7253

7089

The proliferation of textual data containing sensitive personal information across various domains requires robust anonymization techniques to protect privacy and comply with regulations, while preserving data usability for diverse and crucial downstream tasks. This survey provides a comprehen-sive overview of current trends and recent advances in text anonymization techniques. We begin by discussing foundational approaches, primarily centered on Named Entity Recognition, before examining the transformative impact of Large Language Models, detailing their dual role as sophisticated anonymizers and potent de-anonymization threats. The survey further ex-plores domain-specific challenges and tailored solutions in critical sectors such as healthcare, law, finance, and education. We investigate advanced methodologies incorporating formal privacy models and risk-aware frameworks, and address the specialized subfield of authorship anonymization. Additionally, we review evaluation frameworks, comprehensive metrics, benchmarks, and practical toolkits for real-world deployment of anonymization solutions. This review consolidates current knowledge, identifies emerging trends and persistent challenges, including the evolving privacy-utility trade-off, the need to address quasi-identifiers, and the implications of LLM capabilities, and aims to guide future research directions for both academics and practitioners in this field.

Surgical gauze is an essential part of surgical procedures, which is primarily used for controlling bleeding and absorbing bodily fluids. The post-surgical retention of gauze can lead to serious complications in the patient's health and necessitate additional surgery for gauze removal. In the wake of data scarcity, the research on gauze segmentation on the real-world surgical data remains underexplored. In this work, we investigate the use of deep learning methods for gauze segmentation in robotassisted minimally invasive abdominal surgeries, utilizing an inhouse surgical dataset prepared at a university hospital. The training data reflects a realistic surgical setting and extensive diversity in spatial, morphological, and visual attributes of three different gauze categories. We have investigated prevalently used segmentation architectures, including CNN-based, transformer-based, and hybrid architectures, to provide a proof-of-concept for gauze segmentation in a realistic setting. Besides, we investigate the influence of additional sub-optimally annotated, auto-tracked segmentation masks to address the bottleneck of data scarcity and performance optimization. Our results demonstrate the efficacy of real-world data to counter the main challenge reported by prior works - the trade-off between blood presence and gauze detection. The incorporation of auto-track annotations enables performance enhancements, particularly in generic cases. The integration of effective segmentation approaches will benefit robotguided surgical procedures and various downstream applications by providing a precise delineation of foreign objects, enhancing patient safety and surgical outcomes.

Despite Arabic being one of the most widely spoken languages, there is a scarcity of available dialectal Arabic data. In this paper, we address this challenge by proposing a novel approach to data collection through the main use of video captions from TikTok, and other resources such as dictionaries and articles, resulting in the creation of the ArDia dataset. To the best of our knowledge, the ArDia dataset is the largest labeled dialectal Arabic dataset, containing over 900,000 examples, each labeled with its respective dialect. We further leverage this dataset to pretrain transformer-based models, ArDiaBERT and ArDiaGPT. Due to a lack of research on the Arabic models, we present a comprehensive study of Arabic dialect identification using the ArDia dataset on the dialect identification task.

Ever-larger language models with ever-increasing capabilities are by now well-established text processing tools. Alas, information extraction tasks such as named entity recognition are still largely unaffected by this progress as they are primarily based on the previous generation of encoder-only transformer models. Here, we propose a simple yet effective approach, Informed Named Entity Recognition Decoding (iNERD), which treats named entity recognition as a generative process. It leverages the language understanding capabilities of recent generative models in a future-proof manner and employs an informed decoding scheme incorporating the restricted nature of information extraction into open-ended text generation, improving performance and efficiency and eliminating any risk of hallucinations. We coarse-tune our model on a merged named entity corpus to strengthen its performance, evaluate five generative language models on eight named entity recognition datasets, and achieve remarkable results, especially in an environment with an unknown entity class set, demonstrating the adaptability of the approach.

Accurate detection of dementia is crucial for timely intervention and care, and leveraging multimodal data holds significant potential for improving diagnostic accuracy. In this study, we explore deep learning approaches for dementia classification using the Pitt corpus, which includes brief participant descriptions of a cookie theft scene. We analyze 242 control and 307 dementia audio clips to investigate various representation learning techniques. Our best-performing approach fuses audio spectrograms with advanced language models, including Whisper model transcriptions and transformer-based feature extraction. We rigorously evaluate these models and find that our multimodal approach with an F1-score of 86.42% eclipses other single modality approaches by a considerable margin. Our findings underscore the promise of multimodal deep learning techniques in advancing the reliability of dementia detection through audio analysis, possibly paving the way for more robust and accessible diagnostic tools.

