The pre-training models such as BERT have achieved great results in various natural language processing problems. However, a large number of parameters need significant amounts of memory and the consumption of inference time, which makes it difficult to deploy them on edge devices. In this work, we propose a knowledge distillation method LRC-BERT based on contrastive learning to fit the output of the intermediate layer from the angular distance aspect, which is not considered by the existing distillation methods. Furthermore, we introduce a gradient perturbation-based training architecture in the training phase to increase the robustness of LRC-BERT, which is the first attempt in knowledge distillation. Additionally, in order to better capture the distribution characteristics of the intermediate layer, we design a two-stage training method for the total distillation loss. Finally, by verifying 8 datasets on the General Language Understanding Evaluation (GLUE) benchmark, the performance of the proposed LRC-BERT exceeds the existing state-of-the-art methods, which proves the effectiveness of our method.

To reduce the customer service pressure of small and medium-sized enterprises, we propose an intelligent cloud customer service system, called IntelliTag. Unlike traditional customer service, a cloud service based system has difficulty in collecting user personal information. Therefore, we add a tag recommendation function to quickly capture the user's question intent by clicking on the tags. Specifically, IntelliTag is elaborately designed with the consideration of the following three aspects. First, how to mine high-quality tags is a challenging problem. Second, in the tag recommendation tasks, we have multifarious data types and relations that are used to build a sequential recommendation model. Finally, system implementation and deployment also need to be carefully designed to satisfy online service requirements. In this paper, we show the details of data construction, model designs, system implementation and deployment, and the empirical results compared with several state-of-the-art methods. Nowadays, our IntelliTag has already supported hundreds of thousands of enterprises and millions of users in our industrial production environment.

For better user satisfaction and business effectiveness, more and more attention has been paid to the sequence-based recommendation system, which is used to infer the evolution of users’ dynamic preferences, and recent studies have noticed that the evolution of users’ preferences can be better understood from the implicit and explicit feedback sequences. However, most of the existing recommendation techniques do not consider the noise contained in implicit feedback, which will lead to the biased representation of user interest and a suboptimal recommendation performance. Meanwhile, the existing methods utilize item sequence for capturing the evolution of user interest. The performance of these methods is limited by the length of the sequence, and can not effectively model the long-term interest in a long period of time. Based on this observation, we propose a novel CTR model named denoising user-aware memory network (DUMN). Specifically, the framework: (i) proposes a feature purification module based on orthogonal mapping, which use the representation of explicit feedback to purify the representation of implicit feedback, and effectively denoise the implicit feedback; (ii) designs a user memory network to model the long-term interests in a fine-grained way by improving the memory network, which is ignored by the existing methods; and (iii) develops a preference-aware interactive representation component to fuse the long-term and short-term interests of users based on gating to understand the evolution of unbiased preferences of users. Extensive experiments on two real e-commerce user behavior datasets show that DUMN has a significant improvement over the state-of-the-art baselines.

In the past decade, there has been substantial progress at training increasingly deep neural networks. Recent advances within the teacher–student training paradigm have established that information about past training updates show promise as a source of guidance during subsequent training steps. Based on this notion, in this paper, we propose Long Short-Term Sample Distillation, a novel training policy that simultaneously leverages multiple phases of the previous training process to guide the later training updates to a neural network, while efficiently proceeding in just one single generation pass. With Long Short-Term Sample Distillation, the supervision signal for each sample is decomposed into two parts: a long-term signal and a short-term one. The long-term teacher draws on snapshots from several epochs ago in order to provide steadfast guidance and to guarantee teacher–student differences, while the short-term one yields more up-to-date cues with the goal of enabling higher-quality updates. Moreover, the teachers for each sample are unique, such that, overall, the model learns from a very diverse set of teachers. Comprehensive experimental results across a range of vision and NLP tasks demonstrate the effectiveness of this new training method.

