Data quality is crucial in machine learning (ML) applications, as errors in the data can significantly impact the prediction accuracy of the underlying ML model. Therefore, data cleaning is an integral component of any ML pipeline. However, in practical scenarios, data cleaning incurs significant costs, as it often involves domain experts for configuring and executing the cleaning process. Thus, efficient resource allocation during data cleaning can enhance ML prediction accuracy while controlling expenses. This paper presents COMET, a system designed to optimize data cleaning efforts for ML tasks. COMET gives step-by-step recommendations on which feature to clean next, maximizing the efficiency of data cleaning under resource constraints. We evaluated COMET across various datasets, ML algorithms, and data error types, demonstrating its robustness and adaptability. Our results show that COMET consistently outperforms feature importance-based, random, and another well-known cleaning method, achieving up to 52 and on average 5 percentage points higher ML prediction accuracy than the proposed baselines.

Datasets often contain values that naturally reside in a metric space: numbers, strings, geographical locations, machine-learned embeddings in a vector space, and so on. We study the computational complexity of repairing inconsistent databases that violate integrity constraints, where the database values belong to an underlying metric space. The goal is to update the database values to retain consistency while minimizing the total distance between the original values and the repaired ones. We consider what we refer to as coincidence constraints, which include unary key constraints, inclusion constraints, foreign keys, and generally any restriction on the relationship between the numbers of cells of different labels (attributes) coinciding in a single value, for a fixed attribute set. We begin by showing that the problem is APX-hard for general metric spaces. We then present an algorithm solving the problem optimally for tree metrics, which generalize both the line metric (i.e., where repaired values are numbers) and the discrete metric (i.e., where we simply count the number of changed values). Combining our algorithm for tree metrics and a classic result on probabilistic tree embeddings, we design a (high probability) logarithmic-ratio approximation for general metrics. We also study the variant of the problem where we limit the allowed change of each individual value. In this variant, it is already NP-complete to decide the existence of any legal repair for a general metric, and we present a polynomial-time repairing algorithm for the case of a line metric.

Modern artificial intelligence (AI) applications require large quantities of training and test data. This need creates critical challenges not only concerning the availability of such data, but also regarding its quality. For example, incomplete, erroneous, or inappropriate training data can lead to unreliable models that produce ultimately poor decisions. Trustworthy AI applications require high-quality training and test data along many quality dimensions, such as accuracy, completeness, and consistency. We explore empirically the relationship between six data quality dimensions and the performance of 19 popular machine learning algorithms covering the tasks of classification, regression, and clustering, with the goal of explaining their performance in terms of data quality. Our experiments distinguish three scenarios based on the AI pipeline steps that were fed with polluted data: polluted training data, test data, or both. We conclude the paper with an extensive discussion of our observations.

Knowledge graphs and ontologies play an essential role in integrating, standardizing, and reasoning about complex data across domains. In recent studies, leveraging knowledge graphs in AI use cases, instead of traditional relational databases, led to quality improvements by up to 38 percentage points. However, learning ontologies from relational databases remains a challenging task due to the impedance mismatch between both modeling concepts. An understanding of which ontology learning system performs best, and why, is missing, as no established benchmark exists. We present BURR, a benchmark for evaluating ontology learning systems from relational databases. To evaluate the ontology learning space, we introduce a novel mapping-based metric and provide a comprehensive benchmark data collection. This collection of 54 scenarios consists of real-world database-ontology mappings, including industry data, and of a micro-benchmark evaluating the behavior of systems in encapsulated scenarios. We demonstrate the applicability of BURR by evaluating widely used ontology learning systems, including traditional rule-based as well as LLM-based approaches, on the benchmark. The results emphasize the current strengths of simple rule-based approaches compared to LLM-based systems, while also highlighting the significant research potential of LLMs in ontology learning.

Discovering duplicate or high-overlapping tables in table collections is a crucial task for eliminating redundant information, detecting inconsistencies in the evolution of a table across its multiple versions produced over time, and identifying related tables. Candidate duplicate or related tables to support this task can be identified via the estimation of the largest table overlap. Unfortunately, current solutions for finding it present serious scalability issues for heavy workloads: Sloth, the state of-the-art framework for its estimation, requires more than three days of machine time for computing 100k table overlaps. In this paper, we introduce ARMADILLO, an approach based on graph neural networks that learns table embeddings whose cosine similarity approximates the overlap ratio between tables, i.e., the ratio between the area of their largest table overlap and the area of the smaller table in the pair. We also introduce two new annotated datasets based on GitTables and a Wikipedia table corpus containing 1.32 million table pairs overall labeled with their overlap. Evaluating the performance of ARMADILLO on these datasets, we observed that it is able to calculate overlaps between pairs of tables several times faster than the state-of-the-art method while maintaining a good quality in approximating the exact result.

Data regulations like GDPR require systems to support data erasure but leave the definition of "erasure" open to interpretation. This ambiguity makes compliance challenging, especially in databases where data dependencies can lead to erased data being inferred from remaining data. We formally define a precise notion of data erasure that ensures any inference about deleted data, through dependencies, remains bounded to what could have been inferred before its insertion. We design erasure mechanisms that enforce this guarantee at minimal cost. Additionally, we explore strategies to balance cost and throughput, batch multiple erasures, and proactively compute data retention times when possible. We demonstrate the practicality and scalability of our algorithms using both real and synthetic datasets.

Data practitioners often sample their datasets to produce representative subsets for their downstream tasks. When entities in a dataset can be partitioned into multiple groups, stratified sampling is commonly used to produce subsets that match a target group distribution, e.g., to select a balanced subset for training a machine learning model. However, real-world data frequently contains duplicates — multiple representations of the same real-world entity — that can bias sampling, necessitating deduplication. We define deduplicated sampling as the task of producing a clean sample of a dirty dataset according to a target group distribution. The naïve approach to deduplicated sampling would first deduplicate the entire dataset upfront, then perform sampling ex post. However, that approach might be prohibitively expensive for large datasets and time/resource constraints. Deduplicated sampling ondemand with RadlER is a novel approach to produce a clean sample by focusing the cleaning effort only on entities required to appear in that sample. Our experimental evaluation, performed on multiple datasets from different domains, demonstrates that RadlER consistently outperforms baseline approaches, providing data scientists with an efficient solution to quickly produce a clean sample of a dirty dataset according to a target group distribution.

On the web, huge corpora of tables exist, which can include millions of tables, as in the case of Wikipedia. Maintaining them can be a time-consuming task and, in the case of many authors and editors, also requires a great deal of coordination to ensure high quality, complete, consistent, and readable schemata. In this work, we investigate how to provide automatic suggestions to improve the schema of webtables, namely, how to recommend schema changes. For this purpose, we derive rules from past schema changes via a lattice-based approach and then rank these rules to provide the best-fitting suggestions for each webtable. Making use of the entire edit history of Wikipedia tables, we can compare our suggestions with the changes that were actually performed by editors. We show that for 75.13% of the changes in the test set, we make a correct recommendation, namely a change that was also observed subsequently on Wikipedia. In 58.66% of the cases, our recommendation even covers the entire observed change. Finally, we rank the recommendations with a mean reciprocal rank (MRR) of 0.73 and 0.69 for matches and full matches, respectively. A validation of our approach on three Fandom wikis confirms its effectiveness and generality.

Denial constraints (DCs) are data dependencies with high expressive power, offering great flexibility for modeling data quality rules. Specifying DCs manually is problematic, as the required domain expertise is expensive and scarce. Moreover, database updates can invalidate DCs thought to hold and simultaneously uncover new DCs. This fact leads to burdensome scenarios where experts must often revisit DC specifications. Several algorithms have been devised to discover DCs from data, among which only one considers DC discovery on data updates. However, that solution underperforms in many scenarios due to long runtime and excessive memory use. Also, it targets database inserts only, so no previous solution covers deletions. This paper proposes an efficient and flexible algorithm that covers the earlier limitations regarding performance and scope. The algorithm maintains small-footprint intermediate structures during database updates and a method that exploits the changes in this intermediate to update the DCs incrementally. The results of our extensive experimental evaluation show that our algorithm is orders of magnitude faster than the existing one, with much better scalability in the size of the data updates.

