Yu Fu, Chen Luo, Josef Valvoda, Xin Zhang, Xuejing Lei, Xiao Pan, Hui Liu, Yue Dong. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Ziyi Wang, Yuxuan Lu, Yimeng Zhang, Pei Chen, Ziwei Dong, Jing Huang, Jiri Gesi, Xianfeng Tang, Chen Luo, Qun Liu, Yisi Sang, Hanqing Lu, Manling Li, Jin Lai, Dakuo Wang. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Haobo Xu, Sirui Chen, Ruizhong Qiu, Yuchen Yan, Chen Luo, Monica Xiao Cheng, Jingrui He, Hanghang Tong. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2026.

Deploying a conversational shopping agent at production scale remains challenging despite rapid advances in large language models. Unlike research prototypes, production systems must satisfy strict requirements on accuracy, latency, reliability, and cost while serving millions of customers. We share our year-long journey of building a production shopping agent at scale and show that combining agentic reasoning with production-aware retrieval, tool orchestration, and system optimization enables a single shopping agent to support product discovery, shopping question answering, and agentic actions under real-world traffic. Our experience shows that LLM-based agents can be reliably operated at Amazon scale and provides practical design principles for industrial search and recommendation systems.

Weiqi Wang, Limeng Cui, Xin Liu, Sreyashi Nag, Wenju Xu, Chen Luo, Sheikh Muhammad Sarwar, Yang Li, Hansu Gu, Hui Liu, Changlong Yu, Jiaxin Bai, Yifan Gao, Haiyang Zhang, Qi He, Shuiwang Ji, Yangqiu Song. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025.

Few-shot Chain-of-Thought (CoT) prompting has demonstrated strong performance in improving the reasoning capabilities of large language models (LLMs). While theoretical investigations have been conducted to understand CoT, the underlying transformer used in these studies isolates the CoT reasoning process into separated in-context learning steps (Stepwise ICL). In this work, we theoretically show that, compared to Stepwise ICL, the transformer gains better error correction ability and more accurate predictions if the reasoning from earlier steps (Coherent CoT) is integrated. Given that this coherent reasoning changes the behavior of the transformer, we further investigate the sensitivity of the transformer with Coherent CoT when the demonstration examples are corrupted at the inference stage. Our theoretical results indicate that the transformer is more sensitive to errors in intermediate reasoning steps than the final outcome. Building upon this observation, we propose an improvement on CoT by incorporating both correct and incorrect reasoning paths in the demonstration. Our experiments validate the effectiveness of the proposed approach.

Large Language Models (LLMs) are revolutionizing E-Commerce by enabling product recommendation, search, classification, question answering, and advertising applications. Their increasing adoption in real-world systems underscores their potential; however, challenges persist in ensuring accuracy, efficiency, fairness, and privacy. This workshop aims to bring together researchers and industry practitioners to explore both the limitations and opportunities of LLMs in e-commerce. The workshop seeks to foster collaboration, bridge the gap between academia and industry, and drive innovation in the application of LLMs to E-Commerce through discussions on model design, algorithmic advancements, and practical deployment.

Ran Xu, Hui Liu, Sreyashi Nag, Zhenwei Dai, Yaochen Xie, Xianfeng Tang, Chen Luo, Yang Li, Joyce C. Ho, Carl Yang, Qi He. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2025.

Test-time scaling (TTS) enhances the performance of large language models (LLMs) by allocating additional compute resources during inference. However, existing research primarily investigates TTS in single-stage tasks; while many real-world problems are multi-stage complex tasks, composed of a sequence of heterogeneous subtasks with each subtask requires LLM of specific capability. Therefore, we study a novel problem: the test-time compute-optimal scaling in multi-stage complex tasks, aiming to select suitable models and allocate budgets per subtask to maximize overall performance. TTS in multi-stage tasks introduces two fundamental challenges: (i) The combinatorial search space of model and budget allocations, combined with the high cost of inference, makes brute-force search impractical. (ii) The optimal model and budget allocations across subtasks are interdependent, increasing the complexity of the compute-optimal search. To address this gap, we conduct extensive pilot experiments on four tasks across six datasets, deriving three empirical insights characterizing the behavior of LLMs in multi-stage complex tasks. Informed by these insights, we propose AgentTTS, an LLM-agent-based framework that autonomously searches for compute-optimal allocations through iterative feedback-driven interactions with the execution environment. Experimental results demonstrate that AgentTTS significantly outperforms traditional and other LLM-based baselines in search efficiency, and shows improved robustness to varying training set sizes and enhanced interpretability.

