CS Paper Radar

From Gradient-Boosted Trees to Deep Recommenders: Practical Lessons from Migrating a Production Customer Support Recommender

Sonia Sharma, Jeyendran Balakrishnan, Shreya Rajpal, Swapnil Parekh, Nagaraj Janardhana, Andrew Mattarella-Micke

cs.LG cs.AI ·submitted 2026-08-25 06:50 UTC · v1
benchmark

Abstract

Product catalogs in fast-moving service businesses are shifting from static, independently priced SKUs toward dynamically bundled, discount-coupled offerings--a shift that strains the tree-based classifiers traditionally preferred for sparse and highly imbalanced data. These classifiers assume a fixed, slowly changing label space and struggle to incorporate multimodal signals such as tabular data and transcripts. We present the migration of a live, production conversational recommendation system from a gradient-boosted multiclass model to a pairwise-binary deep recommender. Because this system is critical to ecosystem growth initiatives and downstream features like dynamic pitching--surfacing the most relevant pitch text to a support agent in real time during a live customer conversation--maintaining live recommendation quality was a non-negotiable constraint. We detail the techniques that made this migration successful--reformulating recommendation as pairwise binary prediction to learn jointly from user and item features, and enhancing learned representations via negative sampling and noise injection. To efficiently incorporate long, live conversation context, we apply attention pooling over transcript chunks and benchmark it against TF-IDF and sentence-embedding baselines. Finally, we explore multiple architectures (including two-tower models, DeepFM, and their variants) and loss functions such as contrastive loss. Evaluating against a CatBoost baseline across all conversational stages, we demonstrate that our approach achieves parity at conversation beginning and outperforms at later conversational stages.

Comment
10 pages

Closest papers

Nearest neighbours by meaning across the whole index.

LightCRS: Lightweight Multi-Domain Conversational Recommendation System

cs.AI ·2026-10-18 ·He Ma, Lan You, Qiyang Wu, Chen Liu
benchmark similarity 0.70
Abstract

Conversational recommender systems have recently adopted vision-language models to match user dialogue intents with multimodal product information. However, extending these methods to many product domains faces a dilemma: deploying a separately adapted VLM per domain scales cost linearly with the number of domains and requires routing each dialogue to the right model, while naively merging all domains into a single VLM degrades accuracy because product characteristics vary widely across categories. We propose LightCRS, a lightweight, query-conditioned framework for unified multi-domain multimodal conversational recommendation. LightCRS precomputes multimodal product representations once offline with frozen encoders. Online, a compact 5.3M-parameter module fuses this evidence for each dialogue via query-conditioned cross-attention, whose parameters are shared across all domains, and compares candidates jointly with an inter-candidate Transformer. To enable rigorous evaluation, we construct MM-R2A-25, a 25-domain benchmark pairing natural Reddit recommendation-seeking conversations with Amazon product listings. LightCRS achieves state-of-the-art average HR@1, NDCG@5, and MRR, improving HR@1 by 35.4\% over the strongest baseline while reducing per-query latency by 266 times. Joint multi-domain training itself improves HR@1 by 15.2\% over separately trained single-domain models.

0.12

PACERec: Progressive Adaptation of LLM Item Embeddings with Collaborative Evidence for Recommendation

cs.IR cs.AI ·2026-11-02 ·Junzhao Huang, Xi Wu, Zihan Wang, Zijing Wang, Ming Wang +7 more
similarity 0.66
Abstract

Large language model (LLM) embeddings have been increasingly adopted in sequential recommendation to encode static item attributes and alleviate the limitations of item ID embeddings under sparse interactions. However, these embeddings typically capture only static content similarity while neglecting sequential transitions, and since their adaptation quality relies on sparse per-item updates or global geometric calibration, existing methods still struggle to explicitly transfer preceding interaction contexts from frequent to long-tail items under unevenly distributed behavioral evidence, limiting recommendation effectiveness across the catalog. To address this challenge, we propose PACERec, an ID-free progressive adaptation framework that incorporates preceding interaction contexts into semantic representation adaptation by learning a shared velocity field from items with sufficient interactions via frequency-weighted flow matching. In this way, PACERec progressively guides static LLM embeddings toward collaborative behavioral targets while preserving original item semantics, overcoming sparse per-item updates to effectively generalize sequential transition patterns to long-tail items for both user modeling and candidate scoring. Extensive experiments on three real-world datasets across three sequential backbones demonstrate the superiority of our method over competitive baselines in both overall and long-tail recommendation.

