An obstacle to scientific document understanding is the extensive use of acronyms which are shortened forms of long technical phrases. Acronym disambiguation aims to find the correct meaning of an ambiguous acronym in a given text. Recent efforts attempted to incorporate word embeddings and deep learning architectures, and achieved significant effects in this task. In general domains, kinds of fine-grained pretrained language models have sprung up, thanks to the largescale corpora which can usually be obtained through crowdsourcing. However, these models based on domain agnostic knowledge might achieve insufficient performance when directly applied to the scientific domain. Moreover, obtaining large-scale high-quality annotated data and representing high-level semantics in the scientific domain is challenging and expensive. In this paper, we consider both the domain agnostic and specific knowledge, and propose a Hierarchical Dual-path BERT method coined hdBERT to capture the general fine-grained and high-level specific representations for acronym disambiguation. First, the context-based pretrained models, RoBERTa and SciBERT, are elaborately involved in encoding these two kinds of knowledge respectively. Second, multiple layer perceptron is devised to integrate the dualpath representations simultaneously and outputs the prediction. With a widely adopted SciAD dataset contained 62,441 sentences, we investigate the effectiveness of hdBERT. The experimental results exhibit that the proposed approach outperforms state-of-the-art methods among various evaluation metrics. Specifically, its macro F1 achieves 93.73

Acronym identification focuses on finding the acronyms and the phrases that have been abbreviated, which is crucial for scientific document understanding tasks. However, the limited size of manually annotated datasets hinders further improvement for the problem. Recent breakthroughs of language models pre-trained on large corpora clearly show that unsupervised pre-training can vastly improve the performance of downstream tasks. In this paper, we present an Adversarial Training BERT method named AT-BERT, our winning solution to acronym identification task for Scientific Document Understanding (SDU) Challenge of AAAI 2021. Specifically, the pre-trained BERT is adopted to capture better semantic representation. Then we incorporate the FGM adversarial training strategy into the fine-tuning of BERT, which makes the model more robust and generalized. Furthermore, an ensemble mechanism is devised to involve the representations learned from multiple BERT variants. Assembling all these components together, the experimental results on the SciAI dataset show that our proposed approach outperforms all other competitive state-of-the-art methods.

Consumer lending service is escalating in E-Commerce platforms due to its capability in enhancing buyers' purchasing power, improving average order value, and increasing revenue of the platforms. Credit risk forecasting and credit limits setting are two fundamental problems in E-Commerce/online consumer lending services. Currently, the majority of institutes rely on two-separate-step methods to resolve. First, build a rating model to evaluate credit risk, and then design heuristic strategies to set credit limits, which requires a large amount of prior knowledge and lacks theoretical justifications. In this paper, we propose an end-to-end multi-view and multi-task learning based approach named MvMoE (Multi-view-aware Mixture-of-Experts network) to solve these two problems simultaneously. First, a multi-view network with a hierarchical attention mechanism is constructed to distill users' heterogeneous financial information into shared hidden representations. Then, we jointly train these two tasks with a view-aware multi-gate mixture-of-experts network and a subsequent progressive network to improve their performances. With the real-world dataset contained 5.44 million users, we investigate the effectiveness of MvMoE. Experimental results exhibit that the proposed model is able to improve AP over 5.60% on credit risk forecasting and MAE over 9.52% on credit limits setting compared with conventional methods. Meanwhile, MvMoE has good interpretability, which better underpins the imperative demands in financial industries.

Due to the vulnerability of deep neural networks (DNNs) to adversarial examples, a large number of defense techniques have been proposed to alleviate this problem in recent years. However, the progress of building more robust models is usually hampered by the incomplete or incorrect robustness evaluation. To accelerate the research on reliable evaluation of adversarial robustness of the current defense models in image classification, the TSAIL group at Tsinghua University and the Alibaba Security group organized this competition along with a CVPR 2021 workshop on adversarial machine learning (https://aisecure-workshop.github.io/amlcvpr2021/). The purpose of this competition is to motivate novel attack algorithms to evaluate adversarial robustness more effectively and reliably. The participants were encouraged to develop stronger white-box attack algorithms to find the worst-case robustness of different defenses. This competition was conducted on an adversarial robustness evaluation platform -- ARES (https://github.com/thu-ml/ares), and is held on the TianChi platform (https://tianchi.aliyun.com/competition/entrance/531847/introduction) as one of the series of AI Security Challengers Program. After the competition, we summarized the results and established a new adversarial robustness benchmark at https://ml.cs.tsinghua.edu.cn/ares-bench/, which allows users to upload adversarial attack algorithms and defense models for evaluation.

