Cláudia Antunes

Constraint relaxations for discovering unknown sequential patterns

The main drawbacks of sequential pattern mining have been its lack of focus on user expectations and the high number of discovered patterns. However, the solution commonly accepted – the use of constraints – approximates the mining process to a verification of what are the frequent patterns among the specified ones, instead of the discovery of unknown and unexpected patterns. In this paper, we propose a new methodology to mine sequential patterns, keeping the focus on user expectations, without compromising the discovery of unknown patterns. Our methodology is based on the use of constraint relaxations, and it consists on using them to filter accepted patterns during the mining process. We propose a hierarchy of relaxations, applied to constraints expressed as context-free languages, classifying the existing relaxations (legal, valid and naive, previously proposed), and proposing several new classes of relaxations. The new classes range from the approx and non-accepted, to the composition of different types of relaxations, like the approx-legal or the non-prefix-valid relaxations. Finally, we present a case study that shows the results achieved with the application of this methodology to the analysis of the curricular sequences of computer science students.

On the Need of New Methods to Mine Electrodermal Activity in Emotion-Centered Studies

Monitoring the electrodermal activity is increasingly accomplished in agent-based experimental settings as the skin is believed to be the only organ to react only to the sympathetic nervous system. This physiological signal has the potential to reveal paths that lead to excitement, attention, arousal and anxiety. However, electrodermal analysis has been driven by simple feature-extraction, instead of using expressive models that consider a more flexible behavior of the signal for improved emotion recognition. This paper proposes a novel approach centered on sequential patterns to classify the signal into a set of key emotional states. The approach combines SAX for pre-processing the signal and hidden Markov models. This approach was tested over a collected sample of signals using Affectiva-QSensor. An extensive human-to-human and human-to-robot experimental setting is under development for further validation and characterization of emotion-centered patterns.

Pattern mining on stars with FP-growth

Most existing data mining (DM) approaches look for patterns in a single table. Multi-relational DM approaches, on the other hand, look for patterns that involve multiple tables. In recent years, the most common DM techniques have been extended to the multirelational case, but there are few dedicated to star schemas. These schemas are composed of a central fact table, linking a set of dimension tables, and joining all the tables before mining may not be a feasible solution. This work proposes a method for frequent pattern mining in a star schema based on FP-Growth. It does not materialize the entire join between the tables. Instead, it constructs an FP-Tree for each dimension and then combines them to form a super FP-Tree, that will serve as input to FP-Growth.

Multi-relational pattern mining over data streams

The data storage paradigm has changed in the last decade, from operational databases to data repositories that make easier to analyze data and mining information. Among those, the primary multidimensional model represents data through star schemas, where each relation denotes an event involving a set of dimensions or business perspectives. Mining data modeled as a star schema presents two major challenges, namely: mining extremely large amounts of data and dealing with several data tables at the same time. In this paper, we describe an algorithm—Star FP Stream, in detail. This algorithm aims for finding the set of frequent patterns in a large star schema, mining directly the data, in their original structure, and exploring the most efficient techniques for mining data streams. Experiments were conducted over two star schemas, in the healthcare and sales domains.

Accessing Emotion Patterns from Affective Interactions Using Electrodermal Activity

Measuring evocative emotions in affective interactions has become a critical step for effective engagements with computers. Electro dermal activity is believed to accurately isolate sympathetic responses, revealing paths to excitement, attention and arousal, and to differentiate emotional states. However, the inability to deal with varying amplitude and length of responses across individuals has led to its use as a simple intensity barometer. Thus, two questions remain unanswered. To which extent can electro dermal activity be used to recognize emotions? How do electro dermal responses vary between human-to-human and human-to-robot interactions? To answer these questions, we propose a new method to mine the signal that surpasses the referred limitations, and conduct an extensive experiment to study the responses to emotion-evocative stimuli across different settings. Observations reveal emerging electro dermal patterns for each emotion and attractive accuracy levels for emotion recognition that increases when there is a link to the psychological traits of the subjects.

