Federated learning is a decentralized framework that enables client participation in collaborative learning without centralized data collection. However, the framework is susceptible to suboptimal model convergence induced by heterogeneity among the client datasets. These discrepancies, including label imbalance, dissimilarity in data distributions, and uneven data volumes between clients, may cause disagreements among local client updates, affecting the ability of the global model to converge effectively during aggregation. We suggest that one potential solution to this problem lies in weighting the model aggregation by client importance and client-to-client relationships. Based on this idea, we propose HyperGenFL (HG-FL), a hypernetwork that generates aggregation weights from learnable client embeddings without requiring any training or benchmarking data. HG-FL utilizes the attention mechanism to capture inter-client relationships based on learnable client-specific embeddings in order to generate model aggregation weights dynamically during federated learning. By guiding the aggregation process with these learnable relationships between local models, HG-FL reduces update conflicts and improves global model performance. We assess HG-FL under various data-heterogeneous environments based on different benchmark datasets including Fashion-MNIST, CIFAR10, CIFAR100 and Tiny-ImageNet. Experimental results demonstrate that HG-FL can achieve superior performance over a range of existing baseline methods under challenging cases with various heterogeneous environments, large models and a large number of clients.

Aligning Large Language Models (LLMs) with nuanced user instructions is critical for their effective deployment in real-world applications. While prior methods focus on enhancing data diversity and complexity, they often overlook models' sensitivity to fine-grained variations in semantically similar instructions. To address this, we introduce DeMoRecon, a data augmentation framework that decomposes complex instructions into sub-components, modifies individual elements, and reconstructs them into instruction variants. This method preserves contextual integrity while injecting targeted variability essential for fine-grained instruction-following. Based on DeMoRecon, we construct the FGIV dataset, comprising over 1,700 seed instructions and thousands of nuanced variants designed for both supervised fine-tuning and preference-based alignment. Experimental results show that LLMs trained with FGIV achieve up to +10.2% improvement on our fine-grained FGIV-Eval benchmark and up to +8.8% on existing benchmarks such as FollowBench and InfoBench. These findings highlight the value of FGIV in advancing instruction sensitivity and robustness in LLMs.

The detection of methane emissions has drawn significant attention from researchers, as methane is one of the main contributors to global warming. Due to their high accessibility and extensive coverage, satellite-based remote sensing data have become a widely used tool for studying methane emissions. The multispectral satellite Sentinel-2 is particularly prominent for studies related to methane, as it provides shortwave infrared (SWIR) spectral band data that align with the methane absorption line. Previous studies on methane emission detection using Sentinel-2 data rely mainly on synthetic data, which involves simulated methane emission plumes overlaid on real Sentinel-2 backgrounds. However, our analysis reveals that Sentinel-2 images with simulated plumes exhibit different data distributions and visual characteristics from those of real-world data. Moreover, these studies typically use only a single reference image as a clean background along with the plume image, which we find to be insufficient. Finally, the lack of publicly available benchmarks hinders research and development in this domain. To address these challenges, we introduce the MethaneS2CM dataset for methane emission detection, collected from Sentinel-2 multispectral data and the CarbonMapper platform. Our dataset consists of two main products (L1C and L2A) of Sentinel-2, including over 4000 plumes from 43 countries, ranging from 2016 to 2024. The dataset incorporates both long-term and short-term variations in Sentinel-2 images, with each sample consisting of images captured at different time periods. Furthermore, we conduct extensive experiments to evaluate a wide range of existing methane emission detection methods and models based on this dataset. We subsequently propose a deep methane emission classification model, MEECNet. Experimental results demonstrate that MEECNet outperforms all baseline methods. Our dataset is publicly available at https://huggingface.co/datasets/H1deaki/MethaneS2CM.

Federated learning is a decentralized approach to training machine learning models while preserving data privacy. To accommodate data heterogeneity among clients, a longstanding issue in Federated Learning, many Personalized Federated Learning (PFL) strategies decompose each client model into global modules, which are collaboratively learned by all clients and the server, and local modules, which are only trained locally on private data. While these strategies require every client to participate in training, in reality, many client devices lack sufficient data or computing resources to perform meaningful local training, making it difficult to achieve personalization for every client. In this paper, we present HyperFLoRA, a PFL framework that leverages knowledge learned from training-capable clients to enable the immediate creation of personalized models for training-incapable or new clients. HyperFLoRA uses adapters for personalization to minimize communication costs and client training workload while employing a trainable hypernetwork to generate personalized adapter weights for each client using minimal client statistical information. From experiments conducted on both convolutional and Transformer neural networks, HyperFLoRA can achieve superior model personalization performance for new clients that did not participate in training than conventional PFL methods, while significantly reducing training-related communication costs and client workload.