Credit risk forecasting generally aims to evaluate the default probability of users in financial service. It is usually regarded as a binary classification problem, which suffers from the severe class-imbalance problem due to the extremely limited throngs and the concept drift problem brought by the delayed verification. In this paper, we investigate these problems in credit risk forecasting and propose a semi-supervised meta-learning based approach called TRUST (TRainable Undersampling with Self Training) to resolve. First, it decides whether to sample the data through meta-learning based reinforcement learning. Secondly, it learns the distribution of the data that have not yet shown financial performance via self-training and updates the model trained in the first step. Finally, the updated model is evaluated on the validation dataset, the result of which will be fed back through the evaluator. These three steps will be iterated until the model converges. With the real-world industrial dataset containing 1.75 million users, we investigate the effectiveness of our method. Experimental results exhibit that the proposed method is able to improve AP over 5.94% on credit risk forecasting task compared with the recent methods.

Financial credit risk assessment serves as the impetus to evaluate the credit admission or potential business failure of customers in order to make early actions prior to the actual financial crisis. It aims to predict the probability that a customer may belong to a high-risk group, which is usually formulated as a binary classification problem. However, due to the lack of high-risk samples, the prevailing models suffer from the severe class-imbalance problem. Oversampling those high-risk users could alleviate this problem but the effect of noise examples is also amplified. In this paper, we propose a novel adversarial data augmentation method to solve the class imbalance problem in financial credit risk assessment. We train a generator for synthetic sample generation with a discriminator to identify real or fake instances. Besides, an auxiliary risk discriminator is trained cooperatively with the generator to assess the credit risk. Experimental results on three real-world datasets demonstrate the effectiveness of the proposed

Default user detection plays one of the backbones in credit risk forecasting and management. It aims at, given a set of corresponding features, e.g., patterns extracted from trading behaviors, predicting the polarity indicating whether a user will fail to make required payments in the future. Recent efforts attempted to incorporate attributed heterogeneous information network (AHIN) for extracting complex interactive features of users and achieved remarkable success on discovering specific default users such as fraud, cash-out users, etc. In this paper, we consider default users, a more general concept in credit risk, and propose a multi-view attributed heterogeneous information network based approach coined MAHINDER to remedy the special challenges. First, multiple views of user behaviors are adopted to learn personal profile due to the endogenous aspect of financial default. Second, local behavioral patterns are specifically modeled since financial default is adversarial and accumulated. With the real datasets contained 1.38 million users on Alibaba platform, we investigate the effectiveness of MAHINDER, and the experimental results exhibit the proposed approach is able to improve AUC over 2.8% and Recall@Precision=0.1 over 13.1% compared with the state-of-the-art methods. Meanwhile, MAHINDER has as good interpretability as tree-based methods like GBDT, which buoys the deployment in online platforms.

Fraud transactions obtain the rights and interests of e-commerce platforms by illegal ways, and have been the emerging threats to the healthy development of these platforms. Recently, user behavioral data is extensively exploited to detect fraud transactions, and it is usually processed as a sequence consisting of individual actions. However, such sequence-like user behaviors have logical patterns associated with user intentions, which motivates a fine-grained management strategy that binds and cuts off these actions into intention-related segments. In this paper, we devise a tree-like structure named behavior tree to reorganize the user behavioral data, in which a group of successive sequential actions denoting a specific user intention are represented as a branch on the tree. We then propose a novel neural method coined LIC Tree-LSTM(Local Intention Calibrated Tree-LSTM) to utilize the behavior tree for fraud transactions detection. In our LIC Tree-LSTM, the global user intention is captured by an attentional method applied on different branches. Then, we calibrate the entire tree by attentions within tree branches to pinpoint the balance between global and local user intentions. We investigate the effectiveness of LIC Tree-LSTM on a real-world dataset of Alibaba platform, and the experimental results show that our proposed algorithm outperforms state-of-the-art methods in both offline and online modes. Furthermore, our model provides good interpretability which helps us better understand user behaviors.

