--- title: 'TaoSR1: Taobao Search Relevance Model v1' url: https://www.emergentmind.com/topics/taosr1 type: topic --- # TaoSR1: Taobao Search Relevance Model v1 TaoSR1, short for **Taobao Search Relevance Model v1**, is an LLM-based “thinking model” for e-commerce relevance search that is designed to **directly deploy a generative large language model online** for query-product relevance classification [2508.12365]. It is presented as a deployment-ready paradigm for relevance classification in which the LLM remains the final online model, rather than being used only for discriminative fine-tuning or distilled back into a smaller encoder. The framework targets the practical difficulties of applying chain-of-thought (CoT) reasoning to search relevance, specifically **deployment latency**, **CoT error accumulation**, and **discriminative hallucination**, and organizes its solution around supervised fine-tuning with CoT, offline pass@N sampling with Direct Preference Optimization (DPO), difficulty-based dynamic sampling with Group Relative Policy Optimization (GRPO), and online deployment via post-CoT and cumulative probability-based tiering [2508.12365]. ## 1. Task definition and relevance taxonomy TaoSR1 addresses **query-item relevance prediction** in an e-commerce search engine. Given a query and a candidate product, the model predicts one of four relevance labels: **4-Excellent**, **3-Related**, **2-Mismatch**, and **1-Irrelevant** [2508.12365]. In the Taobao setup described in the paper, this is not limited to lexical or shallow semantic matching; it must also handle **negation queries**, **“alternative to X” queries**, **question-answering queries**, **knowledge-based queries**, and **long-tail semantic cases requiring business-rule reasoning** [2508.12365]. The four-class labeling scheme is used internally, while online serving requires a coarser three-tier partition. The production tiers are **Good**, **Mid**, and **Bad**, corresponding respectively to \(\{4\text{-Excellent}, 3\text{-Related}\}\), \(\{2\text{-Mismatch}\}\), and \(\{1\text{-Irrelevant}\}\) [2508.12365]. This distinction matters because the model is simultaneously a fine-grained classifier and a production ranking component. | Four-tier label | Production tier | |---|---| | 4-Excellent | Good | | 3-Related | Good | | 2-Mismatch | Mid | | 1-Irrelevant | Bad | The paper emphasizes that relevance judgments may depend on asymmetric and context-sensitive business rules. Its example is that “Mate50 Pro” retrieved for query “Mate50” may be **3-Related**, whereas the reverse can be **2-Mismatch** [2508.12365]. This indicates that TaoSR1 is intended for a relevance regime in which taxonomy, attributes, product granularity, and domain policy interact nontrivially. ## 2. Motivation and limitations of prior paradigms The motivation for TaoSR1 begins from a limitation attributed to **BERT-based relevance models**. These models are described as strong for ordinary lexical and semantic matching but insufficient for difficult relevance cases because they mainly model text interaction rather than richer reasoning, because long-tail queries require explicit logic and business rules, and because the BERT-style paradigm is limited in scaling and representational ceiling [2508.12365]. The paper also identifies limits in prior **LLM-based relevance methods**. Although recent LLM approaches improve semantic understanding, most still either use **discriminative optimization** rather than true generative reasoning or **distill back into BERT-like models** for deployment [2508.12365]. In that view, the LLM’s reasoning ability is not fully exploited online. A further claim is that simply attaching CoT to relevance classification is insufficient. Three practical obstacles are singled out. First, **deployment latency** becomes prohibitive when full CoT must be generated for hundreds of candidate items per query. Second, **CoT error accumulation** allows hallucinations or deviations in intermediate reasoning steps to propagate into the final label. Third, **discriminative hallucination** can occur when the reasoning appears plausible but the final relevance label is still wrong [2508.12365]. The framework is therefore framed not merely as a larger model, but as a response to an optimization and systems mismatch between generative reasoning and online relevance serving. ## 3. Generative formulation, CoT supervision, and post-CoT At the supervised fine-tuning stage, TaoSR1 does not train a discriminative classifier. Instead, it uses a **generative objective** in which the model directly generates the label text, such as `"4-Excellent"` [2508.12365]. The paper gives the baseline language-model objective as \[ \mathcal{L}_{lm}(\pi)=\mathbb{E}_{(\mathbf{x},\mathbf{y}) \sim \mathcal{D}\left[-\sum_{t=1}^{|\mathbf{y}|-1} \log \pi\left(\mathbf{y}_{t+1} \mid \mathbf{x}_1, \ldots, y_t\right)\right]. \] Here, \(\mathbf{x}\) denotes the input query-item pair and features, \(\mathbf{y}\) the output sequence, \(\pi\) the model distribution, \(\mathcal{D}\) the training data, and \(t\) the token index [2508.12365]. The formulation is standard next-token prediction over the