Functional dependencies (FDs) are among the most important integrity constraints in databases. They serve to normalize datasets and thus resolve redundancies, they contribute to query optimization, and they are frequently used to guide data cleaning efforts. Because the FDs of a particular dataset are usually unknown, automatic profiling algorithms are needed to discover them. These algorithms have made considerable advances in the past few years, but they still require a significant amount of time and memory to process datasets of practically relevant sizes. We present FDHits, a novel FD discovery algorithm that finds all valid, minimal FDs in a given relational dataset. FDHits is based on several discovery optimizations that include a hybrid validation approach, effective hitting set enumeration techniques, one-pass candidate validations, and parallelization. Our experiments show that FDHits, even without parallel execution, has a median speedup of 8.1 compared to state-of-the-art FD discovery algorithms while using significantly less memory. This allows the discovery of all FDs even on datasets that could not be processed by the current state-of-the-art.

Both on the Web and in data lakes, it is possible to detect much redundant data in the form of largely overlapping pairs of tables. In many cases, this overlap is not accidental and provides significant information about the relatedness of the tables. Unfortunately, efficiently quantifying the overlap between two tables is not trivial. In particular, detecting their largest overlap, i.e., their largest common subtable, is a computationally challenging problem. As the information overlap may not occur in contiguous portions of the tables, only the ability to permute columns and rows can reveal it. The detection of the largest overlap can help us in relevant tasks such as the discovery of multiple coexisting versions of the same table, which can present differences in the completeness and correctness of the conveyed information. Automatically detecting these highly similar, matching tables would allow us to guarantee their consistency through data cleaning or change propagation, but also to eliminate redundancy to free up storage space or to save additional work for the editors. We present the first formal definition of this problem, and with it Sloth, our solution to efficiently detect the largest overlap between two tables. We experimentally demonstrate on real-world datasets its efficacy in solving this task, analyzing its performance and showing its impact on multiple use cases.

Denial constraints (DCs) are well-known to express business rules on data. They subsume other integrity constraints (ICs), such as key constraints or functional dependencies. One can use traditional DBMS or specialized algorithms to validate such dependencies on a dataset. However, no known approach exists to detect DC violations incrementally. Data typically changes over time, and recomputing the entire violation set after every update is wasteful. Alerting data practitioners of data quality issues immediately, enables them to take measures earlier and can help prevent follow-up issues. We present Weever, the first incremental approach to detect all violations of a given set of DCs. It uses a novel index structure to process inequality predicates and a new method to plan the execution order of predicates depending on their selectivity, reducing redundant computations when handling multiple DCs. Our evaluation shows that Weever outperforms a DBMS-based baseline by up to two orders of magnitude. And in the same time that a state-of-the-art static approach takes to analyze an entire dataset, Weever processes up to 200 000 insertions.

Detecting anomalous subsequences in time series data is one of the key tasks in time series analytics, having applications in environmental monitoring, preventive healthcare, predictive maintenance, and many further areas. Data scientists have developed various anomaly detection algorithms with individual strengths, such as the ability to detect repeating anomalies, anomalies in non-periodic time series, or anomalies with varying lengths. For a given dataset and task, the best algorithm with a suitable parameterization and, in some cases, sufficient training data, usually solves the anomaly detection problem well. However, given the high number of existing algorithms, their numerous parameters, and a pervasive lack of training data and domain knowledge, effective anomaly detection is still a complex task that heavily relies on manual experimentation. We propose the unsupervised AutoTSAD system, which parameterizes, executes, and ensembles various highly effective anomaly detection algorithms. The ensembling system automatically presents an aggregated anomaly scoring for an arbitrary time series without a need for training data or parameter expertise. Our experiments show that AutoTSAD offers an anomaly detection accuracy comparable to the best manually optimized anomaly detection algorithms, and can significantly outperform existing method selection and ensembling approaches for time series anomaly detection.

Open data portals contain a plethora of data files, with comma-separated value (CSV) files being particularly popular with users and businesses due to their flexible standard. However, this flexibility comes with much responsibility for data consumers, as many files contain various structural problems, e.g., a different number of cells across data rows, multiple value formats within the same column, different variants of quoted fields due to user specifications, etc. We refer to rows that contain such structural inconsistencies as ill-formed. Consequently, ingesting them into a host system, such as a database or an analytics platform, often requires prior data preparation steps.

We present role matching, a novel, fine-grained integrity constraint on temporal fact data, i.e., (subject, predicate, object, timestamp)-quadruples. A role is a combination of subject and predicate and can be associated with different objects as the real world evolves and the data changes over time. A role matching states that the associated object of two or more roles should always match across time. Once discovered, role matchings can serve as integrity constraints to improve data quality, for instance of structured data in Wikipedia[3]. If violated, role matchings can alert data owners or editors and thus allow them to correct the error. Finding all role matchings is challenging due both to the inherent quadratic complexity of the matching problem and the need to identify true matches based on the possibly short history of the facts observed so far. To address the first challenge, we introduce several blocking methods both for clean and dirty input data. For the second challenge, the matching stage, we show how the entity resolution method Ditto[27] can be adapted to achieve satisfactory performance for the role matching task. We evaluate our method on datasets from Wikipedia infoboxes, showing that our blocking approaches can achieve 95% recall, while maintaining a reduction ratio of more than 99.99%, even in the presence of dirty data. In the matching stage, we achieve a macro F1-score of 89% on our datasets, using automatically generated labels.

Inclusion dependencies (INDs) are a well-known type of data dependency, specifying that the values of one column are contained in those of another column. INDs can be used for various purposes, such as foreign-key candidate selection or join partner discovery. The traditional notion of INDs is based on clean data, where the dependencies hold without exceptions. Unfortunately, data often contain errors, preventing otherwise valid INDs from being discovered. A typical response to this problem is to relax the dependency definition using a similarity measure to account for minor data errors, such as typos or different formatting. While this relaxation is known for functional dependencies, for inclusion dependencies no such relaxation has been defined. We formally introduce similarity inclusion dependencies, which relax the inclusion by demanding the existence only of sufficiently similar values. Similarity inclusion dependencies can fulfill traditional IND use cases, such as foreign-key candidate discovery, even in the presence of dirty data. We present Sawfish, the first algorithm to discover all similarity inclusion dependencies in a given dataset efficiently. Our algorithm combines approaches for the discovery of traditional INDs and string similarity joins with a novel sliding-window approach and lazy candidate validation. Our experimental evaluation shows that Sawfish can outperform a baseline by a factor of up to 6.5.

Any system at play in a data-driven project has a fundamental requirement: the ability to load data. The de-facto standard format to distribute and consume raw data is csv. Yet, the plain text and flexible nature of this format make such files often difficult to parse and correctly load their content, requiring cumbersome data preparation steps. We propose a benchmark to assess the robustness of systems in loading data from non-standard csv formats and with structural inconsistencies. First, we formalize a model to describe the issues that affect real-world files and use it to derive a systematic "pollution" process to generate dialects for any given grammar. Our benchmark leverages the pollution framework for the csv format. To guide pollution, we have surveyed thousands of real-world, publicly available csv files, recording the problems we encountered. We demonstrate the applicability of our benchmark by testing and scoring 16 different systems: popular csv parsing frameworks, relational database tools, spreadsheet systems, and a data visualization tool.

The task of entity resolution (ER) aims to detect multiple records describing the same real-world entity in datasets and to consolidate them into a single consistent record. ER plays a fundamental role in guaranteeing good data quality, e.g., as input for data science pipelines. Yet, the traditional approach to ER requires cleaning the entire data before being able to run consistent queries on it; hence, users struggle to tackle common scenarios with limited time or resources (e.g., when the data changes frequently or the user is only interested in a portion of the dataset for the task). We previously introduced BrewER, a framework to evaluate SQL SP queries on dirty data while progressively returning results as if they were issued on cleaned data, according to a priority defined by the user. In this demonstration, we show how BrewER can be exploited to ease the burden of ER, allowing data scientists to save a significant amount of resources for their tasks.

Data change, all the time. The Janus project seeks to address the Variability dimension of Big Data by modeling, exploring, and analyzing such change, providing valuable insights into the evolving real world and ways in which data about it are collected and used.

Spreadsheet datasets are valuable sources of data, but often ill-suited for machine consumption. Their unstructured nature allows users to arrange data and metadata freely in a human-readable format, often in canvas-like layouts. To extract their content, data practitioners need to resort to manual inspection and run cumbersome preparation pipelines. The Mondrian system assists users in identifying and handling multiregion layout templates: spreadsheet layouts composed of independent regions that appear repeatedly across different files. Mondrian comprises an automated approach to detect multiple regions within a single file and an algorithm that leverages mapping region layouts to graphs to compute layout similarity and identify templates. Users interact with Mondrian through a web-based visual interface, that serves as a practical toolkit to handle collections of multiregion spreadsheets and enables their automated preparation.

