LLMCDSR: Enhancing Cross-Domain Sequential Recommendation with Large Language Models (ACM TOIS 2025)

Forum topic · 小凯 · 2026-07-05

Summary

LLMCDSR is a research paper published in ACM Transactions on Information Systems (2025) that explores enhancing cross-domain sequential recommendation using large language models. The work addresses the challenge of transferring user preferences and behavioral patterns across different domains (e.g., from books to movies), a long-standing difficulty caused by sparse overlapping signals and divergent item taxonomies. The paper situates itself at the intersection of sequential recommendation systems and LLM-based semantic understanding, leveraging LLMs' world knowledge and natural-language reasoning to bridge cross-domain item representations. This forum post provides an overview of the paper's background, methodology (representation, core modules, learning strategies, inference pipeline), evaluation protocols, and open questions such as evaluation reliability, latency/cost, hallucination, and multilingual extension. Readers are directed to the original paper at https://dl.acm.org/doi/abs/10.1145/3715099 for quantitative results, and to related work including BERT4Rec, EAGER, Mamba4Rec, and item-ID indexing for recommendation foundation models.

LLMCDSR: Enhancing Cross-Domain Sequential Recommendation with Large Language Models

Source: ACM Digital Library · Venue: ACM Transactions on Information Systems (TOIS), 2025 · Topic: Sequential Recommendation

Overview

LLMCDSR (Large Language Models Cross-Domain Sequential Recommendation) investigates how large language models can improve cross-domain sequential recommendation. In large-scale search and personalization systems, recommendation has long faced challenges around efficiency, scalability, and user intent understanding. Traditional pipelines separate retrieval, ranking, and generation, which struggles to meet the LLM-era demands of natural language interaction, multi-hop reasoning, and up-to-date knowledge. LLMCDSR addresses this by using LLMs' semantic priors and world knowledge to bridge user behavior across domains.

Core Contributions

A unified perspective for the cross-domain recommendation problem, organizing scattered related work into a comparable framework.
A clear decomposition of method components (representation learning, retriever, reranker, planner, generator, feedback mechanisms) for engineering practice.
Reproducible benchmarks, datasets, or classification schemes that lower the entry barrier for follow-up research.
Discussion of interfaces with emerging paradigms such as LLM tool calling, reinforcement learning, and multi-agent collaboration.
Explicitly listed open problems: evaluation trustworthiness, latency and cost, hallucination and safety, and cross-lingual/multimodal extension.

Method Sketch

The work follows a four-stage pipeline: problem formulation → model/system design → training/construction → inference pipeline.

1. Input & representation: encode queries, documents, and user context into dense/sparse representations or structured prompts. 2. Core modules: retriever, reranker, planner, memory, and tool interfaces, composed serially or in parallel. 3. Learning strategies: supervised fine-tuning, contrastive learning, distillation, reinforcement learning (incl. process rewards), synthetic data bootstrapping. 4. Inference: single-pass retrieval, iterative retrieval, parallel sub-queries, early stopping, and budget control.

Evaluation

Typical evaluation protocol for this line of work:

Datasets: MS MARCO, BEIR, Natural Questions, domain corpora, public recommendation benchmarks.
Metrics: nDCG@10, MRR, Recall@k, Hit@k, human preference, task success rate, latency, token cost.
Baselines: BM25, dense retrieval, cross-encoder reranking, retrieval-free LLMs, commercial search APIs.

Quantitative results should be verified against the original PDF; this summary is based on the abstract and public metadata.

Insights for Search / Rec / Personalization

1. Architecture: cascaded retrieve + rerank + generate remains mainstream, but agentic paradigms make retrieval count and policy itself learnable. 2. Data: high-quality instruction data and click/session logs are both critical; synthetic data must avoid knowledge leakage and distribution shift. 3. Evaluation: the gap between offline metrics and online satisfaction is widening; LLM-as-judge needs cross-validation with human evaluation. 4. Production: latency, cost, explainability, and safety are hard constraints for industrial deployment — not just academic benchmarks.

Related Work

BERT4Rec — Sequential Recommendation with Bidirectional Encoder Representations
EAGER — Two-Stream Generative Recommender
Efficient On-Device Session-Based Recommendation (TOIS)
How to Index Item IDs for Recommendation Foundation Models (P5, SIGIR)
Mamba4Rec — Efficient Sequential Recommendation with Selective State Space Models
Multi-Behavior Sequential Transformer Recommender (SIGIR)

Limitations & Future Directions

Potential limitations include experiment scale constrained by GPU budgets, benchmarks mismatched with real user distributions, English-centric data limiting cross-lingual generalization, and safety risks of agentic systems on the open web. Future directions include more efficient test-time compute allocation, deeper integration with knowledge graphs and structured databases, and causal/fairness constraints for recommendation.

Practical Checklist for Engineers

| Item | Question | Suggestion | |------|----------|------------| | Data | PII in training/index? Version control? | Partitioned indexes, anonymization, rollback-safe embedding versions | | Latency | p99 budget? Retrieval steps? | Cascade + early stop, cache hot queries, async rerank | | Quality | Do offline gains translate to online CTR/satisfaction? | Interleaving experiments, human audits, citation checks | | Safety | Open retrieval introduces poisoning/bias? | Source whitelisting, adversarial detection, output filtering | | Cost | Token & GPU cost per query? | Small-model routing, distillation, sparse+dense hybrid |

Glossary

| Term | Meaning | |------|---------| | IR | Information Retrieval | | RAG | Retrieval-Augmented Generation | | LTR | Learning to Rank | | nDCG | Normalized Discounted Cumulative Gain | | Agentic Search | Search modeled as sequential decisions and tool calls | | Gen-IR | Generative Information Retrieval |