Bridging the gap between unstructured documents and relational databases is challenging because document extraction operates locally, whereas databases enforce global semantics through schemas, keys, and constraints. Consequently, existing one-shot large language model (LLM) extraction approaches often fail to reconcile results with relational semantics, yielding inconsistent and hard-to-audit outputs. We present DataMosaic, a document-to-database (Doc2DB) system that explicitly mediates between extraction and database semantics. Given a database schema and constraints, a central orchestrator coordinates entity and relationship extraction alongside verification, repair, and targeted re-extraction within a closed extract-verify-iterate loop. By systematically resolving document ambiguity and constraint violations, DataMosaic incrementally constructs accurate and semantically consistent databases. Featuring pluggable extractors, verifiers and repair operators, experiments in diverse datasets show that DataMosaic substantially reduces constraint violations and improves database-level accuracy over strong Doc2DB baselines.

Large language models (LLMs) are widely applied to data analytics over documents, yet direct reasoning over long, noisy documents remains brittle and error-prone. Hence, we study document question answering (QA) that consolidates dispersed evidence into a structured output (e.g., a table, graph, or chunks) to support reliable, verifiable QA. We propose a two-pillar framework, LiteCoST, to achieve both high accuracy and low latency with small language models (SLMs). Pillar 1: Chain-of-Structured-Thought (CoST). We introduce a CoST template—a schema-aware instruction that guides a strong LLM to produce both a step-wise CoST trace and the corresponding structured output. The process induces a minimal structure, normalizes entities/units, aligns records, serializes the output, and verifies/refines it (optionally with an LLM-as-judge), yielding auditable supervision. Pillar 2: SLM fine-tuning. We then train compact models on the LLM-generated CoST traces/structured data in two phases—Supervised Fine-Tuning for structure/format/steps, followed by Group Relative Policy Optimization with dual rewards for answer/format quality and process consistency—transferring structure-first behavior to SLMs for low-latency deployment. This approach achieves LLM-comparable quality on finance and legal long-document QA (Loong), with 3B/7B SLMs while delivering 2–4×lower latency than GPT-4o and DeepSeek-R1 (671B).

Many real-world conversational settings for knowledge discovery, including podcasts, hiring screens, and marketplaces, require a purpose-driven understanding of a person. We study the Next-Best-Question (NBQ) problem: at each turn, an interviewer should ask the question with the highest expected information gain given what has already been learned and the conversation goal. We propose NBQ, a plug-and-play framework that seeds a diverse pool of candidate questions, maintains a compact and continuously updated user state, adaptively selects the next question within a turn budget, and distills the resulting free-form dialogue into a structured vector-based user profile. As a demanding application, we instantiate NBQ for reciprocal matchmaking, where compatibility must be mutual and each person is modeled by both self-description and counterpart-preference representations. To support large-scale matching, we further introduce QuickMatch, an efficient retrieval layer that recasts reciprocal matching from quadratic pairwise scoring to approximate vector search. Experiments show that NBQ improves user profiling quality by up to 13.6

The rapid growth of online grocery shopping requires recommendation systems that capture cyclical purchasing behavior and diverse user intents. Traditional item-level methods face scalability and accuracy challenges, motivating category-level recommendation as a more structured and practical alternative. We present GROCLM, a fine-tuned language model for grocery category recommendation in a real-world production environment. GROCLM employs a two-stage LoRA-based training strategy to encode cyclical purchasing patterns directly into model parameters, enabling more effective utilization of rebuying signals compared to prompt-based conditioning. To ensure valid and controllable outputs, we further introduce a trie-based constrained decoding mechanism over a predefined category space. Experiments on both proprietary production data and a public benchmark demonstrate that GROCLM consistently outperforms strong baselines. In a live production restocking task, GROCLM achieves a 7.5