Modern recommender systems aim to identify items that are most pertinent to a particular user and are particularly useful when an overwhelming number of items are present. Feature embedding is essential to deep recommender systems, which constructs memory-efficient and semantically meaningful representations by mapping high-dimensional sparse feature vectors into low-dimensional dense vectors. Most existing systems assign a unified dimension to all feature fields, regardless of the diverse importance of different features, which usually results in sub-optimal performance and high memory usage. In this paper, we propose a low-cost embedding dimension search approach named DimReg for recommender systems, by assessing information overlapping between the dimensions within each feature field and pruning unimportant and redundant dimensions progressively during model training via a two-level polarization regularizer, while introducing minimum overhead. Moreover, our method does not require retraining after embedding dimension search, which significantly reduces the computational cost and is more friendly to deployment in real-world recommender systems. Extensive experiments conducted on multiple CTR (Click Through Rate) prediction tasks demonstrate that our method can efficiently reduce the model parameters up to 98.6%, and achieve strong recommendation performance outperforming existing automated embedding dimension search methods.

Nested Named Entity Recognition (NNER), as a subarea of Named Entity Recognition, has presented longstanding challenges to researchers. In NNER, one entity may be part of a larger entity, which can occur at multiple levels. These nested structures prevent traditional sequence labeling methods from properly recognizing all entities. While recent research has focused on designing better recognition methods for NNER in various languages, Chinese Nested Named Entity Recognition (CNNER) is still underdeveloped, largely due to a lack of freely available CNNER benchmarks. To support CNNER research, in this paper, we introduce ChiNesE, a CNNER dataset comprising 20,000 sentences from online passages in multiple domains and containing 117,284 entities that fall into 10 categories, of which 43.8% are nested named entities. Based on ChiNesE, we propose Mulco, a novel method that can recognize named entities in nested structures through multiple scopes. Each scope uses a scope-based sequence labeling method that predicts an anchor and the length of a named entity to recognize it. Experimental results show that Mulco outperforms state-of-the-art baseline methods with different recognition schemes on ChiNesE and ACE 2005 Chinese corpus.

Current recommender systems employ large-sized embedding tables with uniform dimensions for all features, leading to overfitting, high computational cost, and suboptimal generalizing performance. Many techniques aim to solve this issue by feature selection or embedding dimension search. However, these techniques typically select a fixed subset of features or embedding dimensions for all instances and feed all instances into one recommender model without considering heterogeneity between items or users. This paper proposes a novel instance-wise Hierarchical Architecture Search framework, iHAS, which automates neural architecture search at the instance level. Specifically, iHAS incorporates three stages: searching, clustering, and retraining. The searching stage identifies optimal instance-wise embedding dimensions across different field features via carefully designed Bernoulli gates with stochastic selection and regularizers. After obtaining these dimensions, the clustering stage divides samples into distinct groups via a deterministic selection approach of Bernoulli gates. The retraining stage then constructs different recommender models, each one designed with optimal dimensions for the corresponding group. We conduct extensive experiments to evaluate the proposed iHAS on two public benchmark datasets from a real-world recommender system. The experimental results demonstrate the effectiveness of iHAS and its outstanding transferability to widely-used deep recommendation models.

Online advertisements are important elements in e-commerce sites, social media platforms, and search engines. With the increasing popularity of mobile browsing, many online ads are displayed with visual information in the form of a cover image in addition to text descriptions to grab the attention of users. Various recent studies have focused on predicting the click rates of online advertisements aware of visual features or composing optimal advertisement elements to enhance visibility. In this paper, we propose Advertisement Style Editing and Attractiveness Enhancement (AdSEE), which explores whether semantic editing to ads images can affect or alter the popularity of online advertisements. We introduce StyleGAN-based facial semantic editing and inversion to ads images and train a click rate predictor attributing GAN-based face latent representations in addition to traditional visual and textual features to click rates. Through a large collected dataset named QQ-AD, containing 20,527 online ads, we perform extensive offline tests to study how different semantic directions and their edit coefficients may impact click rates. We further design a Genetic Advertisement Editor to efficiently search for the optimal edit directions and intensity given an input ad cover image to enhance its projected click rates. Online A/B tests performed over a period of 5 days have verified the increased click-through rates of AdSEE-edited samples as compared to a control group of original ads, verifying the relation between image styles and ad popularity. We open source the code for AdSEE research at https://github.com/LiyaoJiang1998/adsee.

Matrix Factorization (MF) is essential to many estimation tasks. Most existing matrix factorization methods focus on least squares matrix factorization (LSMF), which aims to minimize a smooth L2 loss between observations and their dependent matrix measurement variables. In reality, however, L1 loss and check loss are widely used in regression to deal with outliers or observations contaminated by skewed or heavy-tailed noise. Although under certain conditions, linear convergence to the global optimality can be established for matrix factorization under the L2 loss, there is a lack of provably efficient algorithms for solving matrix factorization under non-smooth losses. In this paper, we investigate Quantile Matrix Factorization (QMF), the counterpart of Quantile Regression in matrix estimation, that adopts a tunable check loss and introduces robustness to matrix estimation for skewed and heavy- tailed observations, which are prevalent in reality. To deal with the non-smooth loss, we propose Nesterov- smoothed QMF (NsQMF), extending Nesterov's optimal smooth approximation technique to the matrix factorization setting. We then present an alternating minimization algorithm to solve the smooth NsQMF efficiently. We mathematically prove that solving the smoothed NsQMF is equivalent to solving the original non-smooth QMF problem and that our proposed algorithm achieves linear convergence to the global optimality of QMF. Numerical evaluations verify our theoretical findings and demonstrate that NsQMF significantly outperforms the commonly used LSMF and prior approximate smoothing heuristics for QMF under various noise distributions.