The rapid growth of online shopping stores, such as Amazon, has led to services reaching billions of people worldwide. With global retail sales exceeding $6 trillion in 2024, customer expectations for personalized and seamless shopping experiences have heightened. Traditional online shopping experiences, such as search and navigation systems, often fall short in addressing complex shopping journeys. Conversational shopping (such as Amazon Rufus) offers a transformative approach by enabling dynamic, multi-turn dialogues that closely resemble human interactions. This allows customers to explore product options, seek clarifications, and receive personalized recommendations, thereby enhancing product discovery and informed decision-making. In this paper, we share our year-long journey of using language models for conversational shopping at Amazon and introduce how we use LLM fine-tuning techniques to enhance LLMs for a conversational shopping experience like Amazon Rufus. We also introduce innovative strategies for training data collection and demonstrate real-world applications, including product recommendations, clarification mechanisms, and internationalization for global customers.

Prompting LLMs offers an efficient way to guide output generation without explicit model training. In the e-commerce domain, prompt-based applications are widely used in query understanding, recommender systems, and customer support. However, adapting LLMs to different tasks often requires extensive prompt engineering by domain experts, along with frequent updates to align with evolving business needs. Additionally, crafting fully unbiased natural language prompts remains a challenge for humans. To address these challenges, we propose a novel framework, Examples as the Prompt (EaP). Specifically, EaP automatically selects the most representative examples to maximize the few-shot capability of LLMs. It is efficient due to its unsupervised example selection and adaptive to potential data distribution shifts. We validate EaP on four real-world production use cases, demonstrating that it achieves comparable or even superior performance comparing to hand-crafted prompts designed by domain experts. Additionally, we introduce EaPlite, which entirely replaces the natural language components of prompts with labeled examples. EaPlite improves LLM inference speed by up to 70% without compromising performance. The online A/B test shows that using EaP and EaPlite for data labeling can bring significant composite revenue gain by 0.06%.

Regarding computer security, the growth of code vulnerability types presents a persistent challenge. These vulnerabilities, which may cause severe consequences, necessitate precise classification for effective mitigation. However, the rapid emergence of new vulnerability types complicates the classification process. Traditional methodologies, which often involve human expertise and the manual labeling or generation of example instances, are not only resource-intensive but also struggle to adapt to the dynamic nature of these vulnerabilities. This article introducesVulnSense, an innovative method that harnesses the capabilities of Generalized Zero-Shot Learning (GZSL) to address the vulnerability classification problem.VulnSenselearns about unseen vulnerability classes from the descriptions of these unseen classes, while not requiring to see any instances of these unseen classes. Specifically,VulnSenselearns from three main resources: 1) seen classes with labeled code instances; 2) descriptions of these seen classes; and 3) descriptions of “unseen” classes which have no labeled instances. Our experiments underscoreVulnSense's superiority over existing GZSL methods in classifying instances of unseen classes. Concurrently, it maintains a performance parity with traditional labeled-example based learning methods in classifying instances of seen vulnerabilities.VulnSensedemonstrates the potential of using GZSL for vulnerability classification, while also highlighting challenges that inspire future work.

Online shopping platforms, such as Amazon, offer services to billions of people worldwide. Unlike web search or other search engines, product search engines have their unique characteristics, primarily featuring short queries which are mostly a combination of product attributes and structured product search space. The uniqueness of product search underscores the crucial importance of the query understanding component. However, there are limited studies focusing on exploring this impact within real-world product search engines. In this work, we aim to bridge this gap by conducting a comprehensive study and sharing our year-long journey investigating how the query understanding service impacts Amazon Product Search. Firstly, we explore how query understanding-based ranking features influence the ranking process. Next, we delve into how the query understanding system contributes to understanding the performance of a ranking model. Building on the insights gained from our study on the evaluation of the query understanding-based ranking model, we propose a query understanding-based multi-task learning framework for ranking. We present our studies and investigations on Amazon Search.

