant items not lost to propagation, and a cost ratio, and proves monotonicity and convergence properties of that metric.

Load-bearing premise

The error-reduction theorem rests on the unproven assumption, stated in equations (42) and (43), that the unified model's miss probability is the cascade's miss probability improved by an interpolation term $\alpha(P_{\mathrm{LT}}-P_{\mathrm{TTE}})$ and that the transformer's recall exceeds the reranker's by a fixed constant $\beta$; if real miss probabilities do not take these forms, the $(1-\alpha\beta)$ factor is an assumption, not a derived bound.

Editorial extensions

If this is right

  • If Theorem 3 holds for real models, a unified retrieval–reranking system can cut the worst-case upper limit on relevant items lost before reranking by a factor $(1-\alpha\beta)$ relative to a matched cascade.
  • Theorem 4 implies that a single network trained on a weighted sum of retrieval and ranking losses cannot be outperformed, on that same weighted loss, by two separately trained networks; unification costs nothing at the level of loss optimization.
  • The $O(\log(1/\epsilon))$ convergence claim means that under the paper's smoothness and convexity assumptions, training should reach a fixed accuracy in a number of epochs logarithmic in the desired error.
  • The complexity bound $O(d\log N + k^2d)$ says the unified model stays deployable at web scale as long as the reranking set size $k$ is kept small, matching the budget of current two-stage systems.
  • UPQE offers a single number for comparing models on recall, ranking quality, and compute jointly, which could make architecture comparisons in this area more direct.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The constants $\alpha$ and $\beta$ are never derived from the architecture, data, or distillation schedule; a testable extension is to fit them on held-out queries and check whether $(1-\alpha\beta)$ actually predicts the miss-rate ratio.
  • The optimality theorem's force is mostly conceptual: it restates that a weighted sum is minimized by minimizing the sum, so the open question shifts to whether one shared parameter set can serve both the towers and the transformer without capacity conflicts.
  • The same proof template could extend to other hybrid retrieval architectures, such as late-interaction or poly-encoder models, whenever one component's behavior can be written as a distillation-strength interpolation of another's.
  • An empirical study that reports $P_{\mathrm{LT}}$, $P_{\mathrm{TTE}}$, and $P_{\mathrm{L2}}$ separately would convert the paper's posited identities into measurable claims.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes LT-TTD, a unified retrieval-ranking architecture combining a two-tower encoder, a listwise transformer, and a bidirectional knowledge distillation bridge. It claims formal guarantees: Theorem 3 bounds error-propagation reduction relative to a cascade system, Theorem 4 proves global optimality of the multi-objective training, Theorem 5 gives a convergence rate, Theorem 6 analyzes computational complexity, Theorem 7 bounds distillation error, Theorem 8 proves ranking-quality improvement, and Theorems 9–11 establish properties of a new evaluation metric, UPQE. The manuscript is purely theoretical and contains no experiments or empirical validation.

Significance. The problem addressed—error propagation and objective mismatch in cascade retrieval-ranking systems—is practically important, and Section 3 usefully formalizes several limitations of decoupled designs. The paper is clearly written and the complexity analysis in Theorem 6 is plausible. However, the central theoretical contribution, Theorem 3, is not proven: the derivation from Eqs. (42)–(43) to Eq. (44) is algebraically invalid, and Theorem 8 inherits that failure. Because the abstract's headline guarantee depends on Theorem 3, the promised theoretical support is not delivered. Credit is due for explicitly identifying the irreversibility of L1 filtering errors and for proposing a concrete architecture and a propagation-aware evaluation metric, but the core proofs need substantial repair.