0.19

The Vocabulary Is Already Here: Adapting Pretrained Interpretable Features for Generative Recommendation

cs.IR cs.AI ·2026-10-27 ·Ziluowen Luo, Ruochen Liu, Jun Yin, Hao Yan, Zhengxuan Hu +4 more
similarity 0.66
Abstract · 13 pages, 8 figures, 7 tables

Generative recommendation (GR) predicts the next item through its semantic ID (SID). Existing methods commonly fit SID vocabularies to pretrained language model (PLM) representations of item descriptions, incurring additional training costs and making vocabulary coverage dependent on the fitting domains. We observe that pretrained sparse autoencoders expose PLM features with human-interpretable semantic annotations, allowing feature indices to serve as a reusable SID vocabulary. However, reusing this vocabulary requires preserving item distinctions under compact token budgets and learning to favor target items over candidates with shared dominant features. It also requires extending feature interpretability from item content to retrieval decisions. To address these challenges, we propose API, a framework that Adapts the Pretrained Interpretable vocabulary for effective and explainable GR without additional vocabulary learning. Within API, Pretrained Interpretable IDs (PIDs) retain selected feature indices and relative activation strengths for item discrimination. With PIDs as supervision, the Semantic Prediction Adapter (SPA) learns user preferences from interaction histories through joint feature-index and strength prediction. SPA combines both predictions to score candidate items, with item-level supervision encouraging the observed next item to outrank its alternatives. API further explains candidate score differences through feature contributions paired with their pretrained semantic annotations. Extensive experiments show that API achieves competitive performance across backbones and decoding schemes, while preserving item distinctions under compact token budgets and enabling explainable retrieval.

0.24

Preference-Drift-Aware Subsequence Learning and Hierarchical Context Fusion for Long-Sequence Generative Recommendation

cs.IR ·2026-09-11 ·Fei Li, Qingyun Gao, Jianzhe Zhao, Guibing Guo, Beibei Kong +3 more
similarity 0.65
Abstract

Long-sequence generative recommendation methods autoregressively model the user's interaction sequence to generate the next-item representation. Existing methods generally fall into two categories: efficient full-sequence modeling and target-aware context retrieval. Our experiments reveal that as the sequence length increases, the former incurs steadily growing computational cost while its accuracy gains quickly saturate and even degrade due to noise; the latter, though shortening the input sequence, is susceptible to noise that is semantically consistent yet preference-inconsistent, as well as to incomplete contexts. Both paradigms ignore the dynamic changes of user preferences and the cross-subsequence dependencies when handling historical information, thereby limiting accuracy and efficiency. To address these issues, we propose a preference-drift-aware subsequence learning and hierarchical context fusion for long-sequence generative recommendation. Specifically, we learn differentiable soft subsequence boundaries using multidimensional preference-drift information and aggregate items within each subsequence into preference-coherent representations via linear attention with soft assignment weights, thereby circumventing the expense of full-sequence attention. A cross-attention mechanism is then employed to capture dependencies between recent interactions and relevant subsequence contexts, mitigating noise in learning recent-item representations. Finally, a gated fusion mechanism adaptively combines the recent-item representation with the global subsequence context, allowing the resulting target representation to encode both recent and long-term preferences. Extensive experiments demonstrate that our method consistently outperforms existing baselines in both recommendation accuracy and computational efficiency.

0.13

Inherit4Rec: Parameter Inheritance for Efficient Scaling of Recommendation Models

cs.IR ·2026-09-19 ·Ruihao Zhang, Bo Chen, Xiao Wang, Jinlong Jiao, Tijian Hu +6 more
similarity 0.65
Abstract

Scaling model capacity has emerged as an effective approach to overcoming performance bottlenecks in industrial recommender systems. However, repeatedly training new models from scratch requires large amounts of data and substantial training resources. Parameter inheritance provides a promising route for both dense model growth and sparse conversion, yet existing methods for LLMs are primarily designed for static corpora and can suffer sharp performance drops under dynamically evolving recommendation data. To address these challenges, we propose Inherit4Rec, a parameter-inheritance framework that supports both Dense-to-Dense (D2D) growth and Dense-to-Sparse (D2S) conversion. Inherit4Rec-D2D combines hybrid growth with asymmetric training to preserve the forward function at expansion and maintain update continuity. Inherit4Rec-D2S constructs SMoE networks through co-activation-aware partitioning with a load-balancing loss, preserving dense-model capabilities while promoting balanced expert activation. Experiments on KuaiRand-1K and an industrial short-video recommendation dataset show that both transformations consistently outperform the SOTA inheritance baselines across all prediction objectives. These results demonstrate the effectiveness of Inherit4Rec for continual capacity expansion and computation-efficient sparse conversion in industrial recommender systems.