Named Entity Recognition (NER) in the legal domain presents unique challenges due to specialized terminology and complex linguistic structures inherent in legal texts. While large language models (LLMs) like GPT-4, Llama-3, and others have significantly advanced natural language processing, their effectiveness in domain-specific tasks like legal Named Entity Recognition remains underexplored. This study conducts a comprehensive comparative analysis of eleven state-of-the-art LLMs on legal NER tasks across seven diverse datasets in five languages, namely English, Portuguese, German, Turkish, and Ukrainian. We evaluate the models’ performance using F1scores, focusing on their ability to accurately identify and classify legal entities. Our findings reveal significant variability in LLM performance across different languages and legal contexts, with proprietary models like GPT-4 achieving the highest overall scores. The results highlight the influence of model architecture, dataset characteristics, and prompt design on the effectiveness of legal NER tasks. This study provides valuable benchmarks for legal NER applications and offers insights into the strengths and limitations of current LLMs, guiding future research and development in legal natural language processing.

We explore the use of Large Language Models (LLMs) for automating the extraction of Key Performance Indicators (KPIs) from diverse financial reports without any additional fine-tuning. We focus on evaluating various proprietary and open-source LLMs to address the joint named entity recognition and relation extraction tasks essential for accurately linking KPIs to their corresponding values and attributes. Our study highlights the technical challenges involved in the extraction process and presents a comprehensive evaluation of the models’ effectiveness. Our results reveal significant insights into handling these LLMs in such a crucial environment and showcase the transformative potential of LLMs in enhancing financial analysis and decision-making.

Model-agnostic explanation methods for deep learning models are flexible regarding usability and availability. However, due to the fact that they can only manipulate input to see changes in output, they suffer from weak performance when used with complex model architectures. For models with large inputs as, for instance, in object detection, sampling-based methods like KernelSHAP are inefficient due to many computation-heavy forward passes through the model. In this work, we present a framework for using sampling-based explanation methods in a computer vision context shown for body part relevance assessment for pedestrian detection. Furthermore, we introduce a novel sampling-based method similar to KernelSHAP that shows more robustness for lower sampling sizes and, thus, is more efficient for explainability analyses on large-scale datasets. We demonstrate our relevance assessment method on simulation data acquired with the CARLA simulator. In the end, our method enables the benchmarking and performance comparison of various pedestrian detection models based on human-interpretable semantic regions.

Thanks to the progress in precision agriculture and remote sensing, automatic visual disease severity scoring of largescale fields is getting more efficient. This potentially allows the investigation and observation of disease spread dynamics and infection incidence in real-world field trials. Furthermore, phytopathological research has been obtaining comprehensive knowledge about pathogens and epidemiology for decades. In this work, we want to leverage modern data-driven Machine Learning (ML) methods and connect them with available knowledge incorporated by simulation techniques. As a result, we show how to create infection datasets from disease severity annotations and environmental data that can serve as training data for a transformer-based ML model. First preliminary results show coherent, interpretable results that motivate for further optimization and improvements of our modeling strategy.

Table structure recognition (TSR), the task of inferring the layout of tables, including the row, column, and cell structure, is a surprisingly complex task. With the growing amount and importance of digital documents, it has become an increasingly relevant problem, which nonetheless has not yet been solved adequately and still presents a very active area of research. In recent years, a growing number of deep-learning-based approaches to table parsing have been proposed.This paper presents a novel deep-learning-based table structure recognition method that can predict row, column, and cell bounds for table images with a high degree of accuracy. To achieve this goal, a multi-stage pipeline incorporating a Vision-Transformer-based Autoencoder model was devised. This model was trained to predict cell regions for table images, from which accurate cell bounds can be inferred, including spanning cells which cover multiple rows or columns. The goal was to obtain a model that generalizes well and can return accurate predictions on various tables of differing complexity, even if they contain little initial structural information.An additional modification to the model architecture presented in the Masked Autoencoder (MAE) approach was also evaluated.