With the prevalence of deep learning based embedding approaches, recommender systems have become a proven and indispensable tool in various information filtering applications. However, many of them remain difficult to diagnose what aspects of the deep models’ input drive the final ranking decision, thus, they cannot often be understood by human stakeholders. In this paper, we investigate the dilemma between recommendation and explainability, and show that by utilizing the contextual features (e.g., item reviews from users), we can design a series of explainable recommender systems without sacrificing their performance. In particular, we propose three types of explainable recommendation strategies with gradual changes in model transparency: whitebox, graybox, and blackbox. Each strategy explains its ranking decisions via different mechanisms: attention weights, adversarial perturbations, and counterfactual perturbations. We apply these explainable models on five real-world data sets under the contextualized setting where the users provided explicit reviews of the items. The empirical results show that our models can achieve highly competitive ranking performance, and meanwhile, generate accurate and effective explanations in terms of numerous quantitative metrics and qualitative visualizations.

Federated Learning (FL) is a promising distributed machine learning approach that enables collaborative training of a global model using multiple edge devices. The data distributed among the edge devices is highly heterogeneous. Thus, FL faces the challenge of data distribution and heterogeneity, where non-Independent and Identically Distributed (non-IID) data across edge devices may yield in significant accuracy drop. Furthermore, the limited computation and communication capabilities of edge devices increase the likelihood of stragglers, thus leading to slow model convergence. In this paper, we propose the FedDHAD FL framework, which comes with two novel methods: Dynamic Heterogeneous model aggregation (FedDH) and Adaptive Dropout (FedAD). FedDH dynamically adjusts the weights of each local model within the model aggregation process based on the non-IID degree of heterogeneous data to deal with the statistical data heterogeneity. FedAD performs neuron-adaptive operations in response to heterogeneous devices to improve accuracy while achieving superb efficiency. The combination of these two methods makes FedDHAD significantly outperform state-of-the-art solutions in terms of accuracy (up to 6.7% higher), efficiency (up to 2.02 times faster), and computation cost (up to 15.0% smaller).

Efficient and accurate product relevance assessment is critical for user experiences and business success. Training a proficient relevance assessment model requires high-quality query-product pairs, often obtained through negative sampling strategies. Unfortunately, current methods introduce pooling bias by mistakenly sampling false negatives, diminishing performance and business impact. To address this, we present Bias-mitigating Hard Negative Sampling (BHNS), a novel negative sampling strategy tailored to identify and adjust for false negatives, building upon our original False Negative Estimation algorithm. Our experiments in the Instacart search setting confirm BHNS as effective for practical e-commerce use. Furthermore, comparative analyses on public dataset showcase its domain-agnostic potential for diverse applications.

The impact of generative AI on e-commerce is profound. It has significantly improved the understanding of user intent and serves as a comprehensive product knowledge graph. However, the most substantial disruptions are yet to come, partic- ularly through the rise of autonomous agents. In this talk, I will outline a tentative path toward a future where e-commerce not only offers an unparalleled customer experience but also thrives in a world dominated by generative AI and autonomous agents.

Personalization is essential in e-commerce, with item recommendation as a critical task. In this paper, we describe a hybrid embedding-based retrieval system for real-time personalized item recommendations on Instacart. Our system addresses unique challenges in the multi-source retrieval system, and includes several key components to make it highly personalized and dynamic. Specifically, our system features a hybrid embedding model that includes a long-term user interests embedding model and a real-time session-based model, which are combined to capture users’ immediate intents and historical interactions. Additionally, we have developed a contextual bandit solution to dynamically adjust the number of candidates from each source and optimally allocate retrieval slots given a limited computational budget. Our modeling and system optimization efforts have enabled us to provide highly personalized item recommendations in real-time at scale to all our customers, including new and long-standing users.