Neural sequence model, though widely used for modeling sequential data such as the language model, has sequential recency bias (Kuncoro et al. 2018) to the local context, limiting its full potential to capture long-distance context. To address this problem, this paper proposes augmenting sequence models with a span-based neural buffer that efficiently represents long-distance context, allowing a gate policy network to make interpolated predictions from both the neural buffer and the underlying sequence model. Training this policy network to utilize long-distance context is however challenging due to the simple sentence dominance problem (Marvin and Linzen 2018). To alleviate this problem, we propose a novel training algorithm that combines an annealed maximum likelihood estimation with an intrinsic reward-driven reinforcement learning. Sequence models with the proposed span-based neural buffer significantly improve the state-of-the-art perplexities on the benchmark Penn Treebank and WikiText-2 datasets to 43.9 and 35.2 respectively. We conduct extensive analysis and confirm that the proposed architecture and the training algorithm both contribute to the improvements.

Data-driven approaches using neural networks have achieved promising performances in natural language generation (NLG).However, neural generators are prone to make mistakes, e.g., neglecting an input slot value and generating a redundant slot value.Prior works refer this to hallucination phenomenon.In this paper, we study slot consistency for building reliable NLG systems with all slot values of input dialogue act (DA) properly generated in output sentences.We propose Iterative Rectification Network (IRN) for improving general NLG systems to produce both correct and fluent responses.It applies a bootstrapping algorithm to sample training candidates and uses reinforcement learning to incorporate discrete reward related to slot inconsistency into training.Comprehensive studies have been conducted on multiple benchmark datasets, showing that the proposed methods have significantly reduced the slot error rate (ERR) for all strong baselines.Human evaluations also have confirmed its effectiveness.

As one of the core components of customer service bot, User Intent Prediction (UIP) aims at predicting users? intents (usually represented as predefined user questions) before they ask, and has been widely applied in real applications. However, when developing a machine learning system for this problem, two critical issues, i.e., the problem of feature drift and class imbalance, may emerge and seriously deprave the system performance. Moreover, various scenarios may arise due to business demands, making the aforementioned problems much more severe. To address these two problems, we propose an attention-based Deep Multi-instance Sequential Cross Network (aDMSCN) to deal with the UIP task. On the one hand,the UIP task can be subtly formalized as multi-instance learning(MIL) task with an attention-based method proposed to alleviate the influences of feature drift. To the best of our knowledge, this is the first attempt to model the problem from a MIL perspective.On the other hand, a ratio-sensitive loss is also developed in our model, which can mitigate the negative impact of class imbalance. Extensive experiments on both offline real-world datasets and on-line A/B testing show that our proposed framework significantly out performs other state-of-art methods for the UIP task.

Pooling is an essential component to capture long-term speaker characteristics for speaker recognition. This paper proposes simple but effective pooling methods to compute attentive weights for better temporal aggregation over the variable-length input speech, enabling the end-to-end neural network to have improved performance for discriminating among speakers. Particularly, we observe that using multiple heads for attentive pooling over the entire encoded sequence, a method we term as global multi-head attention, significantly improves performance in comparison to various pooling methods, including the recently proposed multi-head attention [1]. To improve diversity of attention heads, we further propose multi-resolution multi-head attention for pooling that has an additional temperature hyperparameter for each head. This leads to even larger performance gain, on top of that achieved using multiple heads. On the benchmark VoxCeleb1 dataset, the proposed method achieves the state-of-the-art performance of Equal Error Rate (EER) of 3.966%. Our analysis shows that using multiple heads and having multiple resolutions on these heads with different temperatures lead to improved certainty of attentive weights in the new state-of-the-art system.

With the rapid growth of internet finance and the booming of financial lending, the intelligent calling for debt collection in FinTech companies has driven increasing attention. Nowadays, the widely used intelligent calling system is based on dialogue flow, namely configuring the interaction flow with the finite-state machine. In our scenario of debt collection, the completed dialogue flow contains more than one thousand interactive paths. All the dialogue procedures are artificially specified, with extremely high maintenance costs and error-prone. To solve this problem, we propose the behavior-cloning-based collection robot framework without any dialogue flow configuration, called two-stage behavior cloning (TSBC). In the first stage, we use multi-label classification model to obtain policies that may be able to cope with the current situation according to the dialogue state; in the second stage, we score several scripts under each obtained policy to select the script with the highest score as the reply for the current state. This framework makes full use of the massive manual collection records without labeling and fully absorbs artificial wisdom and experience. We have conducted extensive experiments in both single-round and multi-round scenarios and showed the effectiveness of the proposed system. The accuracy of a single round of dialogue can be improved by 5%, and the accuracy of multiple rounds of dialogue can be increased by 3.1%.