Understanding and modelling the performance of neural architectures is key to Neural Architecture Search (NAS). Performance predictors have seen widespread use in low-cost NAS and achieve high ranking correlations between predicted and ground truth performance in several NAS benchmarks. However, existing predictors are often designed based on network encodings specific to a predefined search space and are therefore not generalizable to other search spaces or new architecture families. In this paper, we propose a general-purpose neural predictor for NAS that can transfer across search spaces, by representing any given candidate Convolutional Neural Network (CNN) with a Computation Graph (CG) that consists of primitive operators. We further combine our CG network representation with Contrastive Learning (CL) and propose a graph representation learning procedure that leverages the structural information of unlabeled architectures from multiple families to train CG embeddings for our performance predictor. Experimental results on NAS-Bench-101, 201 and 301 demonstrate the efficacy of our scheme as we achieve strong positive Spearman Rank Correlation Coefficient (SRCC) on every search space, outperforming several Zero-Cost Proxies, including Synflow and Jacov, which are also generalizable predictors across search spaces. Moreover, when using our proposed general-purpose predictor in an evolutionary neural architecture search algorithm, we can find highperformance architectures on NAS-Bench-101 and find a MobileNetV3 architecture that attains 79.2% top-1 accuracy on ImageNet.

AutoAugment is an automatic method to design data augmentation policies for deep learning, and has achieved significant improvements on computer vision tasks. However, since early AutoAugment approaches cost thousands of GPU hours, there is a recent demand to investigate low-cost search methods that can still find effective augmentation policies. In this paper, we propose a multi-armed bandit algorithm, named Bandit Data Augment (BDA), to efficiently search for optimal and transferable data augmentation policies. We leverage Successive Halving to make the bandit model progressively focus on more promising augmentation operations during the search, leading to sparse selection of operations and more generalizable augmentation policies. We also propose a computationally efficient rewarding scheme to reduce the evaluation cost of augmentation policies. Extensive experiments demonstrate that BDA can achieve comparable or better performance than prior Auto Augment methods on a wide range of models on CIFAR-10/100 and ImageNet benchmarks. Besides, BDA is 555 times and 536 times faster than AutoAugment on CIFAR-10 and ImageNet, respectively. In addition, BDA is 16 times faster than Fast Auto Augment on ImageNet. More importantly, BDA can discover policies that are transferable across datasets and models, and achieve similar performance to policies found directly on the target dataset.

Cross-domain recommendation is an important method to improve recommender system performance, especially when observations in target domains are sparse. However, most existing techniques focus on single-target or dual-target cross-domain recommendation (CDR) and are hard to be generalized to CDR with multiple target domains. In addition, the negative transfer problem is prevalent in CDR, where the recommendation performance in a target domain may not always be enhanced by knowledge learned from a source domain, especially when the source domain has sparse data. In this study, we propose CAT-ART, a multi-target CDR method that learns to improve recommendations in all participating domains through representation learning and embedding transfer. Our method consists of two parts: a self-supervised Contrastive AuToencoder (CAT) framework to generate global user embeddings based on information from all participating domains, and an Attention-based Representation Transfer (ART) framework which transfers domain-specific user embeddings from other domains to assist with target domain recommendation. CAT-ART boosts the recommendation performance in any target domain through the combined use of the learned global user representation and knowledge transferred from other domains, in addition to the original user embedding in the target domain. We conducted extensive experiments on a collected real-world CDR dataset spanning 5 domains and involving a million users. Experimental results demonstrate the superiority of the proposed method over a range of prior arts. We further conducted ablation studies to verify the effectiveness of the proposed components. Our collected dataset will be open-sourced to facilitate future research in the field of multi-domain recommender systems and user modelling.

Text clustering, as one of the most fundamental challenges in unsupervised learning, aims at grouping semantically similar text segments without relying on human annotations. With the rapid development of deep learning, deep clustering has achieved significant advantages over traditional clustering methods. Despite the effectiveness, most existing deep text clustering methods rely heavily on representations pre-trained in general domains, which may not be the most suitable solution for clustering in specific target domains. To address this issue, we propose CEIL, a novel Classification-Enhanced Iterative Learning framework for short text clustering, which aims at generally promoting the clustering performance by introducing a classification objective to iteratively improve feature representations. In each iteration, we first adopt a language model to retrieve the initial text representations, from which the clustering results are collected using our proposed Category Disentangled Contrastive Clustering (CDCC) algorithm. After strict data filtering and aggregation processes, samples with clean category labels are retrieved, which serve as supervision information to update the language model with the classification objective via a prompt learning approach. Finally, the updated language model with improved representation ability is used to enhance clustering in the next iteration. Extensive experiments demonstrate that the CEIL framework significantly improves the clustering performance over iterations, and is generally effective on various clustering algorithms. Moreover, by incorporating CEIL on CDCC, we achieve the state-of-the-art clustering performance on a wide range of short text clustering benchmarks outperforming other strong baseline methods.