"Bad" data has a direct impact on 88% of companies, with the average company losing 12% of its revenue due to it. Duplicates - multiple but different representations of the same real-world entities - are among the main reasons for poor data quality, so finding and configuring the right deduplication solution is essential. Existing data matching benchmarks focus on the quality of matching results and neglect other important factors, such as business requirements. Additionally, they often do not support the exploration of data matching results. To address this gap between the mere counting of record pairs vs. a comprehensive means to evaluate data matching solutions, we present the Frost platform. It combines existing benchmarks, established quality metrics, cost and effort metrics, and exploration techniques, making it the first platform to allow systematic exploration to understand matching results. Frost is implemented and published in the open-source application Snowman, which includes the visual exploration of matching results, as shown in Figure 1.

Denial constraints (DCs) are an integrity constraint formalism widely used to detect inconsistencies in data. Several algorithms have been devised to discover DCs from data, as manually specifying them is burdensome and, worse yet, error-prone. The existing algorithms follow two basic steps: building an intermediate data structure from records, then enumerating the DCs from that intermediate. However, current algorithms are often inefficient in computing these intermediates. Also, it is still unclear which enumeration algorithm performs best since some of the available algorithms have not yet been compared to each other. In response, we present a set of new algorithms with improved design choices. We introduce a parallel pipeline for rapidly computing the intermediate using custom data representations, algorithms, and indexes. For DC enumeration, we propose an inverted index, pruning, and parallel search strategies. We present hybrid approaches that integrate our techniques with previous enumeration algorithms, improving their performance in many scenarios. Our experimental study shows that the proposed DC discovery algorithms are consistently much faster (up to an order of magnitude) than the current state-of-the-art.

Entity Resolution (ER) aims to identify and merge records that refer to the same real-world entity. ER is typically employed as an expensive cleaning step on the entire data before consuming it. Yet, determining which entities are useful once cleaned depends solely on the user's application, which may need only a fraction of them. For instance, when dealing with Web data, we would like to be able to filter the entities of interest gathered from multiple sources without cleaning the entire, continuously-growing data. Similarly, when querying data lakes, we want to transform data on-demand and return the results in a timely manner---a fundamental requirement of ELT ( Extract-Load-Transform ) pipelines. We propose BrewER , a framework to evaluate SQL SP queries on dirty data while progressively returning results as if they were issued on cleaned data. BrewER tries to focus the cleaning effort on one entity at a time, following an ORDER BY predicate. Thus, it inherently supports top-k and stop-and-resume execution. For a wide range of applications, a significant amount of resources can be saved. We exhaustively evaluate and show the efficacy of BrewER on four real-world datasets.

Abstract Effective query optimization is a core feature of any database management system. While most query optimization techniques make use of simple metadata, such as cardinalities and other basic statistics, other optimization techniques are based on more advanced metadata including data dependencies, such as functional, uniqueness, order, or inclusion dependencies. This survey provides an overview, intuitive descriptions, and classifications of query optimization and execution strategies that are enabled by data dependencies. We consider the most popular types of data dependencies and focus on optimization strategies that target the optimization of relational database queries. The survey supports database vendors to identify optimization opportunities as well as DBMS researchers to find related work and open research questions.

A considerable amount of useful information on the web is (semi-)structured, such as tables and lists. An extensive corpus of prior work addresses the problem of making these human-readable representations interpretable by algorithms. Most of these works focus only on the most recent snapshot of these web objects. However, their evolution over time represents valuable information that has barely been tapped, enabling various applications, including visual change exploration and trust assessment. To realize the full potential of this information, it is critical to match such objects across page revisions.In this work, we present novel techniques that match tables, infoboxes and lists within a page across page revisions. We are, thus, able to extract the evolution of structured information in various forms from a long series of web page revisions. We evaluate our approach on a representative sample of pages and measure the number of correct matches. Our approach achieves a significant improvement in object matching over baselines and over related work.

To assess the quality of data it is useful to extract properties and relationships among them. However, exceptions and approximations need be considered in real-world settings. To this end, relaxed FDs (RFDs) are data dependencies accounting for both exceptions and similarities on data, but their discovery is an extremely complex problem, also due to the necessity of specifying similarity and validity thresholds. The RFD discovery algorithm presented in this paper exploits the concept of dominance to automatically derive similarity thresholds. The discovery performances and the effectiveness of the proposed algorithm are assessed through a comparative evaluation with state-of-art approaches.

Column headers are among the most relevant types of meta-data for relational tables, because they provide meaning and context in which the data is to be interpreted. Headers play an important role in many data integration, exploration, and cleaning scenarios, such as schema matching, knowledge base augmentation, and similarity search. Unfortunately, in many cases column headers are missing, because they were never defined properly, are meaningless, or have been lost during data extraction, transmission, or storage. For example, around one third of the tables on the Web have missing headers. Missing headers leave abundant tabular data shrouded and inaccessible to many data-driven applications.We introduce a fully automated, multi-phase system that discovers table column headers for cases where headers are missing, meaningless, or unrepresentative for the column values. It leverages existing table headers from web tables to suggest human-understandable, representative, and consistent headers for any target table. We evaluate our system on tables extracted from Wikipedia. Overall, 60% of the automatically discovered table headers are exact and complete. Considering more header candidates, top-5 for example, increases this percentage to 72%.

Duplicate detection identifies multiple records in a dataset that represent the same real-world object. Many such approaches exist, both in research and in industry. To investigate essential properties of duplicate detection algorithms, such as their result quality or runtime behavior, they must be executed on suitable test data. The quality evaluation requires that these test data are labeled, constituting a ground truth. Correctly labeled, sizable, and real or at least realistic test datasets, however, are not easy to obtain, creating an obstacle for the advancement of research. In this tutorial, we present common methods to evaluate duplicate detection algorithms and to generate labeled test data. We close with a discussion of open problems.

Knowledge graphs (KGs) form the basis of modern intelligent search systems – their network structure helps with the semantic reasoning and interpretation of complex tasks. A KG is a highly dynamic structure in which facts are continuously updated, added, and removed. A typical approach to ensure data quality in the presence of continuous changes is to apply logic rules. These rules are automatically mined from the data using frequency-based approaches. As a result, these approaches depend on the data quality of the KG and are susceptible to errors and incompleteness.

The detection of constraint-based errors is a critical task in many data cleaning solutions. Previous works perform the task either using traditional data management systems or using specialized systems that speed up error detection. Unfortunately, both approaches may fail to execute in a reasonable time or even exhaust the available memory in the attempt. To address the main drawbacks of previous approaches, we present the FAst Constraint-based Error DeTector (FACET) to detect violations of denial constraints (DCs). FACET uses column sketch information to organize a pipeline of special operators for DC predicates and it implements these operators using a set of efficient algorithms and data structures that adapt to different data characteristics and predicate structures. We evaluate our system on a diverse array of datasets and constraints, showing its robustness and performance gains compared to different types of DBMSs and to a specialized system.

Spreadsheets are among the most commonly used file formats for data management, distribution, and analysis. Their widespread employment makes it easy to gather large collections of data, but their flexible canvas-based structure makes automated analysis difficult without heavy preparation. One of the common problems that practitioners face is the presence of multiple, independent regions in a single spreadsheet, possibly separated by repeated empty cells. We define such files as "multiregion" files. In collections of various spreadsheets, we can observe that some share the same layout. We present the Mondrian approach to automatically identify layout templates across multiple files and systematically extract the corresponding regions. Our approach is composed of three phases: first, each file is rendered as an image and inspected for elements that could form regions; then, using a clustering algorithm, the identified elements are grouped to form regions; finally, every file layout is represented as a graph and compared with others to find layout templates. We compare our method to state-of-the-art table recognition algorithms on two corpora of real-world enterprise spreadsheets. Our approach shows the best performances in detecting reliable region boundaries within each file and can correctly identify recurring layouts across files.

With the advent of big data and data lakes, data are often integrated from multiple sources. Such integrated data are often of poor quality, due to inconsistencies, errors, and so forth. One way to check the quality of data is to infer functional dependencies (fds). However, in many modern applications it might be necessary to extract properties and relationships that are not captured through fds, due to the necessity to admit exceptions, or to consider similarity rather than equality of data values. Relaxed fds (rfds) have been introduced to meet these needs, but their discovery from data adds further complexity to an already complex problem, also due to the necessity of specifying similarity and validity thresholds. We propose Domino, a new discovery algorithm for rfds that exploits the concept of dominance in order to derive similarity thresholds of attribute values while inferring rfds. An experimental evaluation on real datasets demonstrates the discovery performance and the effectiveness of the proposed algorithm.