label text, but its use for relevance classification is central to the framework. For deployment, TaoSR1 extracts a **continuous score** from the probability of the **first generated token** rather than relying only on the final generated label string. The paper defines \[ \mathbf{P}\left(\mathbf{y}_1=c \mid \mathbf{x}\right)=\left(\mathbf{p}_1\right)_{\text{id(c)}=\frac{\exp \left(\pi(\mathbf{x}\right)_{\text{id(c)})}{\sum_{j=1}^{4} \exp \left(\pi(\mathbf{x}\right)_{\text{id}(c_j)})}. \] Its interpretation is that \(c \in \{1,2,3,4\}\) is the class token, \(\mathbf{p}_1\) is the first-token probability vector, and the probabilities are normalized over the four label tokens [2508.12365]. This first-token distribution later supports online tiering. To teach domain-specific reasoning, the training pipeline synthesizes CoT supervision using **Retrieval-Augmented Generation (RAG)**. The structured reasoning template has five steps: **Query Understanding**, **Product Comprehension**, **Category Matching**, **Attribute Matching**, and **Relevance Class Determination** [2508.12365]. Human annotation supplies **atomic factors** or fine-grained reasons; these atomic reasons retrieve matching business rules from a structured knowledge base; the retrieved rules are inserted into the prompt; and **DeepSeek-R1** is used to synthesize CoT examples. Each resulting sample has the form \[ \langle \text{query}, \text{item}, \text{CoT}, \text{label} \rangle. \] A notable architectural result is the comparison between **think-then-respond** and **respond-then-think**. The former outputs \(\langle \text{CoT}, \text{label} \rangle\); the latter outputs \(\langle \text{label}, \text{CoT} \rangle\) [2508.12365]. The paper reports that, counterintuitively, **think-then-respond performs worse** because of **error accumulation** in the reasoning chain. The preferred design is therefore **respond-then-think**, called **post-CoT** or **postCoT**, in which the label is predicted first and the reasoning comes after [2508.12365]. In context, this means the model uses reasoning as a support mechanism without forcing the classification decision to wait for a long and error-prone latent chain. ## 4. Preference optimization and reinforcement learning stages The second stage of TaoSR1 exploits the gap between single-sample accuracy and **pass@N** accuracy. The paper reports that the model’s pass@N from multiple offline samples is much higher than its single-pass performance: **LLM base: 75.01**, **TaoSR1(CoT) Pass@1: 67.38**, **Pass@2: 74.26**, **Pass@3: 77.68**, **Pass@4: 80.18**, and **Pass@5: 81.73** [2508.12365]. This suggests that correct predictions are often present among sampled candidates even when the top sample is wrong. On that basis, TaoSR1 constructs preference data for **DPO** from two disjoint subsets. For **solvable cases** where pass@N \(> 0\), one correct sample is chosen as \(y^+\) and one incorrect sample as \(y^-\), creating self-correction pairs. For **hard cases** where pass@N \(= 0\), **DeepSeek-R1** generates a correct output, which is paired with the model’s incorrect response; approximately **50%** of pass@N \(= 0\) instances are curated into this supplemental preference dataset [2508.12365]. The paper presents the DPO objective in terms of a trainable policy \(\pi_\theta\), a reference model \(\pi_{\mathrm{ref}}\), chosen and rejected responses \(\mathbf{y}^+\) and \(\mathbf{y}^-\), temperature parameter \(\beta\), and sigmoid \(\sigma\), with the effect of pushing the model to assign higher relative probability to chosen responses than rejected ones [2508.12365]. The stated role of DPO is to improve correct-answer preference, robustness on hard cases, and the ability to recover from CoT process errors. The third stage applies **GRPO** to address **discriminative hallucination** and further improve reasoning quality. The paper’s motivation is that even a seemingly good reasoning path may end in the wrong final label, and that naive group-based policy optimization wastes effort on groups that are **all correct** or **all incorrect** [2508.12365]. TaoSR1 therefore uses **difficulty-based dynamic sampling**: it only backpropagates for groups whose empirical accuracy lies in \[ (0,\gamma). \] This excludes groups where everything is right and groups where everything is wrong, focusing optimization on partially solvable, difficult instances [2508.12365]. The paper also reports that a **balanced label distribution** improves GRPO performance, and accordingly majority classes are downsampled to balance the training set. The GRPO objective is given as \[ \mathcal{L}_{\mathrm{GRPO}(\theta)} =\mathbb{E}_{(x,y) \sim \mathcal{D},\left\{o_i\right\}_{i=1}^G \sim \mathcal{D}} \left[\frac{1}{G} \sum_{i=1}^G \frac{1}{\left|o_i\right|} \sum_{t=1}^{o_i} \min \left(r_{i, t}(\theta) \hat{A}_t, \operatorname{clip}\left(r_{i, t}(\theta), 1-\epsilon, 1+\epsilon\right) \hat{A}_t\right) -\beta D_{\mathrm{KL}\left(\pi_\theta \| \pi_{r e f}\right)}\right] \] subject to \[ 0 <\mid\left\{o_i \mid \operatorname { is\_equivalent }\left(y, o_i\right)\right\} \mid<\gamma, \] with \[ r_{i, t}(\theta)=\frac{\pi_\theta\left(o_{i, t} \mid x, o_{i,