arXiv subjects

Changsung Kang

Publications and source records attributed to Changsung Kang.

9 recordsLinked to original sources

arXiv2026

It Takes Two to Match: Co-Evolving Generative Retriever with Reinforcement Learning

Retrieval is the first stage of modern search and advertising systems, selecting a candidate set from a large item universe for downstream ranking and auction. Recent work increasingly leverages LLMs to improve retrieval through query expansion, data synthesis, and retrieval-feedback training. However, the generative component is typically used for query-side augmentation, while final matching is still delegated to a downstream retriever. We introduce CoGR, a retrieval framework that instead trains LLMs to directly construct retrieval representations on both query and item sides. Each generator produces a compact set of keywords, which are matched directly through an inverted index, preserving compatibility with existing keyword-based retrieval infrastructure. CoGR uses a two-stage training pipeline. Supervised fine-tuning first establishes an aligned keyword space, after which co-evolving reinforcement learning alternately optimizes the query- and item-side generators with GRPO against the opposite side's frozen index. Both sides optimize the same query-to-item retrieval $F_1$ objective: the query side receives retrieval $F_1$ directly, while the item side receives a counterfactual marginal reward measuring the change in query-side $F_1$ caused by its generated keywords. Across 10 representative sparse, dense, and generative baselines, CoGR achieves the best performance on both an internal APP Marketplace dataset and the public WANDS benchmark, improving $F_1$ over the strongest baseline by $10.9\%$ and $36.1\%$, respectively. Further analysis shows stable co-evolution and increasingly aligned query--item keyword spaces over training.

cs.IR↗

arXiv2025

Generating Query-Relevant Document Summaries via Reinforcement Learning

E-commerce search engines often rely solely on product titles as input for ranking models with latency constraints. However, this approach can result in suboptimal relevance predictions, as product titles often lack sufficient detail to capture query intent. While product descriptions provide richer information, their verbosity and length make them unsuitable for real-time ranking, particularly for computationally expensive architectures like cross-encoder ranking models. To address this challenge, we propose ReLSum, a novel reinforcement learning framework designed to generate concise, query-relevant summaries of product descriptions optimized for search relevance. ReLSum leverages relevance scores as rewards to align the objectives of summarization and ranking, effectively overcoming limitations of prior methods, such as misaligned learning targets. The framework employs a trainable large language model (LLM) to produce summaries, which are then used as input for a cross-encoder ranking model. Experimental results demonstrate significant improvements in offline metrics, including recall and NDCG, as well as online user engagement metrics. ReLSum provides a scalable and efficient solution for enhancing search relevance in large-scale e-commerce systems.

cs.IR↗

arXiv2025

Knowledge Distillation for Enhancing Walmart E-commerce Search Relevance Using Large Language Models

Ensuring the products displayed in e-commerce search results are relevant to users queries is crucial for improving the user experience. With their advanced semantic understanding, deep learning models have been widely used for relevance matching in search tasks. While large language models (LLMs) offer superior ranking capabilities, it is challenging to deploy LLMs in real-time systems due to the high-latency requirements. To leverage the ranking power of LLMs while meeting the low-latency demands of production systems, we propose a novel framework that distills a high performing LLM into a more efficient, low-latency student model. To help the student model learn more effectively from the teacher model, we first train the teacher LLM as a classification model with soft targets. Then, we train the student model to capture the relevance margin between pairs of products for a given query using mean squared error loss. Instead of using the same training data as the teacher model, we significantly expand the student model dataset by generating unlabeled data and labeling it with the teacher model predictions. Experimental results show that the student model performance continues to improve as the size of the augmented training data increases. In fact, with enough augmented data, the student model can outperform the teacher model. The student model has been successfully deployed in production at Walmart.com with significantly positive metrics.