Although conceptualization has been widely studied in semantics and knowledge representation, it is still challenging to find the most accurate concept terms to tag fast-growing social media content. This is partly attributed to the fact that most traditional knowledge bases contain general terms of the world, such as trees and cars, which are not interesting to users, and do not have the defining power for social media content. Another reason is that the intricate use of tense, negation and grammar in social media content may change the logic or emphasis of the content, thus focusing on different main ideas. In this paper, we present TAG, a high-quality concept matching dataset consisting of 10,000 labeled pairs of fine-grained concepts and web-styled natural language sentences, mined from open-domain social media content. The concepts we provide are the trending terms on social media and have the right granularity to define user interests, e.g., highly educated actors instead of just actors. In the meantime, TAG offers a concept graph which interconnects these fine-grained concepts and entities to provide contextual information. We evaluate a wide range of neural text matching models as well as pre-trained language models for the concept matching task on TAG, and point out their insufficiency to tag social media content to characterize its main idea. We further propose a novel graph-graph matching framework that demonstrates superior abstraction and generalization performance by better utilizing both the structural information in the concept graph and logic interactions between semantic units in the natural language sentence via syntactic dependency parsing.

Cross-domain recommendation can help alleviate the data sparsity issue in traditional sequential recommender systems. In this paper, we propose the RecGURU algorithm framework to generate a Generalized User Representation (GUR) incorporating user information across domains in sequential recommendation, even when there is minimum or no common users in the two domains. We propose a self-attentive autoencoder to derive latent user representations, and a domain discriminator, which aims to predict the origin domain of a generated latent representation. We propose a novel adversarial learning method to train the two modules to unify user embeddings generated from different domains into a single global GUR for each user. The learned GUR captures the overall preferences and characteristics of a user and thus can be used to augment the behavior data and improve recommendations in any single domain in which the user is involved. Extensive experiments have been conducted on two public cross-domain recommendation datasets as well as a large dataset collected from real-world applications. The results demonstrate that RecGURU boosts performance and outperforms various state-of-the-art sequential recommendation and cross-domain recommendation methods. The collected data will be released to facilitate future research.

It is common practice for many large e-commerce operators to analyze daily logged transaction data to predict customer purchase behavior, which may potentially lead to more effective recommendations and increased sales. Traditional recommendation techniques based on collaborative filtering, although having gained success in video and music recommendation, are not sufficient to fully leverage the diverse information contained in the implicit user behavior on e-commerce platforms. In this article, we analyze user action records in the Alibaba Mobile Recommendation dataset from the Alibaba Tianchi Data Lab, as well as the Retailrocket recommender system dataset from the Retail Rocket website. To estimate the probability that a user will purchase a certain item tomorrow, we propose a new model called Time-decayed Multifaceted Factorizing Personalized Markov Chains (Time-decayed Multifaceted-FPMC), taking into account multiple types of user historical actions not only limited to past purchases but also including various behaviors such as clicks, collects and add-to-carts. Our model also considers the time-decay effect of the influence of past actions. To learn the parameters in the proposed model, we further propose a unified framework named Bayesian Sparse Factorization Machines. It generalizes the theory of traditional Factorization Machines to a more flexible learning structure and trains the Time-decayed Multifaceted-FPMC with the Markov Chain Monte Carlo method. Extensive evaluations based on multiple real-world datasets demonstrate that our proposed approaches significantly outperform various existing purchase recommendation algorithms.

Neural architecture search (NAS) has achieved remarkable results in deep neural network design. Differentiable architecture search converts the search over discrete architectures into a hyperparameter optimization problem which can be solved by gradient descent. However, questions have been raised regarding the effectiveness and generalizability of gradient methods for solving non-convex architecture hyperparameter optimization problems. In this paper, we propose L2NAS, which learns to intelligently optimize and update architecture hyperparameters via an actor neural network based on the distribution of high-performing architectures in the search history. We introduce a quantile-driven training procedure which efficiently trains L2NAS in an actor-critic framework via continuous-action reinforcement learning. Experiments show that L2NAS achieves state-of-the-art results on NAS-Bench-201 benchmark as well as DARTS search space and Once-for-All MobileNetV3 search space. We also show that search policies generated by L2NAS are generalizable and transferable across different training datasets with minimal fine-tuning.