0.10

Cut Binary Cross Entropy: Efficient Large-Vocabulary Loss and Gradient Kernels for Sequential Recommendation

cs.LG cs.AIcs.ARcs.IR ·2026-10-04 ·Yaoyiran Li, Haowen Ning, Mohamed Hammad
code/data similarity 0.65
Abstract

Industrial sequential recommender systems operate over massive item catalogs (e.g., 10^5--10^7 items). Multi-label recommendation models are trained with Binary Cross-Entropy (BCE) loss over the full vocabulary, but standard BCE materializes a dense [B, N, V] logits tensor in High Bandwidth Memory (HBM), incurring prohibitive $O(BNV)$ memory and fatal Out-Of-Memory (OOM) errors. While chunked loss optimizations exist for Softmax Cross-Entropy in LLMs, large-scale multi-label BCE optimization remains unexplored across deep learning ecosystems. We propose CutBCE, an exact, hardware-accelerated BCE loss and gradient operator implemented in JAX and Pallas for large-vocabulary workloads. CutBCE introduces (1) an exact fused reformulation evaluating dense background loss and sparse target corrections; (2) a custom Vector-Jacobian Product (VJP) with a dedicated Pallas TPU backward kernel computing logit tiles on-chip in both passes so logits and their gradients never reside in HBM; (3) dynamic VMEM budgeting and sharding-aware collective hoisting for distributed meshes; and (4) count-based zero-overhead training metrics. On single-chip TPU v5e/v6e mini-benchmarks, CutBCE eliminates OOM errors with up to 91.9% speedup. On 8-chip TPU slice training for multi-label SASRec with 876k items (Yambda-50M), CutBCE reduces peak HBM by 65.7% (>14 GiB saved per chip) and increases training speed by 225.9% with comparable accuracy. CutBCE is open-sourced at https://github.com/AI-Hypercomputer/RecML/blob/main/recml/core/ops/binary_cross_entropy_ops.py.

0.30

RedMall: Generative Recommendation for Large-Scale E-commerce

cs.IR ·2026-10-26 ·Kunyang Zhou, Xin Jia, Yi Yang, Yongjun Yang, Manjie Xu +5 more
hot direction similarity 0.64
Abstract · Work in progress

Generative recommendation has shown considerable potential in industrial recommendation systems. However, the performance of existing generative methods is limited by several challenges. 1) Semantic IDs (SIDs) are treated as one-dimensional sequences, overlooking the hierarchical dependencies among different layers. 2) Next-token prediction used is unable to explicitly optimize session-wise recommendation. 3) Although reinforcement learning has been applied to align generated recommendations with user preferences, few existing methods explicitly optimize the ranking of positive items in session-wise recommendation. To address the above issues, we propose RedMall, a generative recommender designed for large-scale e-commerce. Firstly, RedMall constructs hierarchical SIDs through residual quantization over multimodal representations. Secondly, RedMall introduces a lightweight SID encoder to explicitly model layer-wise dependencies in SIDs. Moreover, RedMall incorporates sparse mixture-of-experts, dual gated attention, and multi-target prediction to scale model size, enhance user interest representation, and optimize session-wise recommendation. We further design an exposure-aware loss that treats exposed but unclicked items as negative feedback. To better align user preferences and list-wise ranking objectives, we combine Group Relative Policy Optimization with an auxiliary LambdaRank loss, using both real user feedback and item ranking to guide optimization. Experiments on four public datasets-Amazon Beauty, Amazon Toys, Amazon Food, and MovieLens-1M---show that RedMall outperforms the evaluated baselines across all reported Recall and NDCG metrics. Online A/B tests show improvements of 2.24% in advertiser value (ADVV) and 0.59% in click-through rate (CTR) for e-commerce advertising, as well as gains of 1.08% in GMV and 0.95% in daily active buyers (DAB) for e-commerce recommendations.

0.24

Unified Heterogeneous Feature Interaction via Effective-Rank Scaling for Recommender Systems

cs.IR ·2026-10-26 ·Ao Zhang, Haochuan Li, Boya Du, Jian Yang, Qihang Fang +1 more
similarity 0.64
Abstract · 10 pages, 4 figures

Large-scale industrial ranking models must jointly capture heterogeneous feature interactions, long-term user behaviors, and multi-scenario dependencies under stringent online serving constraints. Existing architectures typically process sequential and non-sequential features with separate modules, limiting bidirectional information exchange. Directly applying homogeneous self-attention to all semantic tokens can also be inefficient when sequential and non-sequential signals exhibit different effective-ranks (eRanks). Meanwhile, shared representations can overlook scenario-specific patterns and induce negative transfer between scenarios. In this paper, we propose HiRank, a unified framework for multi-scenario recommendation that jointly models Heterogeneous feature Interactions via effective-Rank scaling. HiRank integrates gated sequential attention with heterogeneous feature interaction within a unified backbone, enabling bidirectional fusion between sequential and non-sequential representations. HiRank further uses scenario-aware aggregation to compress heterogeneous low-order tokens into a compact representation. Distinct learned projections transform the full compact representation into scenario-aware composite tokens with complementary global feature views and more uniform information density, enabling more effective high-order self-attention. Extensive experiments on industrial-scale datasets demonstrate favorable performance scaling with parameter count and consistent improvements over competitive baselines. Online A/B tests further show a 2.27% increase in orders per user.

0.24