Fenglin Liu, Zheng Li, Hongjian Zhou, Qingyu Yin, Jingfeng Yang, Xianfeng Tang, Chen Luo, Ming Zeng, Haoming Jiang, Yifan Gao, Priyanka Nigam, Sreyashi Nag, Bing Yin, Yining Hua, Xuan Zhou, Omid Rohanian, Anshul Thakur, Lei Clifton, David A. Clifton. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Haoyu Wang, Ruirui Li, Haoming Jiang, Jinjin Tian, Zhengyang Wang, Chen Luo, Xianfeng Tang, Monica Xiao Cheng, Tuo Zhao, Jing Gao. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Baixuan Xu, Weiqi Wang, Haochen Shi, Wenxuan Ding, Huihao Jing, Tianqing Fang, Jiaxin Bai, Xin Liu, Changlong Yu, Zheng Li, Chen Luo, Qingyu Yin, Bing Yin, Long Chen, Yangqiu Song. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024.

Understanding user intentions is essential for improving product recommendations, navigation suggestions, and query reformulations. However, user intentions can be intricate, involving multiple sessions and attribute requirements connected by logical operators such as And, Or, and Not. For instance, a user may search for Nike or Adidas running shoes across various sessions, with a preference for purple. In another example, a user may have purchased a mattress in a previous session and is now looking for a matching bed frame without intending to buy another mattress. Existing research on session understanding has not adequately addressed making product or attribute recommendations for such complex intentions. In this paper, we present the task of logical session complex query answering (LS-CQA), where sessions are treated as hyperedges of items, and we frame the problem of complex intention understanding as an LS-CQA task on an aggregated hypergraph of sessions, items, and attributes. This is a unique complex query answering task with sessions as ordered hyperedges. We also introduce a new model, the Logical Session Graph Transformer (LSGT), which captures interactions among items across different sessions and their logical connections using a transformer structure. We analyze the expressiveness of LSGT and prove the permutation invariance of the inputs for the logical operators. By evaluating LSGT on three datasets, we demonstrate that it achieves state-of-the-art results.

Xiusi Chen, Hongzhi Wen, Sreyashi Nag, Chen Luo, Qingyu Yin, Ruirui Li, Zheng Li, Wei Wang. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers). 2024.

This paper introduces PAG-a novel optimization and decoding approach that guides autoregressive generation of document identifiers in generative retrieval models through simultaneous decoding. To this aim, PAG constructs a set-based and sequential identifier for each document. Motivated by the bag-of-words assumption in information retrieval, the set-based identifier is built on lexical tokens. The sequential identifier, on the other hand, is obtained via quantizing relevance-based representations of documents. Extensive experiments on MSMARCO and TREC Deep Learning Track data reveal that PAG outperforms the state-of-the-art generative retrieval model by a large margin (e.g., 15.6% MRR improvements on MS MARCO), while achieving 22× speed up in terms of query latency.

Over the past few years, consumer behavior has shifted from traditional in-store shopping to online shopping. For example, eCommerce sales have grown from around 5% of total US sales in 2012 to around 15.4% in year 2023. This rapid growth of eCommerce has created new challenges and vital new requirements for intelligent information retrieval systems. Which lead to the primary motivations of this workshop:

Ranking is at the core of many artificial intelligence (AI) applications, including search engines, recommender systems, etc. Modern ranking systems are often constructed with learning-to-rank (LTR) models built from user behavior signals. While previous studies have demonstrated the effectiveness of using user behavior signals (e.g., clicks) as both features and labels of LTR algorithms, we argue that existing LTR algorithms that indiscriminately treat behavior and non-behavior signals in input features could lead to suboptimal performance in practice. Because user behavior signals often have strong correlations with the ranking objective and can only be collected on items that have already been shown to users, directly using behavior signals in LTR could create an exploitation bias that hurts the system performance in the long run.