The auditing of financial documents, historically a labor-intensive process, stands on the precipice of transformation. AI-driven solutions have made inroads into streamlining this process by recommending pertinent text passages from financial reports to align with the legal requirements of accounting standards. However, a glaring limitation remains: these systems commonly fall short in verifying if the recommended excerpts indeed comply with the specific legal mandates. Hence, in this paper, we probe the efficiency of publicly available Large Language Models (LLMs) in the realm of regulatory compliance across different model configurations. We place particular emphasis on comparing cutting-edge open-source LLMs, such as Llama-2, with their proprietary counterparts like OpenAI’s GPT models. This comparative analysis leverages two custom datasets provided by our partner PricewaterhouseCoopers (PwC) Germany. We find that the open-source Llama-2 70 billion model demonstrates outstanding performance in detecting non-compliance or true negative occurrences, beating all their proprietary counterparts. Nevertheless, proprietary models such as GPT-4 perform the best in a broad variety of scenarios, particularly in non-English contexts.

Correct identification and correction of contradictions and inconsistencies within financial reports constitute a fundamental component of the audit process. To streamline and automate this critical task, we introduce a novel approach leveraging large language models and an embedding-based paragraph clustering methodology. This paper assesses our approach across three distinct datasets, including two annotated datasets and one unannotated dataset, all within a zero-shot framework. Our findings reveal highly promising results that significantly enhance the effectiveness and efficiency of the auditing process, ultimately reducing the time required for a thorough and reliable financial report audit.

Auditing financial documents is a very tedious and time-consuming process. As of today, it can already be simplified by employing AI-based solutions to recommend relevant text passages from a report for each legal requirement of rigorous accounting standards. However, these methods need to be fine-tuned regularly, and they require abundant annotated data, which is often lacking in industrial environments. Hence, we present ZeroShotALI, a novel recommender system that leverages a state-of-the-art large language model (LLM) in conjunction with a domain-specifically optimized transformer-based text-matching solution. We find that a two-step approach of first retrieving a number of best matching document sections per legal requirement with a custom BERT-based model and second filtering these selections using an LLM yields significant performance improvements over existing approaches.

Despite its great success, machine learning can have its limits when dealing with insufficient training data. A potential solution is the additional integration of prior knowledge into the training process which leads to the notion of informed machine learning. In this paper, we present a structured overview of various approaches in this field. We provide a definition and propose a concept for informed machine learning which illustrates its building blocks and distinguishes it from conventional machine learning. We introduce a taxonomy that serves as a classification framework for informed machine learning approaches. It considers the source of knowledge, its representation, and its integration into the machine learning pipeline. Based on this taxonomy, we survey related research and describe how different knowledge representations such as algebraic equations, logic rules, or simulation results can be used in learning systems. This evaluation of numerous papers on the basis of our taxonomy uncovers key methods in the field of informed machine learning.

We introduce KPI-Check, a novel system that automatically identifies and cross-checks semantically equivalent key performance indicators (KPIs), e.g. "revenue" or "total costs", in real-world German financial reports. It combines a financial named entity and relation extraction module with a BERT-based filtering and text pair classification component to extract KPIs from unstructured sentences before linking them to synonymous occurrences in the balance sheet and profit & loss statement. The tool achieves a high matching performance of 73.00% micro F1on a hold out test set and is currently being deployed for a globally operating major auditing firm to assist the auditing procedure of financial statements.

We consider Automated List Inspection (ALI), a content-based text recommendation system that assists auditors in matching relevant text passages from notes in financial statements to specific law regulations. ALI follows a ranking paradigm in which a fixed number of requirements per textual passage are shown to the user. Despite achieving impressive ranking performance, the user experience can still be improved by showing a dynamic number of recommendations. Besides, existing models rely on a feature-based language model that needs to be pre-trained on a large corpus of domain-specific datasets. Moreover, they cannot be trained in an end-to-end fashion by jointly optimizing with language model parameters. In this work, we alleviate these concerns by considering a multi-label classification approach that predicts dynamic requirement sequences. We base our model on pre-trained BERT that allows us to fine-tune the whole model in an end-to-end fashion, thereby avoiding the need for training a language representation model. We conclude by presenting a detailed evaluation of the proposed model on two German financial datasets.