Research on data dependencies has experienced a revival as dependency violations can reveal errors in data. Several data cleaning systems use a DBMS to detect such violations. While DBMSs are efficient for some kinds of data dependencies (e.g., unique constraints), they are likely to fall short of satisfactory performance for more complex ones, such as order dependencies.

Wikipedia is the largest encyclopedia to date. Scattered among its articles, there is an enormous number of tables that contain structured, relational information. In contrast to database tables, these webtables lack metadata, making it difficult to automatically interpret the knowledge they harbor. The natural key is a particularly important piece of metadata, which acts as a primary key and consists of attributes inherent to an entity. Determining natural keys is crucial for many tasks, such as information integration, table augmentation, or tracking changes to entities over time.

Duplicate detection is an integral part of data cleaning and serves to identify multiple representations of same real-world entities in (relational) datasets. Existing duplicate detection approaches are effective, but they are also hard to parameterize or require a lot of pre-labeled training data. Both parameterization and pre-labeling are at least domain-specific if not dataset-specific, which is a problem if a new dataset needs to be cleaned. For this reason, we propose a novel, rule-based and fully automatic duplicate detection approach that is based on matching dependencies (MDs). Our system uses automatically discovered MDs, various dataset features, and known gold standards to train a model that selects MDs as duplicate detection rules. Once trained, the model can select useful MDs for duplicate detection on any new dataset. To increase the generally low recall of MD-based data cleaning approaches, we propose an additional boosting step. Our experiments show that this approach reaches up to 94% F-measure and 100% precision on our evaluation datasets, which are good numbers considering that the system does not require domain or target data-specific configuration.

Matching dependencies (MDs) are data profiling results that are often used for data integration, data cleaning, and entity matching. They are a generalization of functional dependencies (FDs) matching similar rather than same elements. As their discovery is very difficult, existing profiling algorithms find either only small subsets of all MDs or their scope is limited to only small datasets. We focus on the efficient discovery of all interesting MDs in real-world datasets. For this purpose, we propose HyMD, a novel MD discovery algorithm that finds all minimal, non-trivial MDs within given similarity boundaries. The algorithm extracts the exact similarity thresholds for the individual MDs from the data instead of using predefined similarity thresholds. For this reason, it is the first approach to solve the MD discovery problem in an exact and truly complete way. If needed, the algorithm can, however, enforce certain properties on the reported MDs, such as disjointness and minimum support, to focus the discovery on such results that are actually required by downstream use cases. HyMD is technically a hybrid approach that combines the two most popular dependency discovery strategies in related work: lattice traversal and inference from record pairs. Despite the additional effort of finding exact similarity thresholds for all MD candidates, the algorithm is still able to efficiently process large datasets, e.g., datasets larger than 3 GB.

Abstract Data analytics are moving beyond the limits of a single platform. In this paper, we present the cost-based optimizer of Rheem, an open-source cross-platform system that copes with these new requirements. The optimizer allocates the subtasks of data analytic tasks to the most suitable platforms. Our main contributions are: (i) a mechanism based on graph transformations to explore alternative execution strategies; (ii) a novel graph-based approach to determine efficient data movement plans among subtasks and platforms; and (iii) an efficient plan enumeration algorithm, based on a novel enumeration algebra. We extensively evaluate our optimizer under diverse real tasks. We show that our optimizer can perform tasks more than one order of magnitude faster when using multiple platforms than when using a single platform.

Inclusion dependencies are an important type of metadata in relational databases, because they indicate foreign key relationships and serve a variety of data management tasks, such as data linkage, query optimization, and data integration. The discovery of inclusion dependencies is, therefore, a well-studied problem and has been addressed by many algorithms. Each of these discovery algorithms follows its own strategy with certain strengths and weaknesses, which makes it difficult for data scientists to choose the optimal algorithm for a given profiling task.

Data analytics are moving beyond the limits of a single data processing platform. A cross-platform query optimizer is necessary to enable applications to run their tasks over multiple platforms efficiently and in a platform-agnostic manner. For the optimizer to be effective, it must consider data movement costs across different data processing platforms. In this paper, we present the graph-based data movement strategy used by Rheem, our open-source cross-platform system. In particular, we (i) model the data movement problem as a new graph problem, which we prove to be NP-hard, and (ii) propose a novel graph exploration algorithm, which allows Rheem to discover multiple hidden opportunities for cross-platform data processing.

Maintaining data consistency is known to be hard. Recent approaches have relied on integrity constraints to deal with the problem - correct and complete constraints naturally work towards data consistency. State-of-the-art data cleaning frameworks have used the formalism known as denial constraint (DC) to handle a wide range of real-world constraints. Each DC expresses a relationship between predicates that indicate which combinations of attribute values are inconsistent. The design of DCs, however, must keep pace with the complexity of data and applications. The alternative to designing DCs by hand is automatically discovering DCs from data, which is computationally expensive due to the large search space of DCs. To tackle this challenging task, we present a novel algorithm to efficiently discover DCs: DCFinder. The algorithm combines data structures called position list indexes with techniques based on predicate selectivity to efficiently validate DC candidates. Because the available data often contain errors, DCFinder is especially designed to discovering approximate DCs, i.e., DCs that may partially hold. Our experimental evaluation uses real and synthetic datasets and shows that DCFinder outperforms all the existing approximate DC discovery algorithms.

The integration of diverse structured and unstructured information sources into a unified, domain-specific knowledge base is an important task in many areas. A well-maintained knowledge base enables data analysis in complex scenarios, such as risk analysis in the financial sector or investigating large data leaks, such as the Paradise or Panama papers. Both the creation of such knowledge bases, as well as their continuous maintenance and curation involves many complex tasks and considerable manual effort.

Functional dependencies (FDs) and unique column combinations (UCCs) form a valuable ingredient for many data management tasks, such as data cleaning, schema recovery, and query optimization. Because these dependencies are unknown in most scenarios, their automatic discovery has been well researched. However, existing methods mostly discover only exact dependencies, i.e., those without violations. Real-world dependencies, in contrast, are frequently approximate due to data exceptions, ambiguities, or data errors. This relaxation to approximate dependencies renders their discovery an even harder task than the already challenging exact dependency discovery. To this end, we propose the novel and highly efficient algorithm P yro to discover both approximate FDs and approximate UCCs. P yro combines a separate-and-conquer search strategy with sampling-based guidance that quickly detects dependency candidates and verifies them. In our broad experimental evaluation, P yro outperforms existing discovery algorithms by a factor of up to 33, scales to larger datasets, and at the same time requires the least main memory.

Functional dependencies (FDs) play an important role in maintaining data quality. They can be used to enforce data consistency and to guide repairs over a database. In this work, we investigate the problem of missing values and its impact on FD discovery. When using existing FD discovery algorithms, some genuine FDs could not be detected precisely due to missing values or some non-genuine FDs can be discovered even though they are caused by missing values with a certain NULL semantics. We define a notion of genuineness and propose algorithms to compute the genuineness score of a discovered FD. This can be used to identify the genuine FDs among the set of all valid dependencies that hold on the data. We evaluate the quality of our method over various real-world and semi-synthetic datasets with extensive experiments. The results show that our method performs well for relatively large FD sets and is able to accurately capture genuine FDs.

Data and metadata in datasets experience many different kinds of change. Values are inserted, deleted or updated; rows appear and disappear; columns are added or repurposed, etc. In such a dynamic situation, users might have many questions related to changes in the dataset, for instance which parts of the data are trustworthy and which are not? Users will wonder: How many changes have there been in the recent minutes, days or years? What kind of changes were made at which points of time? How dirty is the data? Is data cleansing required? The fact that data changed can hint at different hidden processes or agendas: a frequently crowd-updated city name may be controversial; a person whose name has been recently changed may be the target of vandalism; and so on. We show various use cases that benefit from recognizing and exploring such change. We envision a system and methods to interactively explore such change, addressing the variability dimension of big data challenges. To this end, we propose a model to capture change and the process of exploring dynamic data to identify salient changes. We provide exploration primitives along with motivational examples and measures for the volatility of data. We identify technical challenges that need to be addressed to make our vision a reality, and propose directions of future work for the data management community.