major comments (5)
  1. [6.1, Theorem 3 (Eqs. 42–44)] The proof of Eq. (44) does not follow from Eqs. (42) and (43). Substituting (43) into (42) gives a miss probability of 1 - [(1-α)P_TTE + α P_L2 + αβ], whereas Eq. (44) asserts E_LT-TTD = (1 - αβ)E_cascade with E_cascade = 1 - P_L1. No relation among P_L1, P_TTE, and P_L2 is stated that would justify this equality. A concrete assignment satisfying the stated conditions (P_TTE = P_L2 = 0, P_L1 = 0.5, α = 1, β = 0.1) yields E_LT-TTD = 0.9 and (1-αβ)E_cascade = 0.45, violating the claimed inequality. The flagship reduction bound is therefore unproven.
  2. [6.1, definitions of α and β] The constants α and β are not defined in terms of the model, training data, or distillation schedule; Eq. (42) postulates a linear-mix form for the LT-TTD miss probability and Eq. (43) postulates that P_LT = P_L2 + β with arbitrary β. The bound therefore restates the assumed superiority of the LT component rather than deriving a guarantee from the architecture or training objective. To make the theorem meaningful, α and β must be expressed via quantities such as distillation loss, model capacity, or retrieval scores.
  3. [6.5, Theorem 7] The inequality chain in the proof of Theorem 7 does not establish the claimed bound. Pinsker's inequality and the asserted mean-value-theorem estimate give lower bounds on the KL divergence and on the L1 distance between distributions, respectively; Jensen's inequality then yields (E|X|)^2 ≤ E[X^2], which is a lower bound on E[X^2], not the required upper bound. Moreover, δ is never defined. The statement E[(s_TTE - s_LT)^2] ≤ L_forward/λ3 + δ is therefore unsupported.
  4. [6.6, Theorem 8] Theorem 8 inherits the failure of Theorem 3: Δ_error > 0 is claimed to follow from Theorem 3, which is not established. The claim Δ_ranking > 0 is also asserted by reference to listwise models in general without a formal argument that LT-TTD's particular training objective guarantees positive NDCG improvement over the cascade's already-ranked output; the decomposition into Δ_error and Δ_ranking is not derived.
  5. [7.2, Theorem 9] Theorem 9 is misstated: for a perfect unified model, Eq. (67) gives UPQE(q) = γ / NDCG_cascade(q), which is greater than γ whenever NDCG_cascade(q) < 1, so γ is not the maximum value of UPQE. Also, the reuse of α and β in the UPQE definition with the same symbols as the distillation-effectiveness and LT-superiority constants in Theorem 3 is confusing and creates an implicit, unproven link between the theorem's free parameters and the metric's hyperparameters.
minor comments (5)
  1. [6.2, Theorem 4] The theorem is true by definition but carries no information: it states that the minimizer of L_joint is no worse than any fixed parameter vector, which holds for any function. The 'provably better global optimum' claim in the abstract overstates this tautology.
  2. [6.3, Theorem 5] The proof assumes μ-strong convexity and L-smoothness for the weighted loss; these assumptions are not justified for transformer-based ranking losses and the displayed contraction factor (1 - min(ηµ, 1/(ηL))) is not the standard rate for Adam or GD on strongly convex functions; the proof is an outline and should be either made rigorous or presented as a conjecture.
  3. [3.1, Proposition 1] The bound is stated without proof and the scope condition 'any evaluation metric that satisfies the probability ranking principle' is not formalized; please either prove it or cite prior work.
  4. [Notation, Eqs. (41)–(44)] Notation is inconsistent: P_L1, P_L2, P_TTE, and P_LT denote both probabilities and expectations in the same proof, and the distinction between the event d_i ∈ D_L and the probability P(d_i ∈ D_L) should be made explicit.
  5. [Overall] The paper provides no experiments or empirical evaluation of LT-TTD or UPQE; given the strong architectural claims, some form of validation would be necessary in a journal version.

Circularity Check

3 steps flagged · score 8.0 of 10

Theorems 3, 4, and 8 state their conclusions by definition: the reduction factor is posited through unmeasured parameters, and the optimality and ranking guarantees are tautological restatements.