With the rapid growth of B2B (Business-to-Business), how to efficiently respond to various customer questions is becoming an important issue. In this scenario, customer questions always involve many aspects of the products, so there are usually multiple customer service agents to response respectively. To improve efficiency, we propose a human-machine cooperation solution called ServiceGroup, where relevant agents and customers are invited into the same group, and the system can provide a series of intelligent functions, including question notification, question recommendation and knowledge extraction. With the assistance of our developed ServiceGroup, the response rate within 15 minutes is improved twice. Until now, our ServiceGroup has already supported thousands of enterprises by means of millions of groups in instant messaging softwares.

Learning sentence similarity is a fundamental research topic and has been explored using various deep learning methods recently. In this paper, we further propose an enhanced recurrent convolutional neural network (Enhanced-RCNN) model for learning sentence similarity. Compared to the state-of-the-art BERT model, the architecture of our proposed model is far less complex. Experimental results show that our similarity learning method outperforms the baselines and achieves the competitive performance on two real-world paraphrase identification datasets.

As one of the major frauds in financial services, cash-out fraud is that users pursue cash gains with illegal or insincere means. Conventional solutions for the cash-out user detection are to perform subtle feature engineering for each user and then apply a classifier, such as GDBT and Neural Network. However, users in financial services have rich interaction relations, which are seldom fully exploited by conventional solutions. In this paper, with the real datasets in Ant Credit Pay of Ant Financial Services Group, we first study the cashout user detection problem and propose a novel hierarchical attention mechanism based cash-out user detection model, called HACUD. Specifically, we model different types of objects and their rich attributes and interaction relations in the scenario of credit payment service with an Attributed Heterogeneous Information Network (AHIN). The HACUD model enhances feature representation of objects through meta-path based neighbors exploiting different aspects of structure information in AHIN. Furthermore, a hierarchical attention mechanism is elaborately designed to model user’s preferences towards attributes and meta-paths. Experimental results on two real datasets show that the HACUD outperforms the state-of-the-art methods.

We present, GeniePath, a scalable approach for learning adaptive receptive fields of neural networks defined on permutation invariant graph data. In GeniePath, we propose an adaptive path layer consists of two complementary functions designed for breadth and depth exploration respectively, where the former learns the importance of different sized neighborhoods, while the latter extracts and filters signals aggregated from neighbors of different hops away. Our method works in both transductive and inductive settings, and extensive experiments compared with competitive methods show that our approaches yield state-of-the-art results on large graphs.

This paper proposes a method to train deep speaker embed-dings end-to-end that jointly optimizes classification and clustering. A large margin softmax loss is used to reduce classification errors. A novel large margin Gaussian mixture loss is proposed to improve clustering. With the joint optimization, the learned embeddings capture segment-level acoustic representation from variable-length speech segments to discriminate between speakers and to replicate densities of speaker clusters. We compare performance with alternative methods on large-scale text-independent speaker recognition dataset VoxCeleb1 [1] and observe that it outperforms those methods significantly, achieving new state-of-the-art results on the dataset. Moreover, because of the joint optimization, this method exhibits faster and better convergence than using classification loss alone. Our results suggest great potential of joint optimization of classification and clustering for speaker verification and identification.

We create an intention mining system, named AntProphet, for Alipay's intelligent customer service bot, to alleviate the burden of customer service. Whenever users have any questions, AntProphet is the first stop to help users to answer their questions. Our system gathers users' profile and their historical behavioral trajectories, together with contextual information to predict users' intention, i.e., the potential questions that users want to resolve. AntProphet takes care of more than 90% of the customer service demands in the Alipay APP and resolves most of the users' problems on the spot, thus significantly reduces the burden of manpower. With the help of it, the overall satisfaction rate of our customer service bot exceeds 85%.