Databases are one of the great success stories in IT. However, they have been continuously increasing in complexity, hampering operation, maintenance, and upgrades. To face this complexity, sophisticated methods for schema summarization, data cleaning, information integration, and many more have been devised that usually rely on data profiles, such as data statistics, signatures, and integrity constraints. Such data profiles are often extracted by automatic algorithms, which entails various problems: The profiles can be unfiltered and huge in volume; different profile types require different complex data structures; and the various profile types are not integrated with each other. We introduce Metacrate, a system to store, organize, and analyze data profiles of relational databases, thereby following the proven design of databases. In particular, we (i) propose a logical and a physical data model to store all kinds of data profiles in a scalable fashion; (ii) describe an analytics layer to query, integrate, and analyze the profiles efficiently; and (iii) implement on top a library of established algorithms to serve use cases, such as schema discovery, database refactoring, and data cleaning.

Media analysis can reveal interesting patterns in the way newspapers report the news and how these patterns evolve over time. One example pattern is the quoting choices that media make, which could be used as bias indicators. Media slant can be expressed both with the choice of reporting an event, e.g. a person's statement, but also with the words used to describe the event. Thus, automatic discovery of systematic quoting patterns in the news could illustrate to the readers the media' beliefs, such as political preferences. In this paper, we aim to discover political media bias by demonstrating systematic patterns of reporting speech in two major British newspapers. To this end, we analyze news articles from 2000 to 2015. By taking into account different kinds of bias, such as selection, coverage and framing bias, we show that the quoting patterns of newspapers are predictable.

is to understand the dataset at hand and its metadata. The process of metadata discovery is known as data profiling. Profiling activities range from ad-hoc approaches, such as eye-balling random subsets of the data or formulating aggregation queries, to systematic inference of structural information and statistics of a dataset using dedicated profiling tools. In this tutorial, we highlight the importance of data profiling as part of any data-related use-case, and we discuss the area of data profiling by classifying data profiling tasks and reviewing the state-of-the-art data profiling systems and techniques. In particular, we discuss hard problems in data profiling, such as algorithms for dependency discovery and profiling algorithms for dynamic data and streams. We also pay special attention to visualizing and interpreting the results of data profiling. We conclude with directions for future research in the area of data profiling. This tutorial is based on our survey on profiling relational data [2].

Denial constraints (DCs) are a generalization of many other integrity constraints (ICs) widely used in databases, such as key constraints, functional dependencies, or order dependencies. Therefore, they can serve as a unified reasoning framework for all of these ICs and express business rules that cannot be expressed by the more restrictive IC types. The process of formulating DCs by hand is difficult, because it requires not only domain expertise but also database knowledge, and due to DCs' inherent complexity, this process is tedious and error-prone. Hence, an automatic DC discovery is highly desirable: we search for all valid denial constraints in a given database instance. However, due to the large search space, the problem of DC discovery is computationally expensive. We propose a new algorithm H ydra , which overcomes the quadratic runtime complexity in the number of tuples of state-of-the-art DC discovery methods. The new algorithm's experimentally determined runtime grows only linearly in the number of tuples. This results in a speedup by orders of magnitude, especially for datasets with a large number of tuples. Hydra can deliver results in a matter of seconds that to date took hours to compute.

Data preparation and data profiling comprise many both basic and complex tasks to analyze a dataset at hand and extract metadata, such as data distributions, key candidates, and functional dependencies. Among the most important types of metadata is the number of distinct values in a column, also known as the zeroth-frequency moment. Cardinality estimation itself has been an active research topic in the past decades due to its many applications. The aim of this paper is to review the literature of cardinality estimation and to present a detailed experimental study of twelve algorithms, scaling far beyond the original experiments. First, we outline and classify approaches to solve the problem of cardinality estimation - we describe their main idea, error-guarantees, advantages, and disadvantages. Our experimental survey then compares the performance all twelve cardinality estimation algorithms. We evaluate the algorithms' accuracy, runtime, and memory consumption using synthetic and real-world datasets. Our results show that different algorithms excel in different in categories, and we highlight their trade-offs.

Detecting inclusion dependencies, the prerequisite of foreign keys, in relational data is a challenging task. Detecting them among the hundreds of thousands or even millions of tables on the web is daunting. Still, such inclusion dependencies can help connect disparate pieces of information on the Web and reveal unknown relationships among tables. With the algorithm M any , we present a novel inclusion dependency detection algorithm, specialized for the very many—but typically small—tables found on the Web. We make use of Bloom filters and indexed bit-vectors to show the feasibility of our approach. Our evaluation on two corpora of Web tables shows a superior runtime over known approaches and its usefulness to reveal hidden structures on the Web.

Functional dependencies (FDs) are an important prerequisite for various data management tasks, such as schema normalization, query optimization, and data cleansing. However, automatic FD discovery entails an exponentially growing search and solution space, so that even today's fastest FD discovery algorithms are limited to small datasets only, due to long runtimes and high memory consumptions.

Data proling is the discipline of examining an unknown dataset for its structure and statistical information. It is a preprocessing step in a wide range of applications, such as data integration, data cleansing, or query optimization. For this reason, many algorithms have been proposed for the discovery of dierent kinds of metadata. When analyzing a dataset, these proling algorithms are often applied in sequence, but they do not support one another, for instance, by sharing I/O cost or pruning information. We present the holistic algorithm Muds, which jointly discovers the three most important metadata: inclusion dependencies, unique column combinations, and functional dependencies. By sharing I/O cost and data structures across the dierent discovery tasks, Muds can clearly increase the eciency of traditional sequential data proling. The algorithm also introduces novel inter-task pruning rules that build upon dierent types of metadata, e.g., unique column combinations to infer functional dependencies. We evaluate Muds in detail and compare it against the sequential execution of state-of-the-art algorithms. A comprehensive evaluation shows that our holistic algorithm outperforms the baseline by up to factor 48 on datasets with favorable pruning conditions.

One of the crucial requirements before consuming datasets for any application is to understand the dataset at hand and its metadata. The process of metadata discovery is known as data profiling. Profiling activities range from ad-hoc approaches, such as eye-balling random subsets of the data or formulating aggregation queries, to systematic inference of structural information and statistics of a dataset using dedicated profiling tools. In this tutorial, we highlight the importance of data profiling as part of any data-related use-case, and discuss the area of data profiling by classifying data profiling tasks and reviewing the state-of-the-art data profiling systems and techniques. In particular, we discuss hard problems in data profiling, such as algorithms for dependency discovery and profiling algorithms for dynamic data and streams. We conclude with directions for future research in the area of data profiling. This tutorial is based on our survey on profiling relational data [1].

Inclusion dependencies (INDs) form an important integrity constraint on relational databases, supporting data management tasks, such as join path discovery and query optimization. Conditional inclusion dependencies (CINDs), which define including and included data in terms of conditions, allow to transfer these capabilities to RDF data. However, CIND discovery is computationally much more complex than IND discovery and the number of CINDs even on small RDF datasets is intractable. To cope with both problems, we first introduce the notion of pertinent CINDs with an adjustable relevance criterion to filter and rank CINDs based on their extent and implications among each other. Second, we present RDFind, a distributed system to efficiently discover all pertinent CINDs in RDF data. RDFind employs a lazy pruning strategy to drastically reduce the CIND search space. Also, its exhaustive parallelization strategy and robust data structures make it highly scalable. In our experimental evaluation, we show that RDFind is up to 419 times faster than the state-of-the-art, while considering a more general class of CINDs. Furthermore, it is capable of processing a very large dataset of billions of triples, which was entirely infeasible before.

Functional dependencies are structural metadata that can be used for schema normalization, data integration, data cleansing, and many other data management tasks. Despite their importance, the functional dependencies of a specific dataset are usually unknown and almost impossible to discover manually. For this reason, database research has proposed various algorithms for functional dependency discovery. None, however, are able to process datasets of typical real-world size, e.g., datasets with more than 50 attributes and a million records. We present a hybrid discovery algorithm called HyFD, which combines fast approximation techniques with efficient validation techniques in order to find all minimal functional dependencies in a given dataset. While operating on compact data structures, HyFD not only outperforms all existing approaches, it also scales to much larger datasets.

