Association rules are an important technique for gaining insights over large relational datasets consisting of tuples of elements (i.e. attribute-value pairs). However, it is difficult to explain the relative importance of data elements with respect to the rules in which they appear. This paper develops a measure of an element's contribution to a set of association rules based on Shapley values, denoted SHARQ (ShApley Rules Quantification). As is the case with many Shapely-based computations, the cost of a naive calculation of the score is exponential in the number of elements. To that end, we present an efficient framework for computing the exact SHARQ value of a single element whose running time is practically linear in the number of rules. Going one step further, we develop an efficient multi-element SHARQ algorithm which amortizes the cost of the single element SHARQ calculation over a set of elements. Based on the definition of SHARQ for elements we describe two additional use-cases for association rules explainability: rule importance and attribute importance. Extensive experiments over a novel benchmark dataset containing 67 instances of mined rule sets show the effectiveness of our approach.

Explaining the results of clustering pipelines by unraveling the characteristics of each cluster is a challenging task, often addressed manually through visualizations and queries. Existing solutions from the domain of Explainable Artificial Intelligence (XAI) are largely ineffective for cluster explanations, and interpretable-by-design clustering algorithms may be unsuitable when the clustering algorithm does not fit the data properties. To bridge this gap, we introduce Cluster-Explorer, a novel explainability tool for black-box clustering pipelines. Our approach formulates the explanation of clusters as the identification of concise conjunctions of predicates that maximize the coverage of the cluster's data points while minimizing separation from other clusters. We achieve this by reducing the problem to generalized frequent-itemsets mining (gFIM), where items correspond to explanation predicates, and itemset frequency indicates coverage. To enhance efficiency, we leverage inherent problem properties and implement attribute selection to further reduce computational costs. Experimental evaluations on a benchmark collection of 98 clustering results demonstrate the superiority of Cluster-Explorer in both explanation quality and execution times compared to XAI baselines.

Interfaces that rely on the Python programming language have become a popular tool for data analysis and exploration. In particular, the Pandas library allows users to query, manipulate, and visualize data in an easy and intuitive manner. However, users who perform such manipulations over the data in the exploratory process may struggle to justify their results, or understand which part (if any) of the obtained results is interesting and why. To handle such scenarios we developed PD-Explain, a Python library that adapts multiple prevalent query explanation approaches from the literature, and makes them accessible to Pandas users. PD-Explain is seamlessly integrated with Pandas and contains explanation functions that users can employ to choose the explanation approach they wish to use along with the necessary parameters in order to get the explanation in the suitable form. PD-Explain further allows users to automatically detect the interesting parts of a query result and get a visualization of the explanation accompanied by a Natural Language description. Our demonstration will include four different types of query result explanations and three real-world datasets with appropriate analysis tasks that will highlight the intuitive nature and usefulness of PD-Explain in data exploration tasks.

We demonstrate EFFECTS, an automated system for explorable and explainable feature extraction for multivariate time series classification. EFFECTS has a twofold contribution: (1) It significantly facilitates the exploration of MTSC data, and (2) it generates informative yet intuitive and explainable features to be used by the classification model. EFFECTS first mines the MTS data and extracts a set of interpretable features using an optimized transform-slice-aggregate process. To evaluate the quality of EFFECTS features, we gauge how well each feature distinguishes between every two classes, and how well they characterize each single class. Users can then explore the MTS data via the EFFECTS Explorer, which facilitates the visual inspection of important features, dimensions, and time slices. Last, the user can use the top features for each class when building a classification pipeline.

Interpreting clustering results is a challenging, manual task, that often requires the user to perform additional analytical queries and visualizations. To this end, we demonstrate Cluster-Explorer, an interactive, easy-to-use framework that provides explanations for black-box clustering results. Cluster-Explorer takes as input the raw dataset alongside cluster labels, and automatically generates multiple coherent explanations that characterize each cluster.

Data scientists frequently examine the raw content of large tables when exploring an unknown dataset. In such cases, small subsets of the full tables (sub-tables) that accurately capture table contents are useful. We present a framework which, given a large data table T, creates a sub-table of small, fixed dimensions by selecting a subset of T’s rows and projecting them over a subset of T’s columns. The question is: Which rows and columns should be selected to yield an informative sub-table?Our first contribution is an informativeness metric for sub-tables with two complementary dimensions: cell coverage, which measures how well the sub-table captures prominent data patterns in T, and diversity. We use association rules as the patterns captured by sub-tables, and show that computing optimal sub-tables directly using this metric is infeasible. We then develop an efficient algorithm that indirectly accounts for association rules using table embedding. The resulting framework produces sub-tables for the full table as well as for the results of queries over the table, enabling the user to quickly understand results and determine subsequent queries. Experimental results show that high-quality sub-tables can be efficiently computed, and verify the soundness of our metrics as well as the usefulness of selected sub-tables through user studies.