Recent research has shown that transformer networks can be used as differentiable search indexes by representing each document as a sequence of document ID tokens. These generative retrieval models cast the retrieval problem to a document ID generation problem for each query. Despite their elegant design, existing generative retrieval models only perform well on artificially-constructed and small-scale collections. This paper represents an important milestone in generative retrieval research by showing that generative retrieval models can be trained to perform effectively on large-scale standard retrieval benchmarks. In more detail, we propose RIPOR- an optimization framework for generative retrieval that is designed based on two often-overlooked fundamental design considerations. First, RIPOR introduces a novel prefix-oriented ranking optimization algorithm for accurate estimation of relevance score during sequential document ID generation. Second, RIPOR constructs document IDs based on the relevance associations between queries and documents. Evaluation on MSMARCO and TREC Deep Learning Track reveals that RIPOR surpasses state-of-the-art generative retrieval models by a large margin (e.g., 30.5% MRR improvements on MS MARCO Dev Set).

A complex logic query in a knowledge graph refers to a query expressed in logic form that conveys a complex meaning, such as where did the Canadian Turing award winner graduate from? Knowledge graph reasoning-based applications, such as dialogue systems and interactive search engines, rely on the ability to answer complex logic queries as a fundamental task. In most knowledge graphs, edges are typically used to either describe the relationships between entities or their associated attribute values. An attribute value can be in categorical or numerical format, such as dates, years, sizes, etc. However, existing complex query answering (CQA) methods simply treat numerical values in the same way as they treat entities. This can lead to difficulties in answering certain queries, such as which Australian Pulitzer award winner is born before 1927, and which drug is a pain reliever and has fewer side effects than Paracetamol. In this work, inspired by the recent advances in numerical encoding and knowledge graph reasoning, we propose numerical complex query answering. In this task, we introduce new numerical variables and operations to describe queries involving numerical attribute values. To address the difference between entities and numerical values, we also propose the framework of Number Reasoning Network (NRN) for alternatively encoding entities and numerical values into separate encoding structures. During the numerical encoding process, NRN employs a parameterized density function to encode the distribution of numerical values. During the entity encoding process, NRN uses established query encoding methods for the original CQA problem. Experimental results show that NRN consistently improves various query encoding methods on three different knowledge graphs and achieves state-of-the-art results.

Modeling customer shopping intentions is a crucial task for e-commerce, as it directly impacts user experience and engagement. Thus, accurately understanding customer preferences is essential for providing personalized recommendations. Session-based recommendation, which utilizes customer session data to predict their next interaction, has become increasingly popular. However, existing session datasets have limitations in terms of item attributes, user diversity, and dataset scale. As a result, they cannot comprehensively capture the spectrum of user behaviors and preferences.To bridge this gap, we present the Amazon Multilingual Multi-locale Shopping Session Dataset, namely Amazon-M2. It is the first multilingual dataset consisting of millions of user sessions from six different locales, where the major languages of products are English, German, Japanese, French, Italian, and Spanish.Remarkably, the dataset can help us enhance personalization and understanding of user preferences, which can benefit various existing tasks as well as enable new tasks. To test the potential of the dataset, we introduce three tasks in this work:(1) next-product recommendation, (2) next-product recommendation with domain shifts, and (3) next-product title generation.With the above tasks, we benchmark a range of algorithms on our proposed dataset, drawing new insights for further research and practice. In addition, based on the proposed dataset and tasks, we hosted a competition in the KDD CUP 2023 https://www.aicrowd.com/challenges/amazon-kdd-cup-23-multilingual-recommendation-challenge and have attracted thousands of users and submissions. The winning solutions and the associated workshop can be accessed at our website~https://kddcup23.github.io/.

Querying knowledge graphs (KGs) using deep learning approaches can naturally leverage the reasoning and generalization ability to learn to infer better answers. Traditional neural complex query answering (CQA) approaches mostly work on entity-centric KGs. However, in the real world, we also need to make logical inferences about events, states, and activities (i.e., eventualities or situations) to push learning systems from System I to System II, as proposed by Yoshua Bengio. Querying logically from an EVentuality-centric KG (EVKG) can naturally provide references to such kind of intuitive and logical inference. Thus, in this paper, we propose a new framework to leverage neural methods to answer complex logical queries based on an EVKG, which can satisfy not only traditional first-order logic constraints but also implicit logical constraints over eventualities concerning their occurrences and orders. For instance, if we know that *Food is bad* happens before *PersonX adds soy sauce*, then *PersonX adds soy sauce* is unlikely to be the cause of *Food is bad* due to implicit temporal constraint. To facilitate consistent reasoning on EVKGs, we propose Complex Eventuality Query Answering (CEQA), a more rigorous definition of CQA that considers the implicit logical constraints governing the temporal order and occurrence of eventualities. In this manner, we propose to leverage theorem provers for constructing benchmark datasets to ensure the answers satisfy implicit logical constraints. We also propose a Memory-Enhanced Query Encoding (MEQE) approach to significantly improve the performance of state-of-the-art neural query encoders on the CEQA task.