Data cleaning and data integration have been the topic of intensive research for at least the past thirty years, resulting in a multitude of specialized methods and integrated tool suites. All of them require at least some and in most cases significant human input in their configuration, during processing, and for evaluation. For managers (and for developers and scientists) it would be therefore of great value to be able to estimate the effort of cleaning and integrating some given data sets and to know the pitfalls of such an integration project in advance. This helps deciding about an integration project using cost/benefit analysis, budgeting a team with funds and manpower, and monitoring its progress. Further, knowledge of how well a data source fits into a given data ecosystem improves source selection. We present an extensible framework for the automatic effort estimation for mapping and cleaning activities in data integration projects with multiple sources. It comprises a set of measures and methods for estimating integration complexity and ultimately effort, taking into account heterogeneities of both schemas and instances and regarding both integration and cleaning operations. Experiments on two real-world scenarios show that our proposal is two to four times more accurate than a current approach in estimating the time duration of an integration process, and provides a meaningful breakdown of the integration problems as well as the required integration activities.

Data profiling is the discipline of discovering metadata about given datasets. The metadata itself serve a variety of use cases, such as data integration, data cleansing, or query optimization. Due to the importance of data profiling in practice, many tools have emerged that support data scientists and IT professionals in this task. These tools provide good support for profiling statistics that are easy to compute, but they are usually lacking automatic and efficient discovery of complex statistics, such as inclusion dependencies, unique column combinations, or functional dependencies. We present Metanome, an extensible profiling platform that incorporates many state-of-the-art profiling algorithms. While Metanome is able to calculate simple profiling statistics in relational data, its focus lies on the automatic discovery of complex metadata. Metanome's goal is to provide novel profiling algorithms from research, perform comparative evaluations, and to support developers in building and testing new algorithms. In addition, Metanome is able to rank profiling results according to various metrics and to visualize the, at times, large metadata sets.

Functional dependencies are important metadata used for schema normalization, data cleansing and many other tasks. The efficient discovery of functional dependencies in tables is a well-known challenge in database research and has seen several approaches. Because no comprehensive comparison between these algorithms exist at the time, it is hard to choose the best algorithm for a given dataset. In this experimental paper, we describe, evaluate, and compare the seven most cited and most important algorithms, all solving this same problem. First, we classify the algorithms into three different categories, explaining their commonalities. We then describe all algorithms with their main ideas. The descriptions provide additional details where the original papers were ambiguous or incomplete. Our evaluation of careful re-implementations of all algorithms spans a broad test space including synthetic and real-world data. We show that all functional dependency algorithms optimize for certain data characteristics and provide hints on when to choose which algorithm. In summary, however, all current approaches scale surprisingly poorly, showing potential for future research.

The discovery of all inclusion dependencies (INDs) in a dataset is an important part of any data profiling effort. Apart from the detection of foreign key relationships, INDs can help to perform data integration, query optimization, integrity checking, or schema (re-)design. However, the detection of INDs gets harder as datasets become larger in terms of number of tuples as well as attributes. To this end, we propose Binder, an IND detection system that is capable of detecting both unary and n-ary INDs. It is based on a divide & conquer approach, which allows to handle very large datasets -- an important property on the face of the ever increasing size of today's data. In contrast to most related works, we do not rely on existing database functionality nor assume that inspected datasets fit into main memory. This renders Binder an efficient and scalable competitor. Our exhaustive experimental evaluation shows the high superiority of Binder over the state-of-the-art in both unary (Spider) and n-ary (Mind) IND discovery. Binder is up to 26x faster than Spider and more than 2500x faster than Mind.

Duplicate detection is the process of identifying multiple representations of same real world entities. Today, duplicate detection methods need to process ever larger datasets in ever shorter time: maintaining the quality of a dataset becomes increasingly difficult. We present two novel, progressive duplicate detection algorithms that significantly increase the efficiency of finding duplicates if the execution time is limited: They maximize the gain of the overall process within the time available by reporting most results much earlier than traditional approaches. Comprehensive experiments show that our progressive algorithms can double the efficiency over time of traditional duplicate detection and significantly improve upon related work.

The discovery of unknown functional dependencies in a dataset is of great importance for database redesign, anomaly detection and data cleansing applications. However, as the nature of the problem is exponential in the number of attributes none of the existing approaches can be applied on large datasets. We present a new algorithm DFD for discovering all functional dependencies in a dataset following a depth-first traversal strategy of the attribute lattice that combines aggressive pruning and efficient result verification. Our approach is able to scale far beyond existing algorithms for up to 7.5 million tuples, and is up to three orders of magnitude faster than existing approaches on smaller datasets.

Duplicates in a dataset are multiple representations of the same real-world entity and constitute a major data quality problem. This paper investigates the problem of estimating the number and sizes of duplicate record clusters in advance and describes a sampling-based method for solving this problem. In extensive experiments, on multiple datasets, we show that the proposed method reliably estimates the number of duplicate clusters, while being highly efficient.

Before reaping the benefits of open data to add value to an organizations internal data, such new, external datasets must be analyzed and understood already at the basic level of data types, constraints, value patterns etc. Such data profiling, already difficult for large relational data sources, is even more challenging for RDF datasets, the preferred data model for linked open data. We present ProLod++, a novel tool for various profiling and mining tasks to understand and ultimately improve open RDF data. ProLod++ comprises various traditional data profiling tasks, adapted to the RDF data model. In addition, it features many specific profiling results for open data, such as schema discovery for user-generated attributes, association rule discovery to uncover synonymous predicates, and uniqueness discovery along ontology hierarchies. ProLod++ is highly efficient, allowing interactive profiling for users interested in exploring the properties and structure of yet unknown datasets.

The discovery of all unique (and non-unique) column combinations in an unknown dataset is at the core of any data profiling effort. Unique column combinations resemble candidate keys of a relational dataset. Several research approaches have focused on their efficient discovery in a given, static dataset. However, none of these approaches are suitable for applications on dynamic datasets, such as transactional databases, social networks, and scientific applications. In these cases, data profiling techniques should be able to efficiently discover new uniques and non-uniques (and validate old ones) after tuple inserts or deletes, without re-profiling the entire dataset. We present the first approach to efficiently discover unique and non-unique constraints on dynamic datasets that is independent of the initial dataset size. In particular, Swan makes use of intelligently chosen indices to minimize access to old data. We perform an exhaustive analysis of Swan and compare it with two state-of-the-art techniques for unique discovery: Gordian and Ducc. The results show that Swan significantly outperforms both, as well as their incremental adaptations. For inserts, Swan is more than 63x faster than Gordian and up to 50x faster than Ducc. For deletes, Swan is more than 15x faster than Gordian and up to 1 order of magnitude faster than Ducc. In fact, Swan even improves on the static case by dividing the dataset into a static part and a set of inserts.

Efficient top-k retrieval of records from a database has been an active research field for many years. We approach the problem from a real-world application point of view, in which the order of records according to some similarity function on an attribute is not unique: Many records have same values in several attributes and thus their ranking in those attributes is arbitrary. For instance, in large person databases many individuals have the same first name, the same date of birth, or live in the same city. Existing algorithms, such as the Threshold Algorithm (TA), are ill-equipped to handle such cases efficiently.

The discovery of all unique (and non-unique) column combinations in a given dataset is at the core of any data profiling effort. The results are useful for a large number of areas of data management, such as anomaly detection, data integration, data modeling, duplicate detection, indexing, and query optimization. However, discovering all unique and non-unique column combinations is an NP-hard problem, which in principle requires to verify an exponential number of column combinations for uniqueness on all data values. Thus, achieving efficiency and scalability in this context is a tremendous challenge by itself. In this paper, we devise Ducc, a scalable and efficient approach to the problem of finding all unique and non-unique column combinations in big datasets. We first model the problem as a graph coloring problem and analyze the pruning effect of individual combinations. We then present our hybrid column-based pruning technique, which traverses the lattice in a depth-first and random walk combination. This strategy allows Ducc to typically depend on the solution set size and hence to prune large swaths of the lattice. Ducc also incorporates row-based pruning to run uniqueness checks in just few milliseconds. To achieve even higher scalability, Ducc runs on several CPU cores (scale-up) and compute nodes (scale-out) with a very low overhead. We exhaustively evaluate Ducc using three datasets (two real and one synthetic) with several millions rows and hundreds of attributes. We compare Ducc with related work: Gordian and HCA. The results show that Ducc is up to more than 2 orders of magnitude faster than Gordian and HCA (631x faster than Gordian and 398x faster than HCA). Finally, a series of scalability experiments shows the efficiency of Ducc to scale up and out.