  1. self definitional [Section 6.1, Theorem 3, Eqs. (42)-(44)]
    "The probability of a relevant item being missed becomes: P_LT-TTD(...) = 1 − [P_TTE(...) + α·(P_LT(...) − P_TTE(...))] ... Given that the LT component has superior ranking capabilities compared to the L2 component in the cascade system, we can express: P_LT(...) = P_L2(...) + β ... Substituting and taking expectations over all relevant items and queries, we get: E_LT-TTD = (1 − α·β) · E_cascade."

    The quantities α and β are not derived from the model, the data, or the distillation schedule; they are introduced in the proof as 'effectiveness of the knowledge distillation' and 'measure of LT's superiority'. The claimed reduction factor (1−αβ) is simply the product of these two posited parameters. Moreover, substituting (42) and (43) yields 1 − (1−α)P_TTE − αP_L2 − αβ, which is not identically (1−αβ)(1−P_L1); the asserted equality requires an additional unstated relation among P_L1, P_TTE, P_L2, α, and β. The theorem therefore does not derive an error-propagation guarantee from the architecture; it writes down the desired bound as the definition of the parameters.

  2. self definitional [Section 6.2, Theorem 4, proof around Eqs. (45)-(49)]
    "Let θ*_joint be the optimal parameters when optimizing Ljoint(θ). Then: Ljoint(θ*_joint) ≤ Ljoint(θ*_disjoint) ... By definition, θ*_joint = argminθ Ljoint(θ). For any parameter setting θ, we have: Ljoint(θ) = λ1Lretrieve(θ) + λ2Lrank(θ). In particular, for θ*_disjoint ... However, θ*_joint minimizes the weighted sum Ljoint, so Ljoint(θ*_joint) ≤ Ljoint(θ) for any θ, including θ*_disjoint."

    The conclusion of Theorem 4 is exactly the definition of a global minimizer: argminθ Ljoint(θ) is no larger at its optimum than at any other point. The 'disjoint' parameters are just an arbitrary feasible point in the same parameter space, so the inequality holds for any optimizer of Ljoint regardless of how the objectives are trained. No property relating the joint loss to the separate retrieval and ranking objectives is used. The claimed 'provably better global optimum than disjoint training' is therefore a tautology, not a derived guarantee.

1 more flagged steps
  1. self definitional [Section 6.6, Theorem 8, proof around Eqs. (61)-(65)]
    "Combining these two components: Eq[NDCG(q, RLT-TTD)] = Eq[NDCG(q, Rcascade)] + Δ_error + Δ_ranking. Setting γ = Δ_error + Δ_ranking > 0, we get: Eq[NDCG(q, RLT-TTD)] ≥ Eq[NDCG(q, Rcascade)] + γ."

    The constant γ is defined after the fact as the exact NDCG difference it is supposed to guarantee. With γ = Δ_error + Δ_ranking, the displayed inequality is an algebraic identity rather than an independent bound. The positivity of γ rests entirely on the unproved Theorem 3 and on a citation that listwise methods improve ranking, neither of which supplies a numerical or structural value for γ. The theorem thus repackages the desired improvement as a definitional placeholder instead of deriving it.

full rationale

The paper's headline claims are the error-propagation reduction (Theorem 3), better global optimality than disjoint training (Theorem 4), and ranking-quality improvement (Theorem 8). All three are constructed rather than derived. In Theorem 3, the parameters α and β are posited as 'knowledge distillation effectiveness' and 'LT superiority,' and the bound (1−αβ) is the product of those same posited numbers; moreover the algebra from Eqs. (42)–(43) to Eq. (44) does not follow without an extra, unstated relation, so even the restatement is not validly executed. Theorem 4 simply restates the definition of a minimizer, and Theorem 8 defines γ as the very gap to be proved. The remaining results are either conditional routine statements (convergence under strong convexity, complexity arithmetic) or are not load-bearing for the abstract's guarantees. There are no external experiments or benchmarks that would let the architecture's claims be tested independently. Because the central claimed results reduce to definitional posits, the circularity score is high even though no self-citation chain is present.