Query Parsing aims to extract product attributes, such as color, brand, and product type, from search queries. These attributes play a crucial role in search engines for tasks such as matching, ranking, and recommendation. There are two types of attributes: explicit attributes that are mentioned explicitly in the search query, and implicit attributes that are mentioned implicitly. Existing works on query parsing do not differentiate between explicit query parsing and implicit query parsing, which limits their performance in product search engines. In this work, we demonstrate the critical importance of implicit attributes in real-world product search engines. We then present our solution for implicit query parsing at Amazon Search, which is a unified framework combining recent advancements in knowledge graph technologies and customer behavior analysis. We demonstrate the effectiveness of our proposal through offline experiments on Amazon search log data. We also show how to deploy and use the framework on Amazon search to improve customers' shopping experiences.

Rui Feng, Chen Luo, Qingyu Yin, Bing Yin, Tuo Zhao, Chao Zhang. Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies. 2022.

Query understanding models extract attributes from search queries, like color, product type, brand, etc. Search engines rely on these attributes for ranking, advertising, and recommendation, etc. However, product search queries are usually short, three or four words on average. This information shortage limits the search engine’s power to provide high-quality services.

Using implicit feedback collected from user clicks as training labels for learning-to-rank algorithms is a well-developed paradigm that has been extensively studied and used in modern IR systems. Using user clicks as ranking features, on the other hand, has not been fully explored in existing literature. Despite its potential in improving short-term system performance, whether the incorporation of user clicks as ranking features is beneficial for learning-to-rank systems in the long term is still questionable. Two of the most important problems are (1) the explicit bias introduced by noisy user behavior, and (2) the implicit bias, which we refer to as the exploitation bias, introduced by the dynamic training and serving of learning-to-rank systems with behavior features. In this paper, we explore the possibility of incorporating user clicks as both training labels and ranking features for learning to rank. We formally investigate the problems in feature collection and model training, and propose a counterfactual feature projection function and a novel uncertainty-aware learning to rank framework. Experiments on public datasets show that ranking models learned with the proposed framework can significantly outperform models built with raw click features and algorithms that rank items without considering model uncertainty.

Current popular machine learning techniques in natural language processing and data mining rely heavily on high-quality text sources. Nevertheless, real-world text datasets contain a significant amount of spelling errors and improperly punctuated variants where the performance of these models would quickly deteriorate. Moreover, existing text normalization methods are prohibitively expensive to execute over web-scale datasets, can hardly process noisy texts from social networks, or require annotations to learn the corrections in a supervised manner. In this paper, we present Flan (Fast LSH Algorithm for Text Normalization), a scalable randomized algorithm to clean and canonicalize massive text data. Our approach suggests corrections based on the morphology of the words, where lexically similar words are considered the same with high probability. We efficiently handle the pairwise word-to-word comparisons via locality sensitive hashing (LSH). We also propose a novel stabilization process to address the issue of hash collisions between dissimilar words, which is a consequence of the randomized nature of LSH and is exacerbated by the massive scale of real-world datasets. Compared with existing approaches, our method is more efficient, both asymptotically and in empirical evaluations, does not rely on feature engineering, and does not require any annotation. Our experimental results on real-world datasets demonstrate the efficiency and efficacy of Flan. Based on recent advances in densified Minhash, our approach requires much less computational time compared to baseline text normalization techniques on large-scale Twitter and Reddit datasets. In a human evaluation of the quality of the normalization, Flan achieves 5% and 14% improvement against the baselines over the Reddit and Twitter datasets, respectively. Our method also improves performance on Twitter sentiment classification applications and the perturbed GLUE benchmark datasets, where we introduce random errors into the text.

