arXiv subjects

Shuying Sun

Publications and source records attributed to Shuying Sun.

8 recordsLinked to original sources

arXiv2026

Generative Search and Recommendation at Shopify

Search and recommendation at Shopify require retrieving relevant products from a changing catalog of billions of products under strict latency constraints. We present SID-LLM, which casts both tasks as generating semantic IDs: short sequences of content-derived tokens representing products. We jointly pre-train SID-LLM on catalog and behavioral data, giving search and recommendation a shared representation and training objective. We show that reinforcement learning further improves judged search relevance. Furthermore, distillation from task-specific experts improves recommendation quality over joint supervised fine-tuning without a statistically detected loss in search quality. Finally, for efficient serving, we pair multi-token prediction with a lightweight ranker, improving retrieval quality over our autoregressive baseline while reducing latency. We evaluate SID-LLM both as a standalone retriever and as a component of Shopify's production search and recommendation systems. As a standalone search system, SID-LLM achieves quality comparable to Shopify's existing multi-model search stack while meeting production latency requirements. Integrating SID-LLM into the search and recommendation stacks provides significant gains in both offline evaluations and online A/B tests. Together, these results demonstrate the practical value of semantic-ID generation as a unified approach to product discovery at scale.

cs.IR↗

arXiv2026

Revisiting Generative vs. Dense Retrieval: A Static Value Prior Reverses the Gap

Generative retrieval (GR) has emerged as a strong candidate generator for sequential recommendation as language models and semantic-item identifiers have improved. However, GR is not a uniformly stronger replacement of dense retrieval (DR) in online retail candidate generation. We examine their differences on two Amazon Toys datasets under different metrics. GR leads on best relevant item value among users both methods cover, while DR leads on exact-target recall and relevant value across the ranked list. With deep analysis, we find that popularity contributes the most to GR's measured value lead. Adding the same static item value to DR in one reranking step reverses this lead and keeps recall above GR. Finally, on an industrial-scale internal e-commerce dataset with LLM-judged relevance, we show that a 14B SID-LLM still underperforms a $47\times$ smaller dense encoder.

cs.IR↗

arXiv2026

T-RoPE: Time-Aware Rotary Position Embedding for Sequential Recommendation

Large-scale recommenders increasingly adopt the sequential generative recipe behind large language models, bringing the Transformer into recommendation along with design choices made for text, including Rotary Position Embedding (RoPE). In language models, RoPE encodes token indices for relative position reasoning, but in recommendation, an interaction index records only event order, saying nothing about elapsed time, behavioral cycles across scales, or calendar phase. We revisit this choice and propose T-RoPE, a time-aware RoPE for sequential generative recommendation that replaces index-only rotation with timestamp-based angles, learnable temporal coefficients, multiscale frequency banks, shifted query alignment, and non-stationary key rotation. We prove that standard RoPE, even on timestamps, remains time-translation invariant and cannot distinguish seasonal contexts, and that T-RoPE breaks this invariance while preserving the RoPE interface. Across five public benchmarks, T-RoPE achieves the best result on every metric on every dataset, improving over the strongest baseline by 78--130\% in HR@10 on the sparse PixelRec data and 8--12\% across metrics on Amazon Books. On an industrial-scale e-commerce dataset with more than 6B interactions, it improves every metric over the HSTU + Time RAB backbone by 13--82\%, with ablations attributing the largest gains to multiscale frequencies ($+56\%$ NDCG@50) and non-stationary keys ($+4\%$). An online A/B test in the Shop app yields positive lifts in conversion rate ($+0.33\%$) and order count ($+0.63\%$). We also provide forward and backward algorithms whose added cost is linear in sequence length and head dimension, keeping time-aware RoPE practical for large generative recommenders.

cs.AI↗

arXiv2026

Formulation and verification of multiscale gyrokinetic simulation of kinetic-MHD processes in toroidal plasmas