A customer service bot is now a necessary component of an e-commerce platform. As a core module of the customer service bot, user intent prediction can help predict user questions before they ask. A typical solution is to find top candidate questions that a user will be interested in. Such solution ignores the inter-relationship between questions and often aims to maximize the immediate reward such as clicks, which may not be ideal in practice. Hence, we propose to view the problem as a sequential decision making process to better capture the long-term effects of each recommendation in the list. Intuitively, we formulate the problem as a Markov decision process and consider using reinforcement learning for the problem. With this approach, questions presented to users are both relevant and diverse. Experiments on offline real-world dataset and online system demonstrate the effectiveness of our proposed approach.

Fraudulent claim detection is one of the greatest challenges the insurance industry faces. Alibaba's return-freight insurance, providing return-shipping postage compensations over product return on the e-commerce platform, receives thousands of potentially fraudulent claims everyday. Such deliberate abuse of the insurance policy could lead to heavy financial losses. In order to detect and prevent fraudulent insurance claims, we developed a novel data-driven procedure to identify groups of organized fraudsters, one of the major contributions to financial losses, by learning network information. In this paper, we introduce a device-sharing network among claimants, followed by developing an automated solution for fraud detection based on graph learning algorithms, to separate fraudsters from regular customers and uncover groups of organized fraudsters. This solution applied at Alibaba achieves more than 80% precision while covering 44% more suspicious accounts compared with a previously deployed rule-based classifier after human expert investigations. Our approach can easily and effectively generalizes to other types of insurance.

Consumers today face too many reviews to read when shopping online. Presenting the most helpful reviews, instead of all, to them will greatly ease purchase decision making. Most of the existing studies on review helpfulness prediction focused on domains with rich labels, not suitable for domains with insufficient labels. In response, we explore a multi-domain approach that learns domain relationships to help the task by transferring knowledge from data-rich domains to data-deficient domains. To better model domain differences, our approach gates multi-granularity embeddings in a Neural Network (NN) based transfer learning framework to reflect the domain-variant importance of words. Extensive experiments empirically demonstrate that our model outperforms the state-of-the-art baselines and NN-based methods without gating on this task. Our approach facilitates more effective knowledge transfer between domains, especially when the target domain dataset is small. Meanwhile, the domain relationship and domain-specific embedding gating are insightful and interpretable.

With the explosive growth of e-commerce and the booming of e-payment, detecting online transaction fraud in real time has become increasingly important to Fintech business. To tackle this problem, we introduce the TitAnt, a transaction fraud detection system deployed in Ant Financial, one of the largest Fintech companies in the world. The system is able to predict online real-time transaction fraud in mere milliseconds. We present the problem definition, feature extraction, detection methods, implementation and deployment of the system, as well as empirical effectiveness. Extensive experiments have been conducted on large real-world transaction data to show the effectiveness and the efficiency of the proposed system.

Internet companies are facing the need for handling large-scale machine learning applications on a daily basis and distributed implementation of machine learning algorithms which can handle extra-large-scale tasks with great performance is widely needed. Deep forest is a recently proposed deep learning framework which uses tree ensembles as its building blocks and it has achieved highly competitive results on various domains of tasks. However, it has not been tested on extremely large-scale tasks. In this work, based on our parameter server system, we developed the distributed version of deep forest. To meet the need for real-world tasks, many improvements are introduced to the original deep forest model, including MART (Multiple Additive Regression Tree) as base learners for efficiency and effectiveness consideration, the cost-based method for handling prevalent class-imbalanced data, MART based feature selection for high dimension data, and different evaluation metrics for automatically determining the cascade level. We tested the deep forest model on an extra-large-scale task, i.e., automatic detection of cash-out fraud, with more than 100 million training samples. Experimental results showed that the deep forest model has the best performance according to the evaluation metrics from different perspectives even with very little effort for parameter tuning. This model can block fraud transactions in a large amount of money each day. Even compared with the best-deployed model, the deep forest model can additionally bring a significant decrease in economic loss each day.

We propose cw2vec, a novel method for learning Chinese word embeddings. It is based on our observation that exploiting stroke-level information is crucial for improving the learning of Chinese word embeddings. Specifically, we design a minimalist approach to exploit such features, by using stroke n-grams, which capture semantic and morphological level information of Chinese words. Through qualitative analysis, we demonstrate that our model is able to extract semantic information that cannot be captured by existing methods. Empirical results on the word similarity, word analogy, text classification and named entity recognition tasks show that the proposed approach consistently outperforms state-of-the-art approaches such as word-based word2vec and GloVe, character-based CWE, component-based JWE and pixel-based GWE.