Neural architecture search automates neural network design and has achieved state-of-the-art results in many deep learning applications. While recent literature has focused on designing networks to maximize accuracy, little work has been conducted to understand the compatibility of architecture design spaces to varying hardware. In this paper, we analyze the neural blocks used to build Once-for-All (MobileNetV3), ProxylessNAS and ResNet families, in order to understand their predictive power and inference latency on various devices, including Huawei Kirin 9000 NPU, RTX 2080 Ti, AMD Threadripper 2990WX, and Samsung Note10. We introduce a methodology to quantify the friendliness of neural blocks to hardware and the impact of their placement in a macro network on overall network performance via only end-to-end measurements. Based on extensive profiling results, we derive design insights and apply them to hardware-specific search space reduction. We show that searching in the reduced search space generates better accuracy-latency Pareto frontiers than searching in the original search spaces, customizing architecture search according to the hardware. Moreover, insights derived from measurements lead to notably higher ImageNet top-1 scores on all search spaces investigated.

Query feeds recommendation is a new recommended paradigm in mobile search applications, where a stream of queries need to be recommended to improve user engagement. It requires a great quantity of attractive queries for recommendation. A conventional solution is to retrieve queries from a collection of past queries recorded in user search logs. However, these queries usually have poor readability and limited coverage of article content, and are thus not suitable for the query feeds recommendation scenario. Furthermore, to deploy the generated queries for recommendation, human validation, which is costly in practice, is required to filter unsuitable queries. In this paper, we propose TitIE, a query mining system to generate valuable queries using the titles of documents. We employ both an extractive text generator and an abstractive text generator to generate queries from titles. To improve the acceptance rate during human validation, we further propose a model-based scoring strategy to pre-select the queries that are more likely to be accepted during human validation. Finally, we propose a novel dual learning approach to jointly learn the generation model and the selection model by making full use of the unlabeled corpora under a semi-supervised scheme, thereby simultaneously improving the performance of both models. Results from both offline and online evaluations demonstrate the superiority of our approach.

Human activity recognition (HAR) based on mobile sensors plays an important role in ubiquitous computing. However, the rise of data regulatory constraints precludes collecting private and labeled signal data from personal devices at scale. Thanks to the growth of computational power on mobile devices, federated learning has emerged as a decentralized alternative solution to model training, which iteratively aggregates locally updated models into a shared global model, therefore being able to leverage decentralized, private data without central collection. However, the effectiveness of federated learning for HAR is affected by the fact that each user has different activity types and even a different signal distribution for the same activity type. Furthermore, it is uncertain if a single global model trained can generalize well to individual users or new users with heterogeneous data. In this paper, we propose Meta-HAR, a federated representation learning framework, in which a signal embedding network is meta-learned in a federated manner, while the learned signal representations are further fed into a personalized classification network at each user for activity prediction. In order to boost the representation ability of the embedding network, we treat the HAR problem at each user as a different task and train the shared embedding network through a Model-Agnostic Meta-learning framework, such that the embedding network can generalize to any individual user. Personalization is further achieved on top of the robustly learned representations in an adaptation procedure. We conducted extensive experiments based on two publicly available HAR datasets as well as a newly created HAR dataset. Results verify that Meta-HAR is effective at maintaining high test accuracies for individual users, including new users, and significantly outperforms several baselines, including Federated Averaging, Reptile and even centralized learning in certain cases. Our collected dataset will be open-sourced to facilitate future development in the field of sensor-based human activity recognition.

Deep learning models for human activity recognition (HAR) based on sensor data have been heavily studied recently. However, the generalization ability of deep models on complex real-world HAR data is limited by the availability of high-quality labeled activity data, which are hard to obtain. In this article, we design a similarity embedding neural network that maps input sensor signals onto real vectors through carefully designed convolutional and Long Short-Term Memory (LSTM) layers. The embedding network is trained with a pairwise similarity loss, encouraging the clustering of samples from the same class in the embedded real space, and can be effectively trained on a small dataset and even on a noisy dataset with mislabeled samples. Based on the learned embeddings, we further propose both nonparametric and parametric approaches for activity recognition. Extensive evaluation based on two public datasets has shown that the proposed similarity embedding network significantly outperforms state-of-the-art deep models on HAR classification tasks, is robust to mislabeled samples in the training set, and can also be used to effectively denoise a noisy dataset.

Concepts naturally constitute an abstraction for fine-grained entities and knowledge in the open domain. They enable search engines and recommendation systems to enhance user experience by discovering high-level abstraction of a search query and the user intent behind it. In this paper, we study the problem of query conceptualization, which is to find the most appropriate matching concepts for any given search query from a large pool of pre-defined concepts. We propose a coarse-to-fine approach to first reduce the search space for each query through a shortlisting scheme and then identify the matching concepts using pre-trained language models, which are meta-tuned to our query-concept matching task. Our shortlisting scheme involves using a GRU-based Relevant Words Generator (RWG) to first expand and complete the context of the given query and then shortlisting the candidate concepts through a scoring mechanism based on word overlaps. To accurately identify the most appropriate matching concepts for a query, even when the concepts may have zero verbatim overlaps with the query, we meta-fine-tune a BERT pairwise text-matching model under the Reptile meta-learning algorithm, which achieves zero-shot transfer learning on the conceptualization problem. Our two-stage framework can be trained with data completely derived from a search click graph, without requiring any human labelling efforts. For evaluation, we have constructed a large click graph based on more than $7$ million instances of the click history recorded in Tencent QQ browser and performed the query conceptualization task based on a large ontology with $159,148$ unique concepts. Results from a range of evaluation methods, including an offline evaluation procedure on the click graph, human evaluation, online A/B testing and case studies, have demonstrated the superiority of our approach over a number of competitive pre-trained language models and fine-tuned neural network baselines.