Detecting anomalies in dynamic graphs is a vital task, with numerous practical applications in areas such as security, finance, and social media. Existing network embedding based methods have mostly focused on learning good node representations, whereas largely ignoring the subgraph structural changes related to the target nodes in a given time window. In this paper, we propose StrGNN, an end-to-end structural temporal Graph Neural Network model for detecting anomalous edges in dynamic graphs. In particular, we first extract the h-hop enclosing subgraph centered on the target edge and propose a node labeling function to identify the role of each node in the subgraph. Then, we leverage the graph convolution operation and Sortpooling layer to extract the fixed-size feature from each snapshot/timestamp. Based on the extracted features, we utilize the Gated Recurrent Units to capture the temporal information for anomaly detection. We fully implement StrGNN and deploy it into a real enterprise security system, and it greatly helps detect advanced threats and optimize the incident response. Extensive experiments on six benchmark datasets also demonstrate the effectiveness of StrGNN.

We study the problem of query attribute value extraction, which aims to identify named entities from user queries as diverse surface form attribute values and afterward transform them into formally canonical forms. Such a problem consists of two phases: named entity recognition (NER) and attribute value normalization (AVN). However, existing works only focus on the NER phase but neglect equally important AVN. To bridge this gap, this paper proposes a unified query attribute value extraction system in e-commerce search named QUEACO, which involves both two phases. Moreover, by leveraging large-scale weakly-labeled behavior data, we further improve the extraction performance with less supervision cost. Specifically, for the NER phase, QUEACO adopts a novel teacher-student network, where a teacher network that is trained on the strongly-labeled data generates pseudo-labels to refine the weakly-labeled data for training a student network. Meanwhile, the teacher network can be dynamically adapted by the feedback of the student's performance on strongly-labeled data to maximally denoise the noisy supervisions from the weak labels. For the AVN phase, we also leverage the weakly-labeled query-to-attribute behavior data to normalize surface form attribute values from queries into canonical forms from products. Extensive experiments on a real-world large-scale E-commerce dataset demonstrate the effectiveness of QUEACO.

It is still challenging in telecommunication (telco) industry to accurately locate mobile devices (MDs) at city-scale using the measurement report (MR) data, which measure parameters of radio signal strengths when MDs connect with base stations (BSs) in telco networks for making/receiving calls or mobile broadband (MBB) services. In this paper, we find that the widely-used location based services (LBSs) have accumulated lots of over-the-top (OTT) global positioning system (GPS) data in telco networks, which can be automatically used as training labels for learning accurate MR-based positioning systems. Benefiting from these telco big data, we deploy a context-aware coarse-to-fine regression (CCR) model in Spark/Hadoop-based telco big data platform for city-scale localization of MDs with two novel contributions. First, we design map-matching and interpolation algorithms to encode contextual information of road networks. Second, we build a two-layer regression model to capture coarse-to-fine contextual features in a short time window for improved localization performance. In our experiments, we collect 108 GPS-associated MR records in the centroid of Shanghai city with 12 x 11 square kilometers for 30 days, and measure four important properties of real-world MR data related to localization errors: stability, sensitivity, uncertainty and missing values. The proposed CCR works well under different properties of MR data and achieves a mean error of 110m and a median error of $80m$, outperforming the state-of-art range-based and fingerprinting localization methods.

Telecommunication (telco) operators aim to provide users with optimized services and bandwidth in a timely manner. The goal is to increase user experience while retaining profit. To do this, knowing the changing behavior patterns of users through their activity levels in advance can be a great help for operators to adjust their management strategies and reduce operational risk. To achieve this goal, the operators can make use of knowledge discovered from telco’s historical mobile broadband (MBB) records to predict mobile access activity level at an early stage. In this article, we report our research in a real-world telco setting involving more than one million telco users. Our novel contribution includes representing users as documents containing a collection of changing spatiotemporal “words” that express user behavior. By extracting users’ space-time access records in MBB data, we use latent Dirichlet allocation (LDA) to learn user-specific compact topic features for user activity level prediction. We propose a scalable online expectation-maximization (OEM) algorithm that can scale LDA to massive MBB data, which is significantly faster than several state-of-the-art online LDA algorithms. Using these real-world MBB data, we confirm high performance in user activity level prediction. In addition, we show that the inferred topics indicate that future activity level anomalies correlate highly with early skewed bandwidth supply and demand relations. Thus, our prediction system can also guide the telco operators to balance the telecommunication network in terms of supply-demand relations, saving deployment costs and energy of cell towers in the future.