Points of interest (POI) recommendation has been drawn much attention recently due to the increasing popularity of location-based networks, e.g., Foursquare and Yelp. Among the existing approaches to POI recommendation, Matrix Factorization (MF) based techniques have proven to be effective. However, existing MF approaches suffer from two major problems: (1) Expensive computations and storages due to the centralized model training mechanism: the centralized learners have to maintain the whole user-item rating matrix, and potentially huge low rank matrices. (2) Privacy issues: the users' preferences are at risk of leaking to malicious attackers via the centralized learner. To solve these, we present a Decentralized MF (DMF) framework for POI recommendation. Specifically, instead of maintaining all the low rank matrices and sensitive rating data for training, we propose a random walk based decentralized training technique to train MF models on each user's end, e.g., cell phone and Pad. By doing so, the ratings of each user are still kept on one's own hand, and moreover, decentralized learning can be taken as distributed learning with multi-learners (users), and thus alleviates the computation and storage issue. Experimental results on two real-world datasets demonstrate that, comparing with the classic and state-of-the-art latent factor models, DMF significantly improvements the recommendation performance in terms of precision and recall.

Ant Credit Pay is a consumer credit service in Ant Financial Service Group. Similar to credit card, loan default is one of the major risks of this credit product. Hence, effective algorithm for default prediction is the key to losses reduction and profits increment for the company. However, the challenges facing in our scenario are different from those in conventional credit card service. The first one is scalability. The huge volume of users and their behaviors in Ant Financial requires the ability to process industrial-scale data and perform model training efficiently. The second challenges is the cold-start problem. Different from the manual review for credit card application in conventional banks, the credit limit of Ant Credit Pay is automatically offered to users based on the knowledge learned from big data. However, default prediction for new users is suffered from lack of enough credit behaviors. It requires that the proposal should leverage other new data source to alleviate the cold-start problem. Considering the above challenges and the special scenario in Ant Financial, we try to incorporate default prediction with network information to alleviate the cold-start problem. In this paper, we propose an industrial-scale distributed network representation framework, termed NetDP, for default prediction in Ant Credit Pay. The proposal explores network information generated by various interaction between users, and blends unsupervised and supervised network representation in a unified framework for default prediction problem. Moreover, we present a parameter-server-based distributed implement of our proposal to handle the scalability challenge. Experimental results demonstrate the effectiveness of our proposal, especially in cold-start problem, as well as the efficiency for industrial-scale dataset.

We present, GEM, the first heterogeneous graph neural network approach for detecting malicious accounts at Alipay, one of the world's leading mobile cashless payment platform. Our approach, inspired from a connected subgraph approach, adaptively learns discriminative embeddings from heterogeneous account-device graphs based on two fundamental weaknesses of attackers, i.e. device aggregation and activity aggregation. For the heterogeneous graph consists of various types of nodes, we propose an attention mechanism to learn the importance of different types of nodes, while using the sum operator for modeling the aggregation patterns of nodes in each type. Experiments show that our approaches consistently perform promising results compared with competitive methods over time.

Collaborative filtering, especially latent factor model, has been popularly used in personalized recommendation. Latent factor model aims to learn user and item latent factors from user-item historic behaviors. To apply it into real big data scenarios, efficiency becomes the first concern, including offline model training efficiency and online recommendation efficiency. In this paper, we propose a D istributed C ollaborative H ashing ( DCH ) model which can significantly improve both efficiencies. Specifically, we first propose a distributed learning framework, following the state-of-the-art parameter server paradigm, to learn the offline collaborative model. Our model can be learnt efficiently by distributedly computing subgradients in minibatches on workers and updating model parameters on servers asynchronously. We then adopt hashing technique to speedup the online recommendation procedure. Recommendation can be quickly made through exploiting lookup hash tables. We conduct thorough experiments on two real large-scale datasets. The experimental results demonstrate that, comparing with the classic and state-of-the-art (distributed) latent factor models, DCH has comparable performance in terms of recommendation accuracy but has both fast convergence speed in offline model training procedure and realtime efficiency in online recommendation procedure. Furthermore, the encouraging performance of DCH is also shown for several real-world applications in Ant Financial.