Automated machine learning (AutoML) frameworks are gaining popularity among data scientists as they dramatically reduce the manual work devoted to the construction of ML pipelines while obtaining similar and sometimes even better results than manually-built models. Such frameworks intelligently search among millions of possible ML pipeline configurations to finally retrieve an optimal pipeline in terms of predictive accuracy. However, when the training dataset is large, the construction and evaluation of a single ML pipeline take longer, which makes the overall AutoML running times increasingly high.

We demonstrate SubTab, a framework for creating small, informative sub-tables of large data tables to speed up data exploration. Given a table with n rows and m columns where n and m are large, SubTab creates a sub-table T_sub with k<n rows and l

When exploring a new dataset, Data Scientists often apply analysis queries, look for insights in the resulting dataframe, and repeat to apply further queries. We propose in this paper a novel solution that assists data scientists in this laborious process. In a nutshell, our solution pinpoints the most interesting (sets of) rows in each obtained dataframe. Uniquely, our definition of interest is based on the contribution of each row to the interestingness of different columns of the entire dataframe, which, in turn, is defined using standard measures such as diversity and exceptionality. Intuitively, interesting rows are ones that explain why (some column of) the analysis query result is interesting as a whole. Rows are correlated in their contribution and so the interesting score for a set of rows may not be directly computed based on that of individual rows. We address the resulting computational challenge by restricting attention to semantically-related sets, based on multiple notions of semantic relatedness; these sets serve as more informative explanations. Our experimental study across multiple real-world datasets shows the usefulness of our system in various scenarios.

Automated machine learning (AutoML) frameworks have become important tools in the data scientist's arsenal, as they dramatically reduce the manual work devoted to the construction of ML pipelines. Such frameworks intelligently search among millions of possible ML pipelines - typically containing feature engineering, model selection, and hyper parameters tuning steps - and finally output an optimal pipeline in terms of predictive accuracy. However, when the dataset is large, each individual configuration takes longer to execute, therefore the overall AutoML running times become increasingly high. To this end, we present SubStrat, an AutoML optimization strategy that tackles the data size, rather than configuration space. It wraps existing AutoML tools, and instead of executing them directly on the entire dataset, SubStrat uses a genetic-based algorithm to find a small yet representative data subset that preserves a particular characteristic of the full data. It then employs the AutoML tool on the small subset, and finally, it refines the resulting pipeline by executing a restricted, much shorter, AutoML process on the large dataset. Our experimental results, performed on three popular AutoML frameworks, Auto-Sklearn, TPOT, and H2O show that SubStrat reduces their running times by 76.3% (on average), with only a 4.15% average decrease in the accuracy of the resulting ML pipeline.

Exploratory Data Analysis (EDA) is an essential yet highly demanding task. To get a head start before exploring a new dataset, data scientists often prefer to view existing EDA notebooks -- illustrative, curated exploratory sessions, on the same dataset, that were created by fellow data scientists who shared them online. Unfortunately, such notebooks are not always available (e.g., if the dataset is new or confidential). To address this, we present ATENA, a system that takes an input dataset and auto-generates a compelling exploratory session, presented in an EDA notebook. We shape EDA into a control problem, and devise a novel Deep Reinforcement Learning (DRL) architecture to effectively optimize the notebook generation. Though ATENA uses a limited set of EDA operations, our experiments show that it generates useful EDA notebooks, allowing users to gain actual insights.

Exploratory Data Analysis (EDA) is an important initial step for any knowledge discovery process, in which data scientists interactively explore unfamiliar datasets by issuing a sequence of analysis operations (e.g. filter, aggregation, and visualization). Since EDA is long known as a difficult task, requiring profound analytical skills, experience, and domain knowledge, a plethora of systems have been devised over the last decade in order to facilitate EDA.

Exploratory Data Analysis (EDA) is an essential yet highly demanding task. To get a head start before exploring a new dataset, data scientists often prefer to view existing EDA notebooks - illustrative exploratory sessions that were created by fellow data scientists who examined the same dataset and shared their notebooks via online platforms. Unfortunately, creating an illustrative, well-documented notebook is cumbersome and time-consuming, therefore users sometimes share their notebook without explaining their exploratory steps and their results. Such notebooks are difficult to follow and to understand. To address this, we present ExplainED, a system that automatically attaches explanations to views in EDA notebooks. ExplainED analyzes each view in order to detect what elements thereof are particularly interesting, and produces a corresponding textual explanation. The explanations are generated by first evaluating the interestingness of the given view using several measures capturing different interestingness facets, then computing the Shapely values of the elements in the view, w.r.t. the interestingness measure yielding the highest score. These Shapely values are then used to guide the generation of the textual explanation. We demonstrate the usefulness of the explanations generated by ExplainED on real-life, undocumented EDA notebooks.