Methods for the efficient discovery of large item-indexable sequential patterns

An increasingly relevant set of tasks, such as the discovery of biclusters with order-preserving properties, can be mapped as a sequential pattern mining problem on data with item-indexable properties. An item-indexable database, typically observed in biomedical domains, does not allow item repetitions per sequence and is commonly dense. Although multiple methods have been proposed for the efficient discovery of sequential patterns, their performance rapidly degrades over item-indexable databases. The target tasks for these databases benefit from lengthy patterns and tolerate local mismatches. However, existing methods that consider noise relaxations to increase the average short length of sequential patterns scale poorly, aggravating the yet critical efficiency. In this work, we first propose a new sequential pattern mining method, IndexSpan, which is able to mine sequential patterns over item-indexable databases with heightened efficiency. Second, we propose a pattern-merging procedure, MergeIndexBic, to efficiently discover lengthy noise-tolerant sequential patterns. The superior performance of IndexSpan and MergeIndexBic against competitive alternatives is demonstrated on both synthetic and real datasets.

Inference of Sequential Association Rules Guided by Context-Free Grammars

One of the main unresolved problems in data mining is related with the treatment of data that is inherently sequential. Algorithms for the inference of association rules that manipulate sequential data have been proposed and used to some extent but are ineffective, in some cases, because too many candidate rules are extracted and filtering the relevant ones is difficult and inefficient. In this work, we present a method and algorithm for the inference of sequential association rules that uses context-free grammars to guide the discovery process, in order to filter, in an efficient and effective way, the associations discovered by the algorithm.

Constrained pattern mining in the new era

Twenty years of research on frequent itemset mining, or pattern mining, has led to the existence of a set of efficient algorithms for identifying different types of patterns, from transactional to sequential. Despite the great advances in this field, big data brought a completely new context to operate, with new challenges arising from the growth in data size, dynamics and complexity. These challenges include the shift not only from static to dynamic data, but also from tabular to complex data sources, such as social networks (expressed as graphs) and data warehouses (expressed as multi-relational models). In this new context, and more than ever, users need effective ways to control the large number of discovered patterns, and to be able to choose what patterns to consider at each time. The most accepted and common approach to minimize these drawbacks has been to capture and represent the semantics of the domain through constraints, and use them not only to reduce the number of results, but also to focus the algorithms in areas where it is more likely to gain information and return more interesting results. The use of constraints in pattern mining has been widely studied, and there are a lot of proposed types of constraints and pushing strategies. In this paper, we present a new global view of the work done on the incorporation of constraints in the pattern mining process. In particular, we propose a new framework for constrained pattern mining, that allows us to organize and analyze existing algorithms and strategies, based on the different types and properties of constraints, and on the data sources they are able to handle.

Generative modeling of repositories of health records for predictive tasks

Repositories of health records are collections of events with varying number and sparsity of occurrences within and among patients. Although a large number of predictive models have been proposed in the last decade, they are not yet able to simultaneously capture cross-attribute and temporal dependencies associated with these repositories. Two major streams of predictive models can be found. On one hand, deterministic models rely on compact subsets of discriminative events to anticipate medical conditions. On the other hand, generative models offer a more complete and noise-tolerant view based on the likelihood of the testing arrangements of events to discriminate a particular outcome. However, despite the relevance of generative predictive models, they are not easily extensible to deal with complex grids of events. In this work, we rely on the Markov assumption to propose new predictive models able to deal with cross-attribute and temporal dependencies. Experimental results hold evidence for the utility and superior accuracy of generative models to anticipate health conditions, such as the need for surgeries. Additionally, we show that the proposed generative models are able to decode temporal patterns of interest (from the learned lattices) with acceptable completeness and precision levels, and with superior efficiency for voluminous repositories.

Pattern Mining over Star Schemas in the Onto4AR Framework

Storing data according to the multidimensional model, in particular following star schemas, has demonstrated to be one of the most adequate forms to ease the exploration of data. However, this exploration has been limited to be query-based, leaving the discovery of hidden information to a second plan. The main reason for this, relates to the inability of traditional mining techniques to deal with several data tables at the same time. In this paper, we propose a new approach to mine patterns among data stored as a star schema, based in a domain driven framework, where available knowledge is represented in a domain ontology. Pattern mining is performed by an apriori-based algorithm - the D2Apriori, but more efficient algorithms are being implemented and tested, in order to solve performance issues related with the large amount of data stored in data warehouses.