Nowadays, the number of mobile applications (Apps) is rapidly increasing due to different kinds of user needs. Existing App recommendation models can not scale to practical large data scenarios. In this paper, we present a large scale App recommendation framework, which has been applied into real application in Ant Financial for recommending Apps for hundreds of millions of users. We first present feature engineering, including a novel way to high-level features, and then use a linear ranking model to make recommendation. We also report the encouraging results through the experiments in real large scale dataset.

In this work, we propose to actively detect implicit fraudulent transactions. A novel machine learning method is introduced to distinguish anomalous electronic transactions based on the historical records. The transferor will be alerted during the on-going payment when the fraud probability is recognized as large enough. Compared with elaborative rule-based approaches, our model is much more effective in fraud detection.

Nowadays, online payment systems play more and more important roles in people's daily lives. A key component of these systems is to detect and prevent fraud transactions. In industrial practice, such a task is separated into two phases: 1) mining evidential features to describe users, 2) building an effective model based on these features. Generally speaking, the most popular fraud transaction detection systems use elaborately designed features to build tree based models, sometimes a subsequent linear model is added to improve the behaviour. However, the designed features usually contains only static features, while dynamic features are not considered. In addition, the subsequent model can only learn a linear combination, which may always be unsatisfactory. To address these issues, we present a systematic method, which extracts not only users' static features but also dynamic features based on their recent behaviors. Moreover, N-GRAM model is employed to handle the dynamic features so that time series information is addressed. Based on the extracted features, a tree based model is applied and the outputs of it are regarded as new generated feature representations, which will be further inputted into a Deep Neural Network (DNN) to learn the complex relationships and form the final classification model. Extensive experiments show that our proposed model (with both static and dynamic features) significantly outperforms the existing methods.

We present a neural network based graph embedding method for detecting malicious accounts at Alipay, one of the world's leading mobile payment platform. Our method adaptively learns discriminative embeddings from an account-device graph based on two fundamental weaknesses of attackers, i.e. device aggregation and activity aggregation. Experiments show that our method achieves outstanding precision-recall curve compared with existing methods.

This paper describes a PU learning (Positive and Unlabeled learning) based system for potential URL attack detection. Previous machine learning based solutions for this task mainly formalize it as a supervised learning problem. However, in some scenarios, the data obtained always contains only a handful of known attack URLs, along with a large number of unlabeled instances, making the supervised learning paradigms infeasible. In this work, we formalize this setting as a PU learning problem, and solve it by combining two different strategies (two-stage strategy and cost-sensitive strategy). Experimental results show that the developed system can effectively find potential URL attacks. This system can either be deployed as an assistance for existing system or be employed to help cyber-security engineers to effectively discover potential attack mode so that they can improve the existing system with significantly less efforts.

In recent years, due to the emergence of Big Data (terabytes or petabytes) and Big Model (tens of billions of parameters), there has been an ever-increasing need of parallelizing machine learning (ML) algorithms in both academia and industry. Although there are some existing distributed computing systems, such as Hadoop and Spark, for parallelizing ML algorithms, they only provide synchronous and coarse-grained operators (e.g., Map, Reduce, and Join, etc.), which may hinder developers from implementing more efficient algorithms. This motivated us to design a universal distributed platform termed KunPeng, that combines both distributed systems and parallel optimization algorithms to deal with the complexities that arise from large-scale ML. Specifically, KunPeng not only encapsulates the characteristics of data/model parallelism, load balancing, model sync-up, sparse representation, industrial fault-tolerance, etc., but also provides easy-to-use interface to empower users to focus on the core ML logics. Empirical results on terabytes of real datasets with billions of samples and features demonstrate that, such a design brings compelling performance improvements on ML programs ranging from Follow-the-Regularized-Leader Proximal algorithm to Sparse Logistic Regression and Multiple Additive Regression Trees. Furthermore, KunPeng's encouraging performance is also shown for several real-world applications including the Alibaba's Double 11 Online Shopping Festival and Ant Financial's transaction risk estimation.