Understanding what online users may pay attention to on the web is key to content recommendation and search services. These services will benefit from a highly structured and web-scale ontology of entities, concepts, events, topics and categories. While existing knowledge bases and taxonomies embody a large volume of entities and categories, we argue that they fail to discover properly grained concepts, events and topics in the language style of online users. Neither is a logically structured ontology maintained among these notions. In this paper, we present GIANT, a mechanism to construct a user-centered, web-scale, structured ontology, containing a large number of natural language phrases conforming to user attentions at various granularities, mined from the vast volume of web documents and search click logs. Various types of edges are also constructed to maintain a hierarchy in the ontology. We present our detailed techniques used in GIANT, and evaluate the proposed models and methods as compared to a variety of baselines, as well as deploy the resulted Attention Ontology in real-world applications, involving over a billion users, to observe its effect on content recommendation. The online performance of the ontology built by GIANT proves that it can significantly improve the click-through rate in news feeds recommendation.

The ability to ask questions is important in both human and machine intelligence. Learning to ask questions helps knowledge acquisition, improves question-answering and machine reading comprehension tasks, and helps a chatbot to keep the conversation flowing with a human. Existing question generation models are ineffective at generating a large amount of high-quality question-answer pairs from unstructured text, since given an answer and an input passage, question generation is inherently a one-to-many mapping. In this paper, we propose Answer-Clue-Style-aware Question Generation (ACS-QG), which aims at automatically generating high-quality and diverse question-answer pairs from unlabeled text corpus at scale by imitating the way a human asks questions. Our system consists of: i) an information extractor, which samples from the text multiple types of assistive information to guide question generation; ii) neural question generators, which generate diverse and controllable questions, leveraging the extracted assistive information; and iii) a neural quality controller, which removes low-quality generated data based on text entailment. We compare our question generation models with existing approaches and resort to voluntary human evaluation to assess the quality of the generated question-answer pairs. The evaluation results suggest that our system dramatically outperforms state-of-the-art neural question generation models in terms of the generation quality, while being scalable in the meantime. With models trained on a relatively smaller amount of data, we can generate 2.8 million quality-assured question-answer pairs from a million sentences found in Wikipedia.

Extracting events accurately from vast news corpora and organize events logically is critical for news apps and search engines, which aim to organize news information collected from the Internet and present it to users in the most sensible forms. Intuitively speaking, an event is a group of news documents that report the same news incident possibly in different ways. In this article, we describe our experience of implementing a news content organization system at Tencent to discover events from vast streams of breaking news and to evolve news story structures in an online fashion. Our real-world system faces unique challenges in contrast to previous studies on topic detection and tracking (TDT) and event timeline or graph generation, in that we (1) need to accurately and quickly extract distinguishable events from massive streams of long text documents, and (2) must develop the structures of event stories in an online manner, in order to guarantee a consistent user viewing experience. In solving these challenges, we propose Story Forest , a set of online schemes that automatically clusters streaming documents into events, while connecting related events in growing trees to tell evolving stories. A core novelty of our Story Forest system is EventX , a semi-supervised scheme to extract events from massive Internet news corpora. EventX relies on a two-layered, graph-based clustering procedure to group documents into fine-grained events. We conducted extensive evaluations based on (1) 60 GB of real-world Chinese news data, (2) a large Chinese Internet news dataset that contains 11,748 news articles with truth event labels, and (3) the 20 News Groups English dataset, through detailed pilot user experience studies. The results demonstrate the superior capabilities of Story Forest to accurately identify events and organize news text into a logical structure that is appealing to human readers.

Related search query recommendation is a standard feature in many modern search engines. Interesting and relevant queries often increase the active time of users and improve the overall search experience. However, conventional approaches based on tag extraction, keywords matching or click graph link analysis suffer from the common problem of limited coverage and generalizability, which means the system could only make suggestions for a small portion of well-formed search queries.

Most current distributed machine learning systems try to scale up model training by using a data-parallel architecture that divides the computation for different samples among workers. We study distributed machine learning from a different motivation, where the information about the same samples, e.g., users and objects, are owned by several parities that wish to collaborate but do not want to share raw data with each other.

Concepts embody the knowledge of the world and facilitate the cognitive processes of human beings. Mining concepts from web documents and constructing the corresponding taxonomy are core research problems in text understanding and support many downstream tasks such as query analysis, knowledge base construction, recommendation, and search. However, we argue that most prior studies extract formal and overly general concepts from Wikipedia or static web pages, which are not representing the user perspective.