To integrate Linked Open Data, which originates from various and heterogeneous sources, the use of well-defined ontologies is essential. However, oftentimes the utilization of these ontologies by data publishers differs from the intended application envisioned by ontology engineers. This may lead to unspecified properties being used ad-hoc as predicates in RDF triples or it may result in infrequent usage of specified properties. These mismatches impede the goals and propagation of the Web of Data as data consumers face difficulties when trying to discover and integrate domain-specific information. In this work, we identify and classify common misusage patterns by employing frequency analysis and rule mining. Based on this analysis, we introduce an algorithm to propose suggestions for a data-driven ontology re-engineering workflow, which we evaluate on two large-scale RDF datasets.

Data dependencies are used to improve the quality of a database schema, to optimize queries, and to ensure consistency in a database. Conditional dependencies have been introduced to analyze and improve data quality. A conditional dependency is a dependency with a limited scope defined by conditions over one or more attributes. Only the matching part of the instance must adhere to the dependency. In this paper we focus on conditional inclusion dependencies (CINDs).We generalize the definition of CINDs, distinguishing covering and completeness conditions. We present a new use case for such CINDs showing their value for solving complex data quality tasks. Further, we propose efficient algorithms that identify covering and completeness conditions conforming to given quality thresholds. Our algorithms choose not only the condition values but also the condition attributes automatically. Finally, we show that our approach efficiently provides meaningful and helpful results for our use case.

Large amounts of graph-structured data are emerging from various avenues, ranging from natural and life sciences to social and semantic web communities. We address the problem of discovering subgraphs of entities that reflect latent topics in graph-structured data. These topics are structured meta-information providing further insights into the data. The presented approach effectively detects such topics by exploiting only the structure of the underlying graph, thus avoiding the dependency on textual labels, which are a scarce asset in prevalent graph datasets. The viability of our approach is demonstrated in experiments on real-world datasets.

Linked Data has emerged as a powerful way of interconnecting structured data on the Web. However, the cross-linkage between Linked Data sources is not as extensive as one would hope for. In this paper, we formalize the task of automatically creating "sameAs" links across data sources in a globally consistent manner. Our algorithm, presented in a multi-core as well as a distributed version, achieves this link generation by accounting for joint evidence of a match. Experiments confirm that our system scales beyond 100 million entities and delivers highly accurate results despite the vast heterogeneity and daunting scale.

Duplicate detection is the task of identifying all groups of records within a data set that represent the same real-world entity, respectively. This task is difficult, because (i) representations might differ slightly, so some similarity measure must be defined to compare pairs of records and (ii) data sets might have a high volume making a pair-wise comparison of all records infeasible. To tackle the second problem, many algorithms have been suggested that partition the data set and compare all record pairs only within each partition. One well-known such approach is the Sorted Neighborhood Method (SNM), which sorts the data according to some key and then advances a window over the data comparing only records that appear within the same window. We propose with the Duplicate Count Strategy (DCS) a variation of SNM that uses a varying window size. It is based on the intuition that there might be regions of high similarity suggesting a larger window size and regions of lower similarity suggesting a smaller window size. Next to the basic variant of DCS, we also propose and thoroughly evaluate a variant called DCS++ which is provably better than the original SNM in terms of efficiency (same results with fewer comparisons).

Duplicate detection determines different representations of real-world objects in a database. Recent research has considered the use of relationships among object representations to improve duplicate detection. In the general case where relationships form a graph, research has mainly focused on duplicate detection quality/effectiveness. Scalability has been neglected so far, even though it is crucial for large real-world duplicate detection tasks. We scale-up duplicate detection in graph data (DDG) to large amounts of data and pairwise comparisons, using the support of a relational database management system. To this end, we first present a framework that generalizes the DDG process. We then present algorithms to scale DDG in space (amount of data processed with bounded main memory) and in time. Finally, we extend our framework to allow batched and parallel DDG, thus further improving efficiency. Experiments on data of up to two orders of magnitude larger than data considered so far in DDG show that our methods achieve the goal of scaling DDG to large volumes of data.

Unique column combinations of a relational database table are sets of columns that contain only unique values. Discovering such combinations is a fundamental research problem and has many different data management and knowledge discovery applications. Existing discovery algorithms are either brute force or have a high memory load and can thus be applied only to small datasets or samples. In this paper, the well-known Gordian algorithm [9] and "Apriori-based" algorithms [4] are compared and analyzed for further optimization. We greatly improve the Apriori algorithms through efficient candidate generation and statistics-based pruning methods. A hybrid solution HCA-Gordian combines the advantages of Gordian and our new algorithm HCA, and it outperforms all previous work in many situations.

Measuring the similarity of two records is a challenging problem, but necessary for fundamental tasks, such as duplicate detection and similarity search. By exploiting frequencies of attribute values, many similarity measures can be improved: In a person table with U.S. citizens, Arnold Schwarzenegger is a very rare name. If we find several Arnold Schwarzeneggers in it, it is very likely that these are duplicates. We are then less strict when comparing other attribute values, such as birth date or address. We put this intuition to use by partitioning compared record pairs according to frequencies of attribute values. For example, we could create three partitions from our data: Partition 1 contains all pairs with rare names, Partition 2 all pairs with medium frequent names, and Partition 3 all pairs with frequent names. For each partition, we learn a different similarity measure: we apply machine learning techniques to combine a set of base similarity measures into an overall measure. To determine a good partitioning, we compare different partitioning strategies. We achieved best results with a novel algorithm inspired by genetic programming.

Given a (large) set of objects and a query, similarity search aims to find all objects similar to the query. A frequent approach is to define a set of base similarity measures for the different aspects of the objects, and to build light-weight similarity indexes on these measures. To determine the overall similarity of two objects, the results of these base measures are composed, e.g., using simple aggregates or more involved machine learning techniques. We propose the first solution to this search problem that does not place any restrictions on the similarity measures, the composition technique, or the data set size. We define the query plan optimization problem to determine the best query plan using the similarity indexes. A query plan must choose which individual indexes to access and which thresholds to apply. The plan result should be as complete as possible within some cost threshold. We propose the approximative top neighborhood algorithm, which determines a near-optimal plan while significantly reducing the amount of candidate plans to be considered. An exact version of the algorithm determines the optimal solution. Evaluation on real-world data indicates that both versions clearly outperform a complete search of the query plan space.

A large number of statistical indicators (GDP, life expectancy, income, etc.) collected over long periods of time as well as data on historical events (wars, earthquakes, elections, etc.) are published on the World Wide Web. By augmenting statistical outliers with relevant historical occurrences, we provide a means to observe (and predict) the influence and impact of events. The vast amount and size of available data sets enable the detection of recurring connections between classes of events and statistical outliers with the help of association rule mining. The results of this analysis are published at http://www.blackswanevents.org and can be explored interactively.

Graph-structured data abounds and has become the subject of much attention in the past years, for instance when searching and analyzing social network structures. Measures such as the shortest path or the number of paths between two nodes are used as proxies for similarity or relevance[1]. These approaches benefit from the fact that the measures are determined from some context node, e.g., me in a social network. With Sprint, we apply these notions to a new domain, namely ranking web search results using the link-path-structure among pages.Sprint demonstrates the feasibility and effectiveness of Searching by Path Ranks on the INTernet with two use cases: First, we re-rank intranet search results based on the position of the user's homepage on the graph. Second, as a live proof-of-concept we dynamically re-rank Wikipedia search results based on the currently viewed page: When viewing the Java software page, a search for ranks Sun Microsystems higher than the star at the center of our solar system. We evaluate the first use case with a user study. The second use case is the focus of the demonstration and allows users to actively test our system with any combination of context page and search term.

Roughly every third Wikipedia article contains an infobox - a table that displays important facts about the subject in attribute-value form. The schema of an infobox, i.e., the attributes that can be expressed for a concept, is defined by an infobox template. Often, authors do not specify all template attributes, resulting in incomplete infoboxes. With iPopulator, we introduce a system that automatically populates infoboxes of Wikipedia articles by extracting attribute values from the article's text. In contrast to prior work, iPopulator detects and exploits the structure of attribute values for independently extracting value parts. We have tested iPopulator on the entire set of infobox templates and provide a detailed analysis of its effectiveness. For instance, we achieve an average extraction precision of 91% for 1,727 distinct infobox template attributes.