Assumptions & free parameters 5 free parameters · 5 assumptions · 2 invented entities

The central theoretical claims depend on constants α, β, δ, and γ that are introduced ad hoc, and on assumptions such as strong convexity of losses and a posited miss-probability form. These are neither derived from the architecture nor estimated from data, which explains why the paper's guarantees are largely restatements of definitions.

free parameters (5)
  • α (distillation effectiveness)
    Introduced in Theorem 3 as an arbitrary constant in [0,1] to express the probability of missing a relevant item under LT-TTD; no formula or estimation is given.
  • β (LT superiority)
    Introduced in Theorem 3 as the assumed difference between P_LT and P_L2; no derivation from the architecture is provided.
  • δ (distillation error slack)
    Introduced in Theorem 7 to absorb unaccounted errors in the inequality chain; never quantified or estimated.
  • γ (ranking improvement)
    Introduced in Theorem 8 as the sum Δ_error + Δ_ranking; never lower-bounded or defined in terms of model properties.
  • UPQE hyperparameters α, β, γ
    Tuning parameters in the UPQE metric (Section 7); not fitted to data, but the metric itself is proposed without validation or comparison to existing metrics.
assumptions (5)
  • domain assumption All loss components in L_total are L-Lipschitz smooth and μ-strongly convex in a neighborhood of the optimum.
    Assumed in Theorem 5 to obtain the O(log(1/ε)) convergence bound; no evidence that transformer-based deep network losses satisfy strong convexity.
  • ad hoc to paper The miss probability under LT-TTD takes the linear-mix form in equation (42).
    Unjustified modeling assumption in the proof of Theorem 3.
  • ad hoc to paper The difference between LT's and L2's inclusion probabilities is a constant β.
    Equation (43) in Theorem 3; no derivation from the architecture or training dynamics.
  • domain assumption Listwise ranking methods are provably superior to other approaches in this setting, per Cao et al. 2007.
    Used in Theorem 8 to assert Δ_ranking > 0; the cited result does not imply a positive improvement for LT-TTD specifically.
  • standard math Pinsker's inequality and a Lipschitz property of softmax relate the KL divergence to the absolute score difference with constant C.
    Invoked in the proof of Theorem 7; the chain is not made rigorous, and the constant C is not characterized.
invented entities (2)
  • UPQE metric
    purpose: A new evaluation metric for unified ranking architectures combining NDCG ratio, error propagation penalty, and efficiency ratio.
    The metric is proposed in this paper with no external data, ablation, or comparison to existing metrics to establish its validity.
  • Knowledge Distillation Bridge (KDB)
    purpose: Bidirectional knowledge transfer between the two towers and the transformer to align objectives.
    Part of the proposed architecture; no independent support or ablation is provided.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Theoretical Guarantees for LT-TTD: A Unified Transformer-based Architecture for Two-Level Ranking Systems." pith.science (2026). https://pith.science/paper/ESJBITJH

@misc{pith2026250504434,
  author       = {Pith},
  title        = {Pith review of: Theoretical Guarantees for LT-TTD: A Unified Transformer-based Architecture for Two-Level Ranking Systems},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ESJBITJH}},
  note         = {Machine review of arXiv:2505.04434}
}
read the original abstract

Modern recommendation and search systems typically employ multi-stage ranking architectures to efficiently handle billions of candidates. The conventional approach uses distinct L1 (candidate retrieval) and L2 (re-ranking) models with different optimization objectives, introducing critical limitations including irreversible error propagation and suboptimal ranking. This paper identifies and analyzes the fundamental limitations of this decoupled paradigm and proposes LT-TTD (Listwise Transformer with Two-Tower Distillation), a novel unified architecture that bridges retrieval and ranking phases. Our approach combines the computational efficiency of two-tower models with the expressivity of transformers in a unified listwise learning framework. We provide a comprehensive theoretical analysis of our architecture and establish formal guarantees regarding error propagation mitigation, ranking quality improvements, and optimization convergence. We derive theoretical bounds showing that LT-TTD reduces the upper limit on irretrievable relevant items by a factor that depends on the knowledge distillation strength, and prove that our multi-objective optimization framework achieves a provably better global optimum than disjoint training. Additionally, we analyze the computational complexity of our approach, demonstrating that the asymptotic complexity remains within practical bounds for real-world applications. We also introduce UPQE, a novel evaluation metric specifically designed for unified ranking architectures that holistically captures retrieval quality, ranking performance, and computational efficiency.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