Transferable graph learning involves knowledge transferability from a source graph to a relevant target graph. The major challenge of transferable graph learning is the distribution shift between source and target graphs induced by individual node attributes and complex graph structures. To solve this problem, in this paper, we propose a generic graph-structured Gaussian process framework (GraphGP) for adaptively transferring knowledge across graphs with either homophily or heterophily assumptions. Specifically, GraphGP is derived from a novel graph structure-aware neural network in the limit on the layer width. The generalization analysis of GraphGP explicitly investigates the connection between knowledge transferability and graph domain similarity. Extensive experiments on several transferable graph learning benchmarks demonstrate the efficacy of GraphGP over state-of-the-art Gaussian process baselines.

Next Basket Recommendation (NBR) aims to recommend a set of products as a basket to users based on their historical shopping behavior. In this paper, we investigate the problem of NBR in online marketplaces (e.g., Instacart, Uber Eats) that connect users with multiple sellers. In such scenarios, effective NBR can significantly enhance the shopping experience of users by recommending diversified and completed products based on specific sellers, especially when a user purchases from a seller they have not visited before. However, conventional NBR approaches assume that all considered products are from the same sellers, which overlooks the complex relationships between users, sellers, and products. To address such limitations, we develop SecGT, a sequential collaborative graph transformer framework that recommends users with baskets from specific sellers based on seller-aware user preference representations that are generated by collaboratively modeling the joint user-seller-product interactions and sequentially exploring the user-agnostic basket transitions in an interactive way. We evaluate the performance of SecGT on users from a leading online marketplace at multiple cities with various involved sellers. The results show that SecGT outperforms existing NBR and also traditional product recommendation approaches on recommending baskets from cold sellers for different types of users across all cities.

Large language models (LLMs) have recently taken the world by storm, promising potentially game changing opportunities in multiple fields. Naturally, there is significant promise in applying LLMs to the management of structured data, or more generally, to the processes involved in data science. At the very least, LLMs have the potential to provide substantial advancements in long-standing challenges that our community has been tackling for decades. On the other hand, they may introduce completely new capabilities that we have only dreamed of thus far. This panel will bring together a few leading experts who have been thinking about these opportunities from various perspectives and fielding them in research prototypes and even in commercial applications.

A natural language interface (NLI) to databases is an interface that translates a natural language question to a structured query that is executable by database management systems (DBMS). However, an NLI that is trained in the general domain is hard to apply in the spatial domain due to the idiosyncrasy and expressiveness of the spatial semantics. Moreover, there are a wide range of database servers available, and a unilingual NLI model limits its practical usage. In this article, we propose to not only address the spatial domain generalization challenge, but also support multilingual back-end, i.e., supporting different query languages, such as SQL and Prolog. For the challenge of spatial semantics, we propose a spatial comprehension model that is able to recognize the meaning of spatial entities based on the semantics of context and effectively resolve the ambiguity given the spatial semantics. The spatial semantics learned from the spatial comprehension model is then injected to the natural language question to ease the burden of capturing the spatial-specific semantics. We also propose to add a prefix symbol to support the multilingual back-end (e.g., query languages). With our spatial comprehension model and symbol injections, our NLI for the spatial domain, named SpatialNLI, is able to capture the semantic structure of the question and translate it to the corresponding syntax of an executable query accurately. We also experimentally ascertain that SpatialNLI outperforms state-of-the-art methods.

Communities have been shown to coordinate among themselves using public forums such as Twitter and Facebook during disastrous events, and utilize even sparse telecommunication infrastructure to communicate. In this work we design a framework for text mining that will extract crisis situation and location information from tweets about individuals in need. Using these extracted data, we develop a community source map of the disaster. There have been previous efforts to develop real-time disaster maps, but these are limited to manual entry whereas we use AI to create such maps. While much of the prior work focuses on densely populated urban scenarios, we also consider sparsely populated rural contexts where social network updates may be infrequent. We have experimented our proposed solution on around 20 different disaster related tweet databases from different part of the world. The primary results show ~81 percent accuracy in crisis situation information extraction and ~92 percent accuracy in location extraction.