One-vs-Rest (OVR) classification aims to distinguish a single class of interest from other classes. The concept of novelty detection and robustness to dataset shift becomes crucial in OVR when the scope of the rest class extends from the classes observed during training to unseen and possibly unrelated classes. In this work, we propose a novel architecture, namely Decoupling Autoencoder (DAE) to tackle the common issue of robustness w.r.t. out-of-distribution samples which is prevalent in classifiers such as multi-layer perceptrons (MLP) and ensemble architectures. Experiments on plain classification, outlier detection, and dataset shift tasks show DAE to achieve robust performance across these tasks compared to the baselines, which tend to fail completely, when exposed to dataset shift. While DAE and the baselines yield rather uncalibrated predictions on the outlier detection and dataset shift task, we found that DAE calibration is more stable across all tasks. Therefore, calibration measures applied to the classification task could also improve the calibration of the outlier detection and dataset shift scenarios for DAE.

In this paper, we describe the combination of machine learning and simulation towards a hybrid modelling approach. Such a combination of data-based and knowledge-based modelling is motivated by applications that are partly based on causal relationships, while other effects result from hidden dependencies that are represented in huge amounts of data. Our aim is to bridge the knowledge gap between the two individual communities from machine learning and simulation to promote the development of hybrid systems. We present a conceptual framework that helps to identify potential combined approaches and employ it to give a structured overview of different types of combinations using exemplary approaches of simulation-assisted machine learning and machine-learning assisted simulation. We also discuss an advanced pairing in the context of Industry 4.0 where we see particular further potential for hybrid systems.

The combinatorial problem of max-sum diversification asks for a maximally diverse subset of a given set of data. Here, we show that it can be expressed as an Ising energy minimization problem. Given this result, max-sum diversification can be solved on adiabatic quantum computers and we present proof of concept simulations which support this claim. This, in turn, suggests that quantum computing might play a role in data mining. We therefore discuss quantum computing in a tutorial like manner and elaborate on its current strengths and weaknesses for data analysis.

We present a tool that enables benchmarking of companies by means of automatic extraction of key performance indicators from publicly available financial reports. Our tool monitors companies of interest so that their reports are automatically downloaded as soon as they become available. After tables and paragraphs have been extracted from the documents using a table detection module based on convolutional neural networks, relevant key performance indicators are stored in a central database. The extracted values are finally displayed in a user-friendly web application where the user can compare time series of key performance indicators against arbitrary available companies.

We present the Automated List Inspection (ALI) tool that utilizes methods from machine learning, natural language processing, combined with domain expert knowledge to automate financial statement auditing. ALI is a content based context-aware recommender system, that matches relevant text passages from the notes to the financial statement to specific law regulations. In this paper, we present the architecture of the recommender tool which includes text mining, language modeling, unsupervised and supervised methods that range from binary classification models to deep recurrent neural networks. Next to our main findings, we present quantitative and qualitative comparisons of the algorithms as well as concepts for how to further extend the functionality of the tool.

We present a simple, two step procedure that selects central and extremal prototypes from a given set of data. The key idea is to identify minima of the function that characterizes the interior of a kernel minimum enclosing ball of the data. We discuss how to efficiently compute kernel minimim enclosing balls using the Frank-Wolfe algorithm and show that, for Gaussian kernels, the sought after prototypes can be naturally found via a variant of the mean shift procedure. Practical results demonstrate that prototypes found this way are descriptive, meaningful, and interpretable.

We present an online learning algorithm to extract extremal prototypes from a set of data. As an online algorithm, our method can continue to learn during the application phase of a system. However, as a greedy update procedure, it may be sensitive to outliers. We therefore consider the use of extreme value theory for self-assessment and discuss how to incorporate Weibull statistics so as to increase robustness. We evaluate our approaches on synthetic as well as real world datasets to perform profiling. Our empirical results show that incorporating self-assessment not only results in better data representations but also reveals interpretable insights about the analyzed dataset.