In applications with large userbases such as crowdsourcing, social networks or recommender systems, selecting users is a common and challenging task. Different applications require different policies for selecting users, and implementing such policies is applicationspecific and laborious. To this end, we introduce a novel declarative framework that abstracts common components of the user selection problem, while allowing for domain-specific tuning. The framework is based on an ontology view of user profiles, with respect to which we define a query language for policy specification. Our language extends SPARQL with means for capturing soft constraints which are essential for worker selection. At the core of our query engine is then a novel efficient algorithm for handling these constraints. Our experimental study on real-life data indicates the effectiveness and flexibility of our approach, showing in particular that it outperforms existing task-specific solutions in prominent user selection tasks.

Exploratory Data Analysis (EDA), is an important yet challenging task, that requires profound analytical skills and familiarity with the data domain. While Deep Reinforcement Learning (DRL) is nowadays used to solve AI challenges previously considered to be intractable, to our knowledge such solutions have not yet been applied to EDA.

As more and more social network users interact through Internet Memes, an emerging popular type of captioned images, there is a growing need for users to quickly retrieve the right Meme for a given situation. As opposed conventional image search, visually similar Memes may reflect different concepts. Intent is sometimes captured by user annotations (e.g., tags), but these are often incomplete and ambiguous. Thus, a deeper analysis of the relations among Memes is required for an accurate, custom search. To address this problem, we present SimMeme, a Meme-dedicated search engine. SimMeme uses a generic graph-based data model that aligns various types of information about the Memes with a semantic ontology. A novel similarity measure that effectively considers all incorporated data is employed and serves as the foundation of our system. Our experimental results achieve using common evaluation metrics and crowd feedback, over a large repository of real-life annotated Memes, show that in the task of Meme retrieval, SimMeme outperforms state-of-the-art solutions for image retrieval.

Modern Interactive Data Analysis (IDA) platforms, such as Kibana, Splunk, and Tableau, are gradually replacing traditional OLAP/SQL tools, as they allow for easy-to-use data exploration, visualization, and mining, even for users lacking SQL and programming skills. Nevertheless, data analysis is still a di cult task, especially for non-expert users. To that end we present REACT, a recommender system designed for modern IDA platforms. In these platforms, analysis sessions interweave high-level actions of multiple types and operate over diverse datasets . REACT identifies and generalizes relevant (previous) sessions to generate personalized next-action suggestions to the user.

Data analysis may be a difficult task, especially for non-expert users, as it requires deep understanding of the investigated domain and the particular context. In this demo we present REACT, a system that hooks to the analysis UI and provides the users with personalized recommendations of analysis actions. By matching the current user session to previous sessions of analysts working with the same or other data sets, REACT is able to identify the potentially best next analysis actions in the given user context. Unlike previous work that mainly focused on individual components of the analysis work, REACT provides a holistic approach that captures a wider range of analysis action types by utilizing novel notions of similarity in terms of the individual actions, the analyzed data and the entire analysis workflow.

Adequate crowd selection is an important factor in the success of crowdsourcing platforms, increasing the quality and relevance of crowd answers and their performance in different tasks. The optimal crowd selection can greatly vary depending on properties of the crowd and of the task. To this end, we present December, a declarative platform with novel capabilities for flexible crowd selection. December supports the personalized selection of crowd members via a dedicated query language Member-QL. This language enables specifying and combining common crowd selection criteria such as properties of a crowd member's profile and history, similarity between profiles in specific aspects and relevance of the member to a given task. This holistic, customizable approach differs from previous work that has mostly focused on dedicated algorithms for crowd selection in specific settings. To allow efficient query execution, we implement novel algorithms in December based on our generic, semantically-aware definitions of crowd member similarity and expertise. We demonstrate the effectiveness of December and Member-QL by using the VLDB community as crowd members and allowing conference participants to choose from among these members for different purposes and in different contexts.

Crowd data sourcing is increasingly used to gather information from the crowd and to obtain recommendations. In this paper, we explore a novel approach that broadens crowd data sourcing by enabling users to pose general questions, to mine the crowd for potentially relevant data, and to receive concise, relevant answers that represent frequent, significant data patterns. Our approach is based on (1) a simple generic model that captures both ontological knowledge as well as the individual history or habits of crowd members from which frequent patterns are mined; (2) a query language in which users can declaratively specify their information needs and the data patterns of interest; (3) an efficient query evaluation algorithm, which enables mining semantically concise answers while minimizing the number of questions posed to the crowd; and (4) an implementation of these ideas that mines the crowd through an interactive user interface. Experimental results with both real-life crowd and synthetic data demonstrate the feasibility and effectiveness of the approach.

We present OASSIS (for Ontology ASSISted crowd mining), a prototype system which allows users to declaratively specify their information needs, and mines the crowd for answers. The answers that the system computes are concise and relevant , and represent frequent, significant data patterns. The system is based on (1) a generic model that captures both ontological knowledge, as well as the individual knowledge of crowd members from which frequent patterns are mined; (2) a query language in which users can specify their information needs and types of data patterns they seek; and (3) an efficient query evaluation algorithm, for mining semantically concise answers while minimizing the number of questions posed to the crowd. We will demonstrate OASSIS using a couple of real-life scenarios, showing how users can formulate and execute queries through the OASSIS UI and how the relevant data is mined from the crowd.