The goal of data fusion is to combine several representations of one real world object into a single, consistent representation, e.g., in data integration. A very popular operator to perform data fusion is the minimum union operator. It is defined as the outer union and the subsequent removal of subsumed tuples. Minimum union is used in other applications as well, for instance in database query optimization to rewrite outer join queries, in the semantic web community in implementing SPARQL’s OPTIONAL operator, etc. Despite its wide applicability, there are only few efficient implementations, and until now, minimum union is not a relational database primitive. This paper fills this gap as we present implementations of subsumption that serve as a building block for minimum union. Furthermore, we consider this operator as database primitive and show how to perform optimization of query plans in presence of subsumption and minimum union through rule-based plan transformations. Experiments on both artificial and real world data show that our algorithms outperform existing algorithms used for subsumption in terms of runtime and they scale to large volumes of data. In the context of data integration, we observe that performing data fusion calls for more than subsumption and minimum union. Therefore, another contribution of this paper is the definition of the complementation and complement union operators. Intuitively, these allow to merge tuples that have complementing values and thus eliminate unnecessary null-values. Research was partially performed while at Hasso-Plattner-Institut. Research was partially performed while at Hasso-Plattner-Institut

Enterprise Search (ES) is different from traditional IR due to a number of reasons, among which the high level of ambiguity of terms in queries and documents and existence of graph-structured enterprise data (ontologies) that describe the concepts of interest and their relationships to each other, are the most important ones. Our method identifies concepts from the enterprise ontology in the query and corpus. We propose a ranking scheme for ontology sub-graphs on top of approximately matched token q-grams. The ranking leverages the graph-structure of the ontology to incorporate not explicitly mentioned concepts. It improves previous solutions by using a fine-grained ranking function that is specifically designed to cope with high levels of ambiguity. This method is able to capture much more of the semantics of queries and documents than previous techniques. We prove this claim by an evaluation of our method in three real-life scenarios from two different domains, and found it to consistently be superior both in terms of precision and recall.

The amount of information produced in the world increases by 30% every year and this rate will only go up. With advanced network technology, more and more sources are available either over the Internet or in enterprise intranets. Modern data management applications, such as setting up Web portals, managing enterprise data, managing community data, and sharing scientific data, often require integrating available data sources and providing a uniform interface for users to access data from different sources; such requirements have been driving fruitful research on data integration over the last two decades [11, 13].

Duplicate detection determines different representations of real-world objects in a database. Recent research has considered the use of relationships among object representations to improve duplicate detection. In the general case where relationships form a graph, research has mainly focused on duplicate detection quality/effectiveness. Scalability has been neglected so far, even though it is crucial for large real-world duplicate detection tasks.

Duplicate detection is the process of identifying multiple representations of a same real-world object in a data source. Duplicate detection is a problem of critical importance in many applications, including customer relationship management, personal information management, or data mining. In this paper, we present how a research prototype, namely DogmatiX, which was designed to detect duplicates in hierarchical XML data, was successfully extended and applied on a large scale industrial relational database in cooperation with Schufa Holding AG. Schufa's main business line is to store and retrieve credit histories of over 60 million individuals. Here, correctly identifying duplicates is critical both for individuals and companies: On the one hand, an incorrectly identified duplicate potentially results in a false negative credit history for an individual, who will then not be granted credit anymore. On the other hand, it is essential for companies that Schufa detects duplicates of a person that deliberately tries to create a new identity in the database in order to have a clean credit history. Besides the quality of duplicate detection, i.e., its effectiveness, scalability cannot be neglected, because of the considerable size of the database. We describe our solution to coping with both problems and present a comprehensive evaluation based on large volumes of real-world data.

Data sources for data integration often come with spurious schema definitions such as undefined foreign key constraints. Such metadata are important for querying the database and for database integration. We present our algorithm SPIDER (single pass inclusion dependency recognition) for detecting inclusion dependencies, as these are the automatically testable part of a foreign key constraint. For IND detection all pairs of attributes must be tested. SPIDER solves this task very efficiently by testing all attribute pairs in parallel. It analyzes a 2 GB database in ~ 20 min and a 21 GB database in ~ 4 h.

Recent work both in the relational and the XML world have shown that the efficacy and efficiency of duplicate detection is enhanced by regarding relationships between entities. However, most approaches for XML data rely on 1:n parent/child relationships, and do not apply to XML data that represents m:n relationships. We present a novel comparison strategy, which performs duplicate detection effectively for all kinds of parent/child relationships, given dependencies between different XML elements. Due to cyclic dependencies, it is possible that a pairwise classification is performed more than once, which compromises efficiency. We propose an order that reduces the number of such reclassifications and apply it to two algorithms. The first algorithm performs reclassifications, and efficiency is increased by using the order reducing the number of reclassifications. The second algorithm does not perform a comparison more than once, and the order is used to miss few reclassifications and hence few potential duplicates.

Most data integration applications require a matching between the schemas of the respective data sets. We show how the existence of duplicates within these data sets can be exploited to automatically identify matching attributes. We describe an algorithm that first discovers duplicates among data sets with unaligned schemas and then uses these duplicates to perform schema matching between schemas with opaque column names. Discovering duplicates among data sets with unaligned schemas is more difficult than in the usual setting, because it is not clear which fields in one object should be compared with which fields in the other. We have developed a new algorithm that efficiently finds the most likely duplicates in such a setting. Now, our schema matching algorithm is able to identify corresponding attributes by comparing data values within those duplicate records. An experimental study on real-world data shows the effectiveness of this approach.

Duplicate detection is the problem of detecting different entries in a data source representing the same real-world entity. While research abounds in the realm of duplicate detection in relational data, there is yet little work for duplicates in other, more complex data models, such as XML. In this paper, we present a generalized framework for duplicate detection, dividing the problem into three components: candidate definition defining which objects are to be compared, duplicate definition defining when two duplicate candidates are in fact duplicates, and duplicate detection specifying how to efficiently find those duplicates.Using this framework, we propose an XML duplicate detection method, DogmatiX, which compares XML elements based not only on their direct data values, but also on the similarity of their parents, children, structure, etc. We propose heuristics to determine which of these to choose, as well as a similarity measure specifically geared towards the XML data model. An evaluation of our algorithm using several heuristics validates our approach.

The XML wrapper is a new feature of the federated database capabilities of DB2/UDB v8. It enables users and applications to issue SQL queries against XML data from a variety of sources, including files and Web services. The XML wrapper assumes hierarchical XML documents modeled as families of virtual relational tables in a federated schema, which can then be queried to extract information from the XML and combine it with data from other sources. Due to the nature of the problem, using the XML wrapper is complex and several difficult steps must be undertaken: (i) The hierarchical schema of the source must be flattened to a relational form, (ii) Each relation of the flattened schema must be registered in DB2 as a NICKNAME - a complex virtual table definition containing several XPaths as specialized options. (iii) Each NICKNAME must be accompanied by a VIEW - again a complex structure involving join conditions. Chocolate is a tool that alleviates all three tasks: Chocolate provides several flattening strategies and an interface allowing users to modify the automatically generated target schema. Once the user is satisfied with the schema, Chocolate automatically generates the corresponding NICKNAME and VIEW definitions.

Merging and coalescing data from multiple and diverse sources into different data formats continues to be an important problem in modern information systems. Schema matching (the process of matching elements of a source schema with elements of a target schema) and schema mapping (the process of creating a query that maps between two disparate schemas) are at the heart of data integration systems. We demonstrate Clio, a semi-automatic schema mapping tool developed at the IBM Almaden Research Center. In this paper, we showcase Clio's mapping engine which allows mapping to and from relational and XML schemas, and takes advantage of data constraints in order to preserve data associations.

The basis of many systems that integrate data from multiple sources is a set of correspondences between source schemata and a target schema. Correspondences express a relationship between sets of source attributes, possibly from multiple sources, and a set of target attributes. Clio is an integration tool that assists users in defining value correspondences between attributes. In real life scenarios there may be many sources and the source relations may have many attributes. Users can get lost and might miss or be unable to find some correspondences. Also, in many real life schemata the attribute names reveal little or nothing about the semantics of the data values. Only the data values in the attribute columns can convey the semantic meaning of the attribute. Our work relieves users of the problems of too many attributes and meaningless attribute names, by automatically suggesting correspondences between source and target attributes. For each attribute, we analyze the data values and derive a set of features.

Query planning for information integration using a local-as-view approach is exponential in the size of the user query. Furthermore, it may generate an exponential number of plans, many of which will produce results of very poor quality. We propose to use information quality reasoning to speed up query planning. We construct tight upper quality bounds for a branch & bound algorithm. The algorithm uses these quality scores to filter out non-promising plans early on. Experiments show that this approach dramatically improves planning time without compromising the quality of the result. These keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.