E-commerce search relevance modeling faces a critical dilemma: traditional models falter with complex queries, while Large Language Models (LLMs), despite their superior reasoning, suffer from the prohibitive latency of auto-regressive Chain-of-Thought (CoT) generation, rendering them infeasible for production. Knowledge distillation offers a promising solution, yet current methods force an undesirable trade-off: sacrificing the very interpretability that makes LLMs powerful, or relying on expensive, unscalable human-annotated rationales. To address this, we propose STAR—Staged Training with Aligned Reinforcement Learning and Multi-Faceted Distillation, a progressive framework that follows a reasoning, ranking, and transfer pipeline to imbue dense models with both high performance and interpretability. First, STAR aligns a teacher LLM's reasoning with task objectives using a novel multi-granularity reward in Group Relative Policy Optimization (GRPO), leveraging only binary labels. Next, it refines the teacher's ability for calibrated scoring via token-level supervision, enabling efficient ranking through a single forward pass without any additional layers. Finally, this ''white-box'' knowledge is transferred to a compact student via multi-faceted distillation that preserves both reasoning logic and ranking behavior. Offline experiments demonstrate that our 0.6B student model rivals the performance of a strong 8B baseline, making it highly efficient and fully deployable. Real-world effectiveness is validated by significant online A/B test gains, including a +0.93% GoodRate lift and a +1.04% increase in GMV. STAR has been fully deployed to 100% of main search traffic on 1688.com.

In large-scale e-commerce search, accurately modeling multi-modal relevance is paramount for matching user intent—especially given the growing influence of visual content on shopping decisions. However, existing methods often fail to perform fine-grained reasoning. For instance, they struggle when a product title is irrelevant due to marketing language, but its image is highly relevant to the query. Furthermore, they cannot effectively disambiguate which specific query terms are satisfied by the visual versus the textual modality. While Large Language Models (LLMs) excel at such reasoning, their high computational overhead makes direct online deployment infeasible. To bridge this gap, we propose TRACE (Term-level Reasoning And Chain-of-thought Enhanced distillation), a framework designed for deploying advanced reasoning capabilities at scale. TRACE operates in two stages. First, it enhances an LLM's multi-modal reasoning by employing Group Relative Policy Optimization (GRPO) guided by a term-level Chain-of-Thought (CoT) reward function, enabling it to generate detailed, step-by-step relevance judgments. Second, it efficiently transfers this fine-grained reasoning to a lightweight, deployable model using a novel term-level knowledge distillation strategy that inherits reasoning ability. Offline evaluations show significant improvements across different datasets. More critically, online A/B tests on 1688.com resulted in a +1.04% GMV uplift, a +0.906% LTV increase, and a +0.523% improvement in UV_L2O, demonstrating its significant value in a real-world production environment.

In e-commerce search, the diverse ways in which users express intentions lead to lexical and semantic gaps between queries and product descriptions, making query rewriting (QR) indispensable for improving matching efficiency. With the development of LLMs, QR has evolved from discriminative approaches to various LLM-based alignment methods. However, these methods typically treat all queries uniformly, without fundamentally distinguishing their rewriting difficulty or underlying linguistic issues, making the rewritten query deviate from human expectations. To address this limitation, we propose AWHCP (Aligning with Human Cognition and Preference), a novel framework that adopts a human-centric perspective and introduces the Problem–Intention–Fix–Rewrite (PIFR) paradigm. Built upon PIFR, AWHCP establishes a multi-granularity alignment training framework that simultaneously aligns with both system retrieval preferences and human rewriting behaviors. First, we construct high-quality PIFR-structured data and perform supervised fine-tuning to enable the model to learn human-like rewriting patterns. Second, we apply beam search to generate multiple candidates and leverage system-side feedback signals to conduct coarse-grained direct preference alignment, endowing the model with initial difficulty-aware reasoning capabilities. Third, we introduce a multi-dimensional rewrite quality judgment model trained via Group Relative Policy Optimization (GRPO), enabling fine-grained alignment with nuanced human rewriting preferences. Deployed on 1688's main search engine since August 2025, AWHCP has demonstrated strong effectiveness through extensive offline evaluations and large-scale online A/B tests, leading to a +3.9% gain in UV-L2O.

Accurate prediction of click-through rates (CTR) for cold-start entities (CSEs) within search engine ecosystems presents significant challenges. Notably, CSEs encompass novel users/items and new session search queries, each characterized by their limited interaction data and poor-quality embeddings, which collectively contribute to the complexity of CTR estimation.Existing studies predominantly address cold-start challenges in isolation, such as focusing separately on new users or new items, and lack a comprehensive framework to effectively integrate atomic ID features with group-level representations. To address these limitations, we propose GRAIN (Group Reinforced Adaptive Interaction Network), a novel framework that enhances CTR prediction across all maturity phases, namely Cold-Start, Warm-Up, and Common. GRAIN consists of three key components: 1) a Graph-based Id-to-Cluster (GIC) module that aggregates atomic ID features into cluster-level representations; 2) an ID-Cluster Cross (ICC) module that aligns ID-level and cluster-level features through contrastive learning and cross-grained interaction mechanism; 3) a lightweight auxiliary task that classifies entities into different maturity stages using a data-driven phase partitioning algorithm. Extensive experiments demonstrate GRAIN's effectiveness in improving CTR prediction accuracy across multiple maturity phases. GRAIN has been successfully deployed on the 1688 App, handling billions of daily requests.

In e-commerce, search and recommendation rankings require a deep understanding of user behaviors and personalized scoring of products. While existing systems maintain separate pipelines for search and recommendation, these two scenarios share aligned objectives and exhibit consistent data patterns during ranking. To address this, we propose a joint modeling approach for search-recommendation ranking that enables information gain exchange between the two scenarios, thus facilitating enhanced modeling of users' cross-scenario behaviors. Our proposed SuperRS framework employs a Dual-layer Multi-MoE (DualMoE) architecture to tackle scenario-specific disparities and achieve multi-interest fusion perception. A key aspect is the Search-Recommendation Sequence Fusion Unit, which integrates user interaction sequences from both scenarios. Additionally, we introduce a unified Representation Extraction method utilizing Reciprocal Scenario Interest Attention (RSIA) for feature alignment. Dynamic Feature Integration (DFI) employs a dual gating mechanism for controlled information fusion while preserving scenario identities, combined with multi-objective optimization. On the 1688 App, our framework demonstrates superior performance to baseline models across both offline evaluation metrics and online business indicators.

Nowadays, online shoppers have paid more and more attention to detailed product descriptions, since a well-written description is a huge factor in making online sales. However, for a website with billions of product data like Alibaba, the writing efficiency of human copywriters cannot match the growth rate of new products. To address this issue, we propose a novel pointer-generator neural network to generate product description. In particular, coordinate encoders and a pattern-controlled decoder are utilized to improve generation quality with an attention mechanism. The coordinate encoders equipped with a Transformer and a gated convolutional unit is introduced to learn the source input representations. In the decoding phase, a pattern controlled decoder is proposed to control the output description pattern (such as category, length, and style) to ensure the quality of the description. For evaluation, we build a substantial collection of real-world products along with human-written descriptions. An extensive set of experiments with both human annotated data demonstrate the advantage of the proposed method for generation qualities. Finally, an online deployment shows significant benefits of our model in a real online shopping scenario, as measured by the click-through rate.