cs.IR↗

arXiv2024

Meta Learning to Rank for Sparsely Supervised Queries

Supervisory signals are a critical resource for training learning to rank models. In many real-world search and retrieval scenarios, these signals may not be readily available or could be costly to obtain for some queries. The examples include domains where labeling requires professional expertise, applications with strong privacy constraints, and user engagement information that are too scarce. We refer to these scenarios as sparsely supervised queries which pose significant challenges to traditional learning to rank models. In this work, we address sparsely supervised queries by proposing a novel meta learning to rank framework which leverages fast learning and adaption capability of meta-learning. The proposed approach accounts for the fact that different queries have different optimal parameters for their rankers, in contrast to traditional learning to rank models which only learn a global ranking model applied to all the queries. In consequence, the proposed method would yield significant advantages especially when new queries are of different characteristics with the training queries. Moreover, the proposed meta learning to rank framework is generic and flexible. We conduct a set of comprehensive experiments on both public datasets and a real-world e-commerce dataset. The results demonstrate that the proposed meta-learning approach can significantly enhance the performance of learning to rank models with sparsely labeled queries.

cs.IR↗

arXiv2021

Classifying Emails into Human vs Machine Category

It is an essential product requirement of Yahoo Mail to distinguish between personal and machine-generated emails. The old production classifier in Yahoo Mail was based on a simple logistic regression model. That model was trained by aggregating features at the SMTP address level. We propose building deep learning models at the message level. We built and trained four individual CNN models: (1) a content model with subject and content as input; (2) a sender model with sender email address and name as input; (3) an action model by analyzing email recipients' action patterns and correspondingly generating target labels based on senders' opening/deleting behaviors; (4) a salutation model by utilizing senders' "explicit salutation" signal as positive labels. Next, we built a final full model after exploring different combinations of the above four models. Experimental results on editorial data show that our full model improves the adjusted-recall from 70.5% to 78.8% compared to the old production model, while at the same time lifts the precision from 94.7% to 96.0%. Our full model also significantly beats the state-of-the-art Bert model at this task. This full model has been deployed into the current production system (Yahoo Mail 6).

cs.CL↗

arXiv2012

Local Markov Property for Models Satisfying Composition Axiom

The local Markov condition for a DAG to be an independence map of a probability distribution is well known. For DAGs with latent variables, represented as bi-directed edges in the graph, the local Markov property may invoke exponential number of conditional independencies. This paper shows that the number of conditional independence relations required may be reduced if the probability distributions satisfy the composition axiom. In certain types of graphs, only linear number of conditional independencies are required. The result has applications in testing linear structural equation models with correlated errors.

cs.AI↗

arXiv2012

Inequality Constraints in Causal Models with Hidden Variables

We present a class of inequality constraints on the set of distributions induced by local interventions on variables governed by a causal Bayesian network, in which some of the variables remain unmeasured. We derive bounds on causal effects that are not directly measured in randomized experiments. We derive instrumental inequality type of constraints on nonexperimental distributions. The results have applications in testing causal models with observational or experimental data.

cs.AI↗

arXiv2012

Polynomial Constraints in Causal Bayesian Networks

We use the implicitization procedure to generate polynomial equality constraints on the set of distributions induced by local interventions on variables governed by a causal Bayesian network with hidden variables. We show how we may reduce the complexity of the implicitization problem and make the problem tractable in certain causal Bayesian networks. We also show some preliminary results on the algebraic structure of polynomial constraints. The results have applications in distinguishing between causal models and in testing causal models with combined observational and experimental data.

cs.AI↗

arXiv2011

Modeling Perceived Relevance for Tail Queries without Click-Through Data

Click-through data has been used in various ways in Web search such as estimating relevance between documents and queries. Since only search snippets are perceived by users before issuing any clicks, the relevance induced by clicks are usually called \emph{perceived relevance} which has proven to be quite useful for Web search. While there is plenty of click data for popular queries, very little information is available for unpopular tail ones. These tail queries take a large portion of the search volume but search accuracy for these queries is usually unsatisfactory due to data sparseness such as limited click information. In this paper, we study the problem of modeling perceived relevance for queries without click-through data. Instead of relying on users' click data, we carefully design a set of snippet features and use them to approximately capture the perceived relevance. We study the effectiveness of this set of snippet features in two settings: (1) predicting perceived relevance and (2) enhancing search engine ranking. Experimental results show that our proposed model is effective to predict the relative perceived relevance of Web search results. Furthermore, our proposed snippet features are effective to improve search accuracy for longer tail queries without click-through data.

cs.IR↗