Inquiry performs one of the backbones in current E-commerce websites. Detecting spams in inquiries is essential for these platforms but is surprisingly underexplored by current research. In this paper, we propose a system coined ISTBEL to detect spam inquiries. Motivated by the observations on both behavioral and linguistic differences between spammers and benign users, ISTBEL jointly utilizes the temporal-categorical behavioral sequence and text sequence to accomplish its purpose. At its heart, a variant of LSTM equipped interactive attentions is devised to strengthen the coherence between the categorical and temporal behaviors. A character-level CNN is adopted to capture the capricious linguistic patterns. These two modules are integrated and collaboratively contribute to the final predictions. Our system is trained through an end-to-end manner and can directly utilize raw data as input without tedious feature engineering. We have applied the system to inquiry spam detection on Alibaba.com, and results show ISTBEL is promising in detection spammers and outperforms baselines by a large margin. Meanwhile, ISTBEL can be easily applied to other applications which contain both behavioral and linguistic information.

Evaluation is instrumental to developing and managing effective information retrieval systems. For this process, enlisting crowdsourcing has proven viable. However, less understood are crowdsourcing's limits for evaluation, particularly for domain-specific search. The authors compare relevance assessments gathered using crowdsourcing with those from a domain expert to evaluate different search engines in a large government archive. Although crowdsourced judgments rank the tested search engines in the same order as expert judgments, crowdsourced workers appear unable to distinguish different levels of highly accurate search results the way expert assessors can.

Purpose - The National Archives (TNA) is the UK Government's official archive. It stores and maintains records spanning over a 1,000 years in both physical and digital form. Much of the information held by TNA includes references to place and frequently user queries to TNA's online catalogue involve searches for location. The purpose of this paper is to illustrate how TNA have extracted the geographic references in their historic data to improve access to the archives.Design/methodology/approach - To be able to quickly enhance the existing archival data with geographic information, existing technologies from Natural Language Processing (NLP) and Geographical Information Retrieval (GIR) have been utilised and adapted to historical archives.Findings - Enhancing the archival records with geographic information has enabled TNA to quickly develop a number of case studies highlighting how geographic information can improve access to large-scale archival collections. The use of existing methods from the GIR domain and technologies, such as OpenLayers, enabled one to quickly implement this process in a way that is easily transferable to other institutions.Practical implications - The methods and technologies described in this paper can be adapted, by other archives, to similarly enhance access to their historic data. Also the data-sharing methods described can be used to enable the integration of knowledge held at different archival institutions.Originality/value - Place is one of the core dimensions for TNA's archival data. Many of the records which are held make reference to place data (wills, legislation, court cases), and approximately one fifth of users' searches involve place names. However, there are still a number of open questions regarding the adaptation of existing GIR methods to the history domain. This paper presents an overview over available GIR methods and the challenges in applying them to historical data.

Spatial diversity is a relatively new branch of research in the context of spatial information retrieval. Although the assumption that spatially diversified results may meet users’ needs better seems reasonable, there has been little hard evidence in the literature indicating so. In this paper, we will show the potentials of spatial diversity by not only the traditional evaluation metrics (precision and cluster recall), but also through a user preference study using Amazon Mechanical Turk. The encouraging results from the latter prove that users do have strong preference on spatially diversified results.

Spatial diversity is a relatively new branch of research in the context of spatial information retrieval. It tries to answer user's query with results that are not only relevant but also spatially diversified so that they are from many different locations. Although the assumption that spatially diversified results may meet users' needs better seems reasonable, there has been little hard evidence in the literature indicating so. In this paper, we will show our follow-up work on the novel approach to investigating user preference on spatial diversity by using Amazon Mechanical Turk.