Spectral hashing (SH) seeks compact binary codes of data points so that Hamming distances between codes correlate with data similarity. Quickly learning such codes typically boils down to principle component analysis (PCA). However, this is only justified for normally distributed data. For proportional data (normalized histograms), this is not the case. Due to the sum-to-unity constraint, features that are as independent as possible will not all be uncorrelated. In this paper, we show that a linear-time transformation efficiently copes with sum-to-unity constraints: first, we select a small number K of diverse data points by maximizing the volume of the simplex spanned by these prototypes; second, we represent each data point by means of its cosine similarities to the K selected prototypes. This maximum volume hashing is sensible since each dimension in the transformed space is likely to follow a von Mises (vM) distribution, and, in very high dimensions, the vM distribution closely resembles a Gaussian distribution. This justifies to employ PCA on the transformed data. Our extensive experiments validate this: maximum volume hashing outperforms spectral hashing and other state of the art techniques.

Early stress recognition is of great relevance in precision plant protection. Pre-symptomatic water stress detection is of particular interest, ultimately helping to meet the challenge of “How to feed a hungry world?”. Due to the climate change, this is of considerable political and public interest. Due to its large-scale and temporal nature, e.g., when monitoring plants using hyper-spectral imaging, and the demand of physical meaning of the results, it presents unique computational problems in scale and interpretability. However, big data matrices over time also arise in several other real-life applications such as stock market monitoring where a business sector is characterized by the ups and downs of each of its companies per year or topic monitoring of document collections. Therefore, we consider the general problem of embedding data matrices into Euclidean space over time without making any assumption on the generating distribution of each matrix. To do so, we represent all data samples by means of convex combinations of only few extreme ones computable in linear time. On the simplex spanned by the extremes, there are then natural candidates for distributions inducing distances between and in turn embeddings of the data matrices. We evaluate our method across several domains, including synthetic, text, and financial data as well as a large-scale dataset on water stress detection in plants with more than 3 billion matrix entries. The results demonstrate that the embeddings are meaningful and fast to compute. The stress detection results were validated by a domain expert and conform to existing plant physiological knowledge.

Low-rank approximations which are computed from selected rows and columns of a given data matrix have attracted considerable attention lately. They have been proposed as an alternative to the SVD because they naturally lead to interpretable decompositions which was shown to be successful in application such as fraud detection, fMRI segmentation, and collaborative filtering. The CUR decomposition of large matrices, for example, samples rows and columns according to a probability distribution that depends on the Euclidean norm of rows or columns or on other measures of statistical leverage. At the same time, there are various deterministic approaches that do not resort to sampling and were found to often yield factorization of superior quality with respect to reconstruction accuracy. However, these are hardly applicable to large matrices as they typically suffer from high computational costs. Consequently, many practitioners in the field of data mining have abandon deterministic approaches in favor of randomized ones when dealing with today's large-scale data sets. In this paper, we empirically disprove this prejudice. We do so by introducing a novel, linear-time, deterministic CUR approach that adopts the recently introduced Simplex Volume Maximization approach for column selection. The latter has already been proven to be successful for NMF-like decompositions of matrices of billions of entries. Our exhaustive empirical study on more than 30 synthetic and real-world data sets demonstrates that it is also beneficial for CUR-like decompositions. Compared to other deterministic CUR-like methods, it provides comparable reconstruction quality but operates much faster so that it easily scales to matrices of billions of elements. Compared to sampling-based methods, it provides competitive reconstruction quality while staying in the same run-time complexity class.

There have recently been considerable advances in fast inference for (online) latent Dirichlet allocation (LDA). While it is widely recognized that the scheduling of documents in stochastic optimization and in turn in LDA may have significant consequences, this issue remains largely unexplored. Instead, practitioners schedule documents essentially uniformly at random, due perhaps to ease of implementation, and to the lack of clear guidelines on scheduling the documents.

We introduce and study the spectral evolution model, which characterizes the growth of large networks in terms of the eigenvalue decomposition of their adjacency matrices: In large networks, changes over time result in a change of a graph's spectrum, leaving the eigenvectors unchanged. We validate this hypothesis for several large social, collaboration, authorship, rating, citation, communication and tagging networks, covering unipartite, bipartite, signed and unsigned graphs. Following these observations, we introduce a link prediction algorithm based on the extrapolation of a network's spectral evolution. This new link prediction method generalizes several common graph kernels that can be expressed as spectral transformations. In contrast to these graph kernels, the spectral extrapolation algorithm does not make assumptions about specific growth patterns beyond the spectral evolution model. We thus show that it performs particularly well for networks with irregular, but spectral, growth patterns.