We study the problem of search query inference from web documents, where a short, comprehensive natural language query is inferred from a long article. Search query generation or inference is of great value to search engines and recommenders in terms of locating potential target users and ranking content. Despite being closely related to other NLP tasks like abstract generation and keyword extraction, we point out that search query inference is, in fact, a new problem, in that the generated natural language query, which consists of a few words, is expected to be comprehensive enough to lead to the click-through of the corresponding document. Therefore, query generation requires an accurate inference of query words, as well as a deeper level of understanding on document semantic structures. Toward this end, we propose a novel generative model called the Graph-augmented Sequence to Attention (G-S2A) network. Adopting an Encoder-Decoder architecture, G-S2A incorporates a sentence-level Graph Convolutional Network (GCN), a keyword-level GCN, as well as a hierarchical recurrent neural network (RNN) into the encoder to generate structural document representations. An attentional Transformer decoder is then applied to combine different types of encoded features to generate a target query. On a query-document dataset from a real-world search engine, our model outperforms several neural generative models on a wide range of metrics.

Automatic question generation is an important technique that can improve the training of question answering, help chatbots to start or continue a conversation with humans, and provide assessment materials for educational purposes. Existing neural question generation models are not sufficient mainly due to their inability to properly model the process of how each word in the question is selected, i.e., whether repeating the given passage or being generated from a vocabulary. In this paper, we propose our Clue Guided Copy Network for Question Generation (CGC-QG), which is a sequence-to-sequence generative model with copying mechanism, yet employing a variety of novel components and techniques to boost the performance of question generation. In CGC-QG, we design a multi-task labeling strategy to identify whether a question word should be copied from the input passage or be generated instead, guiding the model to learn the accurate boundaries between copying and generation. Furthermore, our input passage encoder takes as input, among a diverse range of other features, the prediction made by a clue word predictor, which helps identify whether each word in the input passage is a potential clue to be copied into the target question. The clue word predictor is designed based on a novel application of Graph Convolutional Networks onto a syntactic dependency tree representation of each passage, thus being able to predict clue words only based on their context in the passage and their relative positions to the answer in the tree. We jointly train the clue prediction as well as question generation with multi-task learning and a number of practical strategies to reduce the complexity. Extensive evaluations show that our model significantly improves the performance of question generation and out-performs all previous state-of-the-art neural question generation models by a substantial margin.

User action modeling and prediction has long been a topic of importance to recommender systems and user profiling. The quality of the model or accuracy of prediction plays a vital role in related applications like recommendation, advertisement displaying, searching, etc. For large scale systems with a massive number of users, beside the pure prediction performance, there are other practical factors like training and prediction latency, memory overhead, that must be optimized to ensure smooth operation of the system. We propose a fast linear computational framework to handle a vast number of second order crossed features with dimensionality reduction. By leveraging the training and serving system architecture, we shift heavy calculation burden from online serving to offline preprocessing, at the cost of a reasonable amount of memory overhead. The experiments on a 15-day data trace from Tencent MyApp shows that our proposed framework can achieve comparable prediction performance to much complex models like the field-aware factorization machine (FFM) while being served in 2 ms with a reasonable amount of memory overhead.

A large number of deep learning models have been proposed for the text matching problem, which is at the core of various typical natural language processing (NLP) tasks. However, existing deep models are mainly designed for the semantic matching between a pair of short texts, such as paraphrase identification and question answering, and do not perform well on the task of relevance matching between short-long text pairs. This is partially due to the fact that the essential characteristics of short-long text matching have not been well considered in these deep models. More specifically, these methods fail to handle extreme length discrepancy between text pieces and neither can they fully characterize the underlying structural information in long text documents. In this paper, we are especially interested in relevance matching between a piece of short text and a long document, which is critical to problems like query-document matching in information retrieval and web searching. To extract the structural information of documents, an undirected graph is constructed, with each vertex representing a keyword and the weight of an edge indicating the degree of interaction between keywords. Based on the keyword graph, we further propose a Multiresolution Graph Attention Network to learn multi-layered representations of vertices through a Graph Convolutional Network (GCN), and then match the short text snippet with the graphical representation of the document with an attention mechanism applied over each layer of the GCN. Experimental results on two datasets demonstrate that our graph approach outperforms other state-of-the-art deep matching models.

Short Text Matching plays an important role in many natural language processing tasks such as information retrieval, question answering, and conversational system. Conventional text matching methods rely on predefined templates and rules, which are not applicable to short text with limited numebr of words and limit their ability to generalize to unobserved data. Many recent efforts have been made to apply deep neural network models to natural language processing tasks, which reduces the cost of feature engineering. In this paper, we present the design of Multi-Channel Information Crossing , a multi-channel convolutional neural network model for text matching, with additional attention mechanisms from sentence and text semantics. MIX compares text snippets at varied granularities to form a series of multi-channel similarity matrices, which are crossed with another set of carefully designed attention matrices to expose the rich structures of sentences to deep neural networks. We implemented MIX and deployed the system on Tencent's Venus distributed computation platform. Thanks to carefully engineered multi-channel information crossing, evaluation results suggest that MIX outperforms a wide range of state-of-the-art deep neural network models by at least 11.1% in terms of the normalized discounted cumulative gain ([email protected]), on the English WikiQA dataset. Moreover, we also performed online A/B tests with real users on the search service of Tencent QQ Browser. Results suggest that MIX raised the number of clicks on the returned results by 5.7%, due to an increased accuracy in query-document matching, which demonstrates the superior performance of MIX in production environments.