40 extracted references · 26 canonical work pages

  1. [1]

    Effectiveness/efficiency tradeoffs for candidate generation in multi-stage retrieval architectures

    Nima Asadi and Jimmy Lin. Effectiveness/efficiency tradeoffs for candidate generation in multi-stage retrieval architectures. In Proceedings of the 36th international ACM SIGIR conference on Research and development in information retrieval, pages 997--1000, 2013

    work page 2013

  2. [2]

    From ranknet to lambdarank to lambdamart: An overview

    Christopher JC Burges. From ranknet to lambdarank to lambdamart: An overview. Learning, 11 0 (23-581): 0 81, 2010

    2010

  3. [3]

    Learning to rank using gradient descent

    Christopher Burges, Tal Shaked, Erin Renshaw, Ari Lazier, Matt Deeds, Nicole Hamilton, and Greg Hullender. Learning to rank using gradient descent. In Proceedings of the 22nd international conference on Machine learning, pages 89--96, 2005

    work page 2005

  4. [4]

    Learning to rank with nonsmooth cost functions

    Christopher Burges, Robert Ragno, and Quoc Le. Learning to rank with nonsmooth cost functions. In Advances in neural information processing systems, pages 193--200, 2006

    work page 2006

  5. [5]

    Learning to rank: from pairwise approach to listwise approach

    Zhe Cao, Tao Qin, Tie-Yan Liu, Ming-Feng Tsai, and Hang Li. Learning to rank: from pairwise approach to listwise approach. In Proceedings of the 24th international conference on Machine learning, pages 129--136, 2007

    2007

  6. [6]

    Universal sentence encoder

    Daniel Cer, Yinfei Yang, Sheng-yi Kong, Nan Hua, Nicole Limtiaco, Rhomni St John, Noah Constant, Mario Guajardo-Cespedes, Steve Yuan, Chris Tar, et al. Universal sentence encoder. arXiv preprint arXiv:1803.11175, 2018

    arXiv 2018

  7. [7]

    Deep neural networks for youtube recommendations

    Paul Covington, Jay Adams, and Emre Sargin. Deep neural networks for youtube recommendations. In Proceedings of the 10th ACM conference on recommender systems, pages 191--198, 2016

    2016

  8. [8]

    Deeper text understanding for ir with contextual neural language modeling

    Zhuyun Dai and Jamie Callan. Deeper text understanding for ir with contextual neural language modeling. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 985--988, 2019

    work page 2019

Show all 40 references
  1. [9]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Langu...

    2019

  2. [10]

    Real-time personalization using embeddings for search ranking at airbnb

    Mihajlo Grbovic and Haibin Cheng. Real-time personalization using embeddings for search ranking at airbnb. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 311--320, 2018

    2018

  3. [11]

    Distilling the knowledge in a neural network

    Geoffrey Hinton, Oriol Vinyals, and Jeff Dean. Distilling the knowledge in a neural network. arXiv preprint arXiv:1503.02531, 2015

    2015 arXiv

  4. [12]

    Learning deep structured semantic models for web search using clickthrough data

    Po-Sen Huang, Xiaodong He, Jianfeng Gao, Li Deng, Alex Acero, and Larry Heck. Learning deep structured semantic models for web search using clickthrough data. In Proceedings of the 22nd ACM international conference on Information & Knowledge Management, pages 2333--2338, 2013

    2013

  5. [13]

    Embedding-based retrieval in facebook search

    Jui-Ting Huang, Ashish Sharma, Shuying Sun, Linjun Xia, David Zhang, Philip Pronin, Janani Padmanabhan, Giuseppe Ottaviano, and Linjun Yang. Embedding-based retrieval in facebook search. In Proceedings of the 26th ACM SIGKDD International Conference on Knowledge Discovery & Da...