Matrix factorization methods are among the most common techniques for detecting latent components in data. Popular examples include the Singular Value Decomposition or Non-negative Matrix Factorization. Unfortunately, most methods suffer from high computational complexity and therefore do not scale to massive data. In this paper, we present a linear time algorithm for the factorization of gigantic matrices that iteratively yields latent components. We consider a constrained matrix factorization s.t.~the latent components form a simplex that encloses most of the remaining data. The algorithm maximizes the volume of that simplex and thereby reduces the displacement of data from the space spanned by the latent components. Hence, it also lowers the Frobenius norm, a common criterion for matrix factorization quality. Our algorithm is efficient, well-grounded in distance geometry, and easily applicable to matrices with billions of entries. In addition, the resulting factors allow for an intuitive interpretation of data: every data point can now be expressed as a convex combination of the most extreme and thereby often most descriptive instances in a collection of data. Extensive experimental validations on web-scale data, including 80 million images and 1.5 million twitter tweets, demonstrate superior performance compared to related factorization or clustering techniques.

Collaborative tagging services (folksonomies) have been among the stars of theWeb 2.0 era. They allow their users to label diverse resources with freely chosen keywords (tags). Our studies of two real-world folksonomies unveil that individual users develop highly personalized vocabularies of tags. While these meet individual needs and preferences, the considerable differences between personal tag vocabularies (personomies) impede services such as social search or customized tag recommendation. In this paper, we introduce a novel user-centric tag model that allows us to derive mappings between personal tag vocabularies and the corresponding folksonomies. Using these mappings, we can infer the meaning of user-assigned tags and can predict choices of tags a user may want to assign to new items. Furthermore, our translational approach helps in reducing common problems related to tag ambiguity, synonymous tags, or multilingualism. We evaluate the applicability of our method in tag recommendation and tag-based social search. Extensive experiments show that our translational model improves the prediction accuracy in both scenarios.

Non-negative matrix factorization (NMF) has recently received a lot of attention in data mining, information retrieval, and computer vision. It factorizes a non-negative input matrix V into two non-negative matrix factors V = WH such that W describes "clusters" of the datasets. Analyzing genotypes, social networks, or images, it can be beneficial to ensure V to contain meaningful "cluster centroids", i.e., to restrict W to be convex combinations of data points. But how can we run this convex NMF in the wild, i.e., given millions of data points? Triggered by the simple observation that each data point is a convex combination of vertices of the data convex hull, we propose to restrict W further to be vertices of the convex hull. The benefits of this convex-hull NMF approach are twofold. First, the expected size of the convex hull of, for example, n random Gaussian points in the plane is ¿(¿log n), i.e., the candidate set typically grows much slower than the data set. Second, distance preserving low-dimensional embeddings allow one to compute candidate vertices efficiently. Our extensive experimental evaluation shows that convex-hull NMF compares favorably to convex NMF for large data sets both in terms of speed and reconstruction quality. Moreover, we show that our method can easily be applied to large-scale, real-world data sets, in our case consisting of 1.6 million images respectively 150 million votes on World of Warcraft ® guilds.

We analyse the corpus of user relationships of the Slashdot technology news site. The data was collected from the Slashdot Zoo feature where users of the website can tag other users as friends and foes, providing positive and negative endorsements. We adapt social network analysis techniques to the problem of negative edge weights. In particular, we consider signed variants of global network characteristics such as the clustering coefficient, node-level characteristics such as centrality and popularity measures, and link-level characteristics such as distances and similarity measures. We evaluate these measures on the task of identifying unpopular users, as well as on the task of predicting the sign of links and show that the network exhibits multiplicative transitivity which allows algebraic methods based on matrix multiplication to be used. We compare our methods to traditional methods which are only suitable for positively weighted edges.

We propose a hierarchical approach to document categorization that requires no pre-configuration and maps the semantic document space to a predefined taxonomy. The utilization of search engines to train a hierarchical classifier makes our approach more flexible than existing solutions which rely on (human) labeled data and are bound to a specific domain. We show that the structural information given by the taxonomy allows for a context aware construction of search queries and leads to higher tagging accuracy. We test our approach on different benchmark datasets and evaluate its performance on the single- and multi-tag assignment tasks. The experimental results show that our solution is as accurate as supervised classifiers for web page classification and still performs well when categorizing domain specific documents.