This paper describes a diversity featured retrieval system which is built for the task of ImageCLEFPhoto 2008. Two existing tools are used: Solr and Carrot2. We have experimented with different settings of the system to see how the performance changes. The results suggest that the system can indeed increase diversity of the retrieved results, without sacrificing too much of the precision.

We propose a generic diversity and two novel spatial diversity algorithms for (image) search result diversification. The outputs of the algorithms are compared with the standard search results (which contains no diversity implementation) and found to be promising. In particular, the geometric mean spatial diversity algorithm manages to achieve good geographical diversity while not significantly reducing precision. To the best of our knowledge, such a quantitive evaluation of spatial diversity algorithms for context based image retrieval is new to the community.

This paper describes a study on diversity in image search results. One of the first test collections explicitly built to study diversity – the ImageCLEFPhoto 2008 collection – was used in an evaluation exercise in the summer of 2008. Analyzing 200 of the runs submitted by 24 research groups enabled the relationship between precision and result diversity to be studied. In addition, the level of diversity present in search results produced by retrieval systems built without explicit support for diversity was computed. The remaining potential to improve on diversity was calculated and finally, a significant preference by users for diverse search results was shown.

This paper describes the adaptation of an existing test collection for image retrieval to enable diversity in the results set to be measured. Previous research has shown that a more diverse set of results often satisfies the needs of more users better than standard document rankings. To enable diversity to be quantified, it is necessary to classify images relevant to a given theme to one or more sub-topics or clusters. We describe the challenges in building (as far as we are aware) the first test collection for evaluating diversity in image retrieval. This includes selecting appropriate topics, creating sub-topics, and quantifying the overall effectiveness of a retrieval system. A total of 39 topics were augmented for cluster-based relevance and we also provide an initial analysis of assessor agreement for grouping relevant images into sub-topics or clusters.

In information retrieval, sub-space techniques are usually used to reveal the latent semantic structure of a data-set by projecting it to a low dimensional space. Non-negative matrix factorisation (NMF), which generates a non-negative representation of data through matrix decomposition, is one such technique. It is different from other similar techniques, such as singular vector decomposition (SVD), in its non-negativity constraints which lead to its parts-based representation characteristic. In this paper, we present the novel use of NMF in two tasks; object class detection and automatic annotation of images. Experimental results imply that NMF is a promising sub-space technique for discovering the latent structure of image data-sets, with the ability of encoding the latent topics that correspond to object classes in the basis vectors generated.

We live in the midst of the information era, during which organising and indexing information more effectively is a matter of essential importance. With the fast development of digital imagery, how to search images - a rich form of information - more efficiently by their content has become one of the biggest challenges. Content-based image retrieval (CBIR) has been the traditional and dominant technique for searching images for decades. However, not until recently have researchers started to realise some vital problems existing in CBIR systems. One of the most important is perhaps what people call the \textit{semantic gap}, which refers to the gap between the information that can be extracted from images and the interpretation of the images for humans. As an attempt to bridge the semantic gap, automatic image annotation has been gaining more and more attentions in recent years. This thesis aims to explore a number of different approaches to automatic image annotation and some related issues. It begins with an introduction into different techniques for image description, which forms the foundation of the research on image auto-annotation. The thesis then goes on to give an in-depth examination of some of the quality issues of the data-set used for evaluating auto-annotation systems. A series of approaches to auto-annotation are presented in the follow-up chapters. Firstly, we describe an approach that incorporates the salient based image representation into a statistical model for better annotation performance. Secondly, we explore the use of non-negative matrix factorisation (NMF), a matrix decomposition tehcnique, for two tasks; object class detection and automatic annotation of images. The results imply that NMF is a promising sub-space technique for these purposes. Finally, we propose a model named the image based feature space (IBFS) model for linking image regions and keywords, and for image auto-annotation. Both image regions and keywords are mapped into the same space in which their relationships can be measured. The idea of multiple segmentations is then implemented in the model, and better results are achieved than using a single segmentation.