    2020

  6. [14]

    Poly-encoders: Architectures and pre-training strategies for fast and accurate multi-sentence scoring

    Samuel Humeau, Kurt Shuster, Marie-Anne Lachaux, and Jason Weston. Poly-encoders: Architectures and pre-training strategies for fast and accurate multi-sentence scoring. In International Conference on Learning Representations, 2020

    2020

  7. [15]

    a rvelin and Jaana Kek \

    Kalervo J \"a rvelin and Jaana Kek \"a l \"a inen. Cumulated gain-based evaluation of ir techniques. In ACM Transactions on Information Systems (TOIS), volume 20, pages 422--446. ACM New York, NY, USA, 2002

    2002

  8. [16]

    Optimizing search engines using clickthrough data

    Thorsten Joachims. Optimizing search engines using clickthrough data. In Proceedings of the eighth ACM SIGKDD international conference on Knowledge discovery and data mining, pages 133--142, 2002

    2002

  9. [17]

    Colbert: Efficient and effective passage search via contextualized late interaction over bert

    Omar Khattab and Matei Zaharia. Colbert: Efficient and effective passage search via contextualized late interaction over bert. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 39--48, 2020

    2020

  10. [18]

    Click-through prediction for advertising in twitter timeline

    Cheng Li, Yiming Lu, Qiaozhu Mei, Dong Wang, and Sandeep Pandey. Click-through prediction for advertising in twitter timeline. In Proceedings of the 21th ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, pages 1959--1968, 2015

    1959

  11. [19]

    Learning to rank for information retrieval

    Tie-Yan Liu. Learning to rank for information retrieval. Foundations and Trends in Information Retrieval, 3 0 (3): 0 225--331, 2009

    2009

  12. [20]

    Anatomy of a web-scale recommender system

    Yiming Lu, Ruobing Fang, Vanja Josifovski, and Caiming Xiong. Anatomy of a web-scale recommender system. In Proceedings of the 14th ACM International Conference on Web Search and Data Mining, pages 294--303, 2021

    2021

  13. [21]

    Repbert: Contextualized text embeddings for first-stage retrieval

    Dan Luan, ChengXiang Zhai, and Heng Ji. Repbert: Contextualized text embeddings for first-stage retrieval. arXiv preprint arXiv:2006.15498, 2020

    2006 arXiv

  14. [22]

    Cedr: Contextualized embeddings for document ranking

    Sean MacAvaney, Andrew Yates, Arman Cohan, and Nazli Goharian. Cedr: Contextualized embeddings for document ranking. In Proceedings of the 42nd International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1101--1104, 2019

    2019

  15. [23]

    Discriminative models for information retrieval

    Ramesh Nallapati. Discriminative models for information retrieval. In Proceedings of the 27th annual international ACM SIGIR conference on Research and development in information retrieval, pages 64--71, 2004

    2004

  16. [24]

    Passage re-ranking with bert

    Rodrigo Nogueira and Kyunghyun Cho. Passage re-ranking with bert. arXiv preprint arXiv:1901.04085, 2019

    1901 arXiv

  17. [25]

    Multi-stage document ranking with bert

    Rodrigo Nogueira, Wei Yang, Kyunghyun Cho, and Jimmy Lin. Multi-stage document ranking with bert. arXiv preprint arXiv:1910.14424, 2019

    1910 arXiv

  18. [26]

    Setrank: Learning a permutation-invariant ranking model for information retrieval

    Liang Pang, Yanyan Lan, Jiafeng Guo, Jun Xu, Jingfang Xu, and Xueqi Cheng. Setrank: Learning a permutation-invariant ranking model for information retrieval. In Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, pag...

