Large Language Models (LLMs) are increasingly deployed in interactive information access systems over structured data, enabling natural language querying of large-scale enterprise tables in domains such as digital advertising. However, ensuring factually accurate and hallucination-free responses remains a critical challenge for reliable Table Question Answering (Table QA), particularly in production analytics workflows. While Direct Preference Optimization (DPO) has emerged as a popular approach for aligning LLMs using pairwise feedback, it overlooks the underlying rationale behind preferences, limiting the effectiveness of alignment. In this work, we introduce Reason-Aware Preference Optimization (RAPO), a novel alignment framework that incorporates reason-augmented preference signals to better guide LLM learning and improve the reliability of structured information retrieval. Evaluated on real-world advertising datasets, RAPO demonstrates substantial improvements over DPO in factual accuracy, enabling more trustworthy retrieval of business-critical insights.

Position bias, the phenomenon whereby users tend to focus on higher-ranked items of the search result list regardless of the actual relevance to queries, is prevailing in many ranking systems. Position bias in training data biases the ranking model, leading to increasingly unfair item rankings, click-through-rate (CTR), and conversion rate (CVR) predictions. To jointly mitigate position bias in both item CTR and CVR prediction, we propose two position-bias-free CTR and CVR prediction models: Position-Aware Click-Conversion (PACC) and PACC via Position Embedding (PACC-PE). PACC is built upon probability decomposition and models position information as a probability. PACC-PE utilizes neural networks to model product-specific position information as embedding. Experiments on the E-commerce sponsored product search dataset show that our proposed models have better ranking effectiveness and can greatly alleviate position bias in both CTR and CVR prediction.

Time series data is ubiquitous, and accurate time series forecasting is vital for many real-world application domains, including retail, healthcare, supply chain, climate science, e-commerce and economics. Forecasting, in general, has led to broad impact and a diverse range of applications. However, with large-scale, high-dimensional time-series data available, more advanced techniques must be invented or improved for highly accurate predictions. Latest data mining and machine learning techniques play a crucial role in the next generation of forecasting models. In this Applied Machine Learning Methods for Time Series Forecasting (AMLTS) workshop, we focus on effective and accurate latest machine learning approaches to solve various real-world problems. With this workshop's ability to attract audiences across various domains, we invite experienced industrial practitioners and researchers to help uncover new approaches and break new ground in time-series modelings' challenging and vital settings.

Named-Entity-Recognition (NER) or Item Aspect Recognition task is fundamental to e-commerce marketplace. A structured listing (i.e. item aspect and catalog) is critical to the success of the marketplace - it helps the seller to put their listings to the right catalog and aspects so that the buyers can easily find the most relevant and accurate listings they are looking for. For e-commerce search engine, item aspects (brand, color, size, texture, etc.) should be automatically recognized from users' shopping queries in order to accurate understand their shopping intent. An accurate query and item aspect understanding helps to match seller's listing to buyer's purchase intend. However, in practice, this still remains challenge in e-commerce marketplace due to a couple reasons, e.g. the sparsity of the data - hundreds of millions of item aspect-name and aspect-value pairs (e.g. brand=Nike); noisy (low quality), not scalable but expensive label data that are obtained via human labeling \citeb12 effort; the lack of context in generally very short search queries; the disperse of context and dis-orderedness in listing title or descriptions, etc. among others. All those imposes a big challenge that affect the performance of aspect recognition for e-commerce practice. In this paper, we introduce an end-to-end machine learning system to build an effective aspect recognition system for search engine that leverages the generated label data from different legacy systems and effort, without any extra human-label effort (and thus no incurred cost). The framework is constructed with multiple machine learning components that are built to optimize the search relevance and conversion as an end goal. We show that the proposed aspect recognition machine learning system improved search results quality and relevance greatly compared to the existing already strong baseline system for e-commerce search.