Semantic matching of natural language sentences or identifying the relationship between two sentences is a core research problem underlying many natural language tasks. Depending on whether training data is available, prior research has proposed both unsupervised distance-based schemes and supervised deep learning schemes for sentence matching. However, previous approaches either omit or fail to fully utilize the ordered, hierarchical, and flexible structures of language objects, as well as the interactions between them. In this paper, we propose Hierarchical Sentence Factorization---a technique to factorize a sentence into a hierarchical representation, with the components at each different scale reordered into a "predicate-argument" form. The proposed sentence factorization technique leads to the invention of: 1) a new unsupervised distance metric which calculates the semantic distance between a pair of text snippets by solving a penalized optimal transport problem while preserving the logical relationship of words in the reordered sentences, and 2) new multi-scale deep learning models for supervised semantic training, based on factorized sentence hierarchies. We apply our techniques to text-pair similarity estimation and text-pair relationship classification tasks, based on multiple datasets such as STSbenchmark, the Microsoft Research paraphrase identification (MSRP) dataset, the SICK dataset, etc. Extensive experiments show that the proposed hierarchical sentence factorization can be used to significantly improve the performance of existing unsupervised distance-based metrics as well as multiple supervised deep learning models based on the convolutional neural network (CNN) and long short-term memory (LSTM).

We describe our experience of implementing a news content organization system at Tencent that discovers events from vast streams of breaking news and evolves news story structures in an online fashion. Our real-world system has distinct requirements in contrast to previous studies on topic detection and tracking (TDT) and event timeline or graph generation, in that we 1) need to accurately and quickly extract distinguishable events from massive streams of long text documents that cover diverse topics and contain highly redundant information, and 2) must develop the structures of event stories in an online manner, without repeatedly restructuring previously formed stories, in order to guarantee a consistent user viewing experience. In solving these challenges, we propose Story Forest, a set of online schemes that automatically clusters streaming documents into events, while connecting related events in growing trees to tell evolving stories. We conducted extensive evaluation based on 60 GB of real-world Chinese news data, although our ideas are not language-dependent and can easily be extended to other languages, through detailed pilot user experience studies. The results demonstrate the superior capability of Story Forest to accurately identify events and organize news text into a logical structure that is appealing to human readers, compared to multiple existing algorithm frameworks.

In this paper, we study a new type of spatial sparse recovery problem, that is to infer the fine-grained spatial distribution of certain density data in a region only based on the aggregate observations recorded for each of its subregions. One typical example of this spatial sparse recovery problem is to infer spatial distribution of cellphone activities based on aggregate mobile traffic volumes observed at sparsely scattered base stations. We propose a novel Constrained Spatial Smoothing (CSS) approach, which exploits the local continuity that exists in many types of spatial data to perform sparse recovery via finite-element methods, while enforcing the aggregated observation constraints through an innovative use of the ADMM algorithm. We also improve the approach to further utilize additional geographical attributes. Extensive evaluations based on a large dataset of phone call records and a demographical dataset from the city of Milan show that our approach significantly outperforms various state-of-the-art approaches, including Spatial Spline Regression (SSR).

We study the video recommendation problem based on a large amount of user viewing logs instead of explicit ratings. As viewing records are implicitly suggest user preferences, existing matrix factorization methods fail to generate discriminative recommendations based on such one-class positive samples. We propose a scalable approach called separating-plane matrix factorization (SPMF) to make effective recommendations based on positive implicit feedback, with a learning complexity that is comparable to traditional matrix factorization. With extensive offline evaluation in Tencent Data Warehouse (TDW) based on a large amount of data, we show that our approach outperforms a wide range of state-of-the-art methods. We also deployed our system in the QQ Browser App of Tencent and performed online A/B testing with real users. Results suggest that our approach increased the video click through rate by $23% over implicit-feedback collaborative filtering (IFCF), a scheme available in Apache Spark's MLlib.

Online real-estate information systems such as Zillow and Trulia have gained increasing popularity in recent years. One important feature offered by these systems is the online home price estimate through automated data-intensive computation based on housing information and comparative market value analysis. State-of-the-art approaches model house prices as a combination of a latent land desirability surface and a regression from house features. However, by using uniformly damping kernels, they are unable to handle irregularly shaped regions or capture land value discontinuities within the same region due to the existence of implicit sub-communities, which are common in real-world scenarios. In this paper, we explore the novel application of recent advances in spatial functional analysis to house price modeling and propose the Hierarchical Spatial Functional Model (HSFM), which decomposes house values into land desirability at both the global scale and hidden local scales as well as the feature regression component. We propose statistical learning algorithms based on finite-element spatial functional analysis and spatial constrained clustering to train our model. Extensive evaluations based on housing data in a major Canadian city show that our proposed approach can reduce the mean relative house price estimation error down to 6.60%.