    2020

  19. [27]

    Learning diverse rankings with multi-armed bandits

    Filip Radlinski, Robert Kleinberg, and Thorsten Joachims. Learning diverse rankings with multi-armed bandits. In Proceedings of the 25th international conference on Machine learning, pages 784--791, 2008

    2008

  20. [28]

    Sentence-bert: Sentence embeddings using siamese bert-networks

    Nils Reimers and Iryna Gurevych. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP), ...

    2019

  21. [29]

    The probabilistic relevance framework: Bm25 and beyond

    Stephen Robertson and Hugo Zaragoza. The probabilistic relevance framework: Bm25 and beyond. Foundations and Trends in Information Retrieval, 3 0 (4): 0 333--389, 2009

    2009

  22. [30]

    Okapi at trec-3

    Stephen Robertson, Steve Walker, Susan Jones, Micheline Hancock-Beaulieu, and Mike Gatford. Okapi at trec-3. In TREC, volume 3, pages 109--126, 1994

    1994

  23. [31]

    Ranking distillation: Learning compact ranking models with high performance for recommender system

    Jiaxi Tang and Ke Wang. Ranking distillation: Learning compact ranking models with high performance for recommender system. In Proceedings of the 24th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining, pages 2289--2298, 2018

    2018

  24. [32]

    Softrank: optimizing non-smooth rank metrics

    Michael Taylor, John Guiver, Stephen Robertson, and Tom Minka. Softrank: optimizing non-smooth rank metrics. In Proceedings of the 2008 International Conference on Web Search and Data Mining, pages 77--86, 2008

    2008

  25. [33]

    Attention is all you need

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N Gomez, ukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Advances in neural information processing systems, pages 5998--6008, 2017

    2017

  26. [34]

    Joint optimization of cascade ranking models

    Xuanhui Wang, Jost F Gemmeke, Cheng Zhu, Dhivya K Chivukula, Boulos Charbuty, and Aditya Muralidharan. Joint optimization of cascade ranking models. In Proceedings of the 14th ACM International Conference on Web Search and Data Mining, pages 882--890, 2021

    2021

  27. [35]

    Adapting boosting for information retrieval measures

    Qiang Wu, Christopher JC Burges, Krysta M Svore, and Jianfeng Gao. Adapting boosting for information retrieval measures. Information Retrieval, 13 0 (3): 0 254--270, 2010

    2010

  28. [36]

    Listwise approach to learning to rank: theory and algorithm

    Fen Xia, Tie-Yan Liu, Jue Wang, Wensheng Zhang, and Hang Li. Listwise approach to learning to rank: theory and algorithm. In Proceedings of the 25th international conference on Machine learning, pages 1192--1199, 2008

    2008

  29. [37]

    Mixed negative sampling for learning two-tower neural networks in recommendations

    Ruoxi Yang, Xu Yi, Zhen Chen, and Edward Y Chang. Mixed negative sampling for learning two-tower neural networks in recommendations. Proceedings of the 29th International Conference on World Wide Web, pages 441--443, 2020

    2020

  30. [38]

    A two-tower deep neural network for ad relevance prediction

    Xing Yi, Bo Zhou, Qing Zhang, and Alexander Wang. A two-tower deep neural network for ad relevance prediction. In Proceedings of the 3rd International Workshop on Ad Targeting at Scale, pages 1--6, 2019

    2019

  31. [39]

    A support vector method for optimizing average precision

    Yisong Yue, Thomas Finley, Filip Radlinski, and Thorsten Joachims. A support vector method for optimizing average precision. In Proceedings of the 30th annual international ACM SIGIR conference on Research and development in information retrieval, pages 271--278, 2007

    2007

  32. [40]

    Atrank: An attention-based user behavior modeling framework for recommendation

    Chang Zhou, Jinze Bai, Junshuai Song, Xiaofei Liu, Zhiwei Zhao, Xiaolong Chen, and Jun Gao. Atrank: An attention-based user behavior modeling framework for recommendation. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 32, 2018

    2018

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.