A comprehensive gyrokinetic simulation model has been implemented in the global toroidal gyrokinetic code (GTC) and verified for studying low-frequency waves and turbulence in magnetic fusion plasmas by treating all kinetic-MHD processes on an equal footing. A theoretical framework has been formulated to unify various methods for efficiently solving the electron drift kinetic equation in multiscale simulations by separating electron responses into analytic and non-analytic parts based on the smallness parameter of electron-to-ion mass ratio. The model can be reduced to the ideal MHD model with both the linear dispersion relation and the nonlinear ponderomotive force in theory and simulation. The model is used for the verification and validation of simulating internal kink modes in the DIII-D tokamak with accurate calculations of equilibrium parallel current and compressible magnetic perturbation. A large simulation database has been generated to train a surrogate model to predict the kink instability. Statistical analysis shows that the radial location of safety factor q=1 flux-surface and the plasma beta inside the q=1 surface are the most important parameters for predicting the kink instability.

physics.plasm-ph↗

arXiv2022

Reconstruction of tokamak plasma safety factor profile using deep learning

In tokamak operations, accurate equilibrium reconstruction is essential for reliable real-time control and realistic post-shot instability analysis. The safety factor (q) profile defines the magnetic field line pitch angle, which is the central element in equilibrium reconstruction. The motional Stark effect (MSE) diagnostic has been a standard measurement for the magnetic field line pitch angle in tokamaks that are equipped with neutral beams. However, the MSE data are not always available due to experimental constraints, especially in future devices without neutral beams. Here we develop a deep learning-based surrogate model of the gyrokinetic toroidal code for q profile reconstruction (SGTC-QR) that can reconstruct the q profile with the measurements without MSE to mimic the traditional equilibrium reconstruction with the MSE constraint. The model demonstrates promising performance, and the sub-millisecond inference time is compatible with the real-time plasma control system.

physics.plasm-ph↗

arXiv2020

Embedding-based Retrieval in Facebook Search

Search in social networks such as Facebook poses different challenges than in classical web search: besides the query text, it is important to take into account the searcher's context to provide relevant results. Their social graph is an integral part of this context and is a unique aspect of Facebook search. While embedding-based retrieval (EBR) has been applied in eb search engines for years, Facebook search was still mainly based on a Boolean matching model. In this paper, we discuss the techniques for applying EBR to a Facebook Search system. We introduce the unified embedding framework developed to model semantic embeddings for personalized search, and the system to serve embedding-based retrieval in a typical search system based on an inverted index. We discuss various tricks and experiences on end-to-end optimization of the whole system, including ANN parameter tuning and full-stack optimization. Finally, we present our progress on two selected advanced topics about modeling. We evaluated EBR on verticals for Facebook Search with significant metrics gains observed in online A/B experiments. We believe this paper will provide useful insights and experiences to help people on developing embedding-based retrieval systems in search engines.

cs.IR↗

arXiv2017

Spectral and Combinatorial Properties of Some Algebraically Defined Graphs

Let $k\ge 3$ be an integer, $q$ be a prime power, and $\mathbb{F}_q$ denote the field of $q$ elements. Let $f_i, g_i\in\mathbb{F}_q[X]$, $3\le i\le k$, such that $g_i(-X) = -\, g_i(X)$. We define a graph $S(k,q) = S(k,q;f_3,g_3,\cdots,f_k,g_k)$ as a graph with the vertex set $\mathbb{F}_q^k$ and edges defined as follows: vertices $a = (a_1,a_2,\ldots,a_k)$ and $b = (b_1,b_2,\ldots,b_k)$ are adjacent if $a_1\ne b_1$ and the following $k-2$ relations on their components hold: $$ b_i-a_i = g_i(b_1-a_1)f_i\Bigl(\frac{b_2-a_2}{b_1-a_1}\Bigr)\;,\quad 3\le i\le k. $$ We show that graphs $S(k,q)$ generalize several recently studied examples of regular expanders and can provide many new such examples.

math.CO↗

arXiv2017

The Eigenvalues of the Graphs $D(4,q)$

The graphs $D(k,q)$ have connected components $CD(k,q)$ giving the best known bounds on extremal problems with {\em forbidden\/} even cycles, and are denser than the well-known graphs of Lubotzky, Phillips, Sarnak and Margulis. Despite this, little about the spectrum and expansion properties of these graphs is known. In this paper we find the spectrum for $k=4$, the smallest open case. For each prime power $q$, the graph $D(4,q)$ is $q$-regular graph on $2q^4$ vertices, all of whose eigenvalues other than $\pm q$ are bounded in absolute value by $2\sqrt{q}$. Accordingly, these graphs are good expanders, in fact very close to Ramanujan.

math.CO↗