The advent of Massive Open Online Courses (MOOCs) allows massive volume of registrants to enrol in these MOOCs. This research aims to offer MOOCs registrants with automatic content related feedback to fulfil their cognitive needs. A framework is proposed which consists of three modules which are the subject ontology learning module, the short text classification module, and the question answering module. Unlike previous research, to identify relevant concepts for ontology learning a regular expression parser approach is used. Also, the relevant concepts are extracted from unstructured documents. To build the concept hierarchy, a frequent pattern mining approach is used which is guided by a heuristic function to ensure that sibling concepts are at the same level in the hierarchy. As this process does not require specific lexical or syntactic information, it can be applied to any subject. To validate the approach, the resulting ontology is used in a question-answering system which analyses students' content-related questions and generates answers for them. Textbook end of chapter questions/answers are used to validate the question-answering system. The resulting ontology is compared vs. the use of Text2Onto for the question-answering system, and it achieved favourable results. Finally, different indexing approaches based on a subject's ontology are investigated when classifying short text in MOOCs forum discussion data; the investigated indexing approaches are: unigram-based, concept-based and hierarchical concept indexing. The experimental results show that the ontology-based feature indexing approaches outperform the unigram-based indexing approach. Experiments are done in binary classification and multiple labels classification settings . The results are consistent and show that hierarchical concept indexing outperforms both concept-based and unigram-based indexing. The BAGGING and random forests classifiers achieved the best result among the tested classifiers.

La conception des systèmes multimédia embarqués présente de nombreux défis comme la croissante complexité du logiciel et du matériel sous-jacent, ou les pressions liées aux délais de mise en marche. L'optimisation du processus de débogage et validation du logiciel peut aider à réduire sensiblement le temps de développement. Parmi les outils de débogage de systèmes embarqués, un puissant outil largement utilise est l'analyse de traces d'exécution. Cependant, l'évolution des techniques de traçage dans les systèmes embarqués se traduit par des traces d'exécution avec une grande quantité d'information, à tel point que leur analyse manuelle devient ingérable. Dans ce cas, les techniques de recherche de motifs peuvent aider en trouvant des motifs intéressants dans de grandes quantités d'information. Concrètement, dans cette thèse, nous nous intéressons à la découverte de comportements périodiques sur des applications multimédia. Donc, les contributions de cette thèse concernent l'analyse des traces d'exécution d'applications multimédia en utilisant des techniques de recherche de motifs périodiques fréquents. Concernant la recherche de motifs périodiques, nous proposons une définition de motif périodique adaptée aux caractéristiques de la programmation parallèle. Nous proposons ensuite une représentation condensée de l'ensemble de motifs périodiques fréquents, appelée Core Periodic Concepts (CPC), en adoptant une approche basée sur les relations triadiques. De plus, nous définissons quelques propriétés de connexion entre ces motifs, ce qui nous permet de mettre en oeuvre un algorithme efficace de recherche de CPC, appelé PerMiner. Pour montrer l'efficacité et le passage à l'échelle de PerMiner, nous réalisons une analyse rigoureuse qui montre que PerMiner est au moins deux ordres de grandeur plus rapide que l'état de l'art. En plus, nous réalisons un analyse de l'efficacité de PerMiner sur une trace d'exécution d'une application multimédia réelle en présentant l'accélération accompli par la version parallèle de l'algorithme. Concernant les systèmes embarqués, nous proposons un premier pas vers une méthodologie qui explique comment utiliser notre approche dans l'analyse de traces d'exécution d'applications multimédia. Avant d'appliquer la recherche de motifs fréquents, les traces d'exécution doivent être traitées, et pour cela nous proposons plusieurs techniques de pré-traitement des traces. En plus, pour le post-traitement des motifs périodiques, nous proposons deux outils : un outil qui trouve des pairs de motifs en compétition ; et un outil de visualisation de CPC, appelé CPCViewer. Finalement, nous montrons que notre approche peut aider dans le débogage des applications multimédia à travers deux études de cas sur des traces d'exécution d'applications multimédia réelles. / Increasing complexity in both the software and the underlying hardware, and ever tighter time-to-market pressures are some of the key challenges faced when designing multimedia embedded systems. Optimizing software debugging and validation phases can help to reduce development time significantly. A powerful tool used extensively when debugging embedded systems is the analysis of execution traces. However, evolution in embedded system tracing techniques leads to execution traces with a huge amount of information, making manual trace analysis unmanageable. In such situations, pattern mining techniques can help by automatically discovering interesting patterns in large amounts of data. Concretely, in this thesis, we are interested in discovering periodic behaviors in multimedia applications. Therefore, the contributions of this thesis are focused on the definition of periodic pattern mining techniques for the analysis of multimedia applications execution traces. Regarding periodic pattern mining contributions, we propose a definition of periodic pattern adapted to the characteristics of concurrent software. We then propose a condensed representation of the set of frequent periodic patterns, called core periodic concepts (CPC), by adopting an approach originated in triadic concept approach. Moreover, we define certain connectivity properties of these patterns that allow us to implement an efficient CPC mining algorithm, called PerMiner. Then, we perform a thorough analysis to show the efficiency and scalability of PerMiner algorithm. We show that PerMiner algorithm is at least two orders of magnitude faster than the state of the art. Moreover, we evaluate the efficiency of PerMiner algorithm over a real multimedia application trace and also present the speedup achieved by a parallel version of the algorithm. Then, regarding embedded systems contributions, we propose a first step towards a methodology which aims at giving the first guidelines of how to use our approach in the analysis of multimedia applications execution traces. Besides, we propose several ways of preprocessing execution traces and a competitors finder tool to postprocess the mining results. Moreover, we present a CPC visualization tool, called CPCViewer, that facilitates the analysis of a set of CPCs. Finally, we show that our approach can help in debugging multimedia applications through the study of two use cases over real multimedia application execution traces.

Les radiologues utilisent au quotidien des solutions d'imagerie médicale pour le diagnostic. L'amélioration de l'expérience utilisateur est toujours un axe majeur de l'effort continu visant à améliorer la qualité globale et l'ergonomie des produits logiciels. Les applications de monitoring permettent en particulier d'enregistrer les actions successives effectuées par les utilisateurs dans l'interface du logiciel. Ces interactions peuvent être représentées sous forme de séquences d'actions. Sur la base de ces données, ce travail traite de deux sujets industriels : les pannes logicielles et l'ergonomie des logiciels. Ces deux thèmes impliquent d'une part la compréhension des modes d'utilisation, et d'autre part le développement d'outils de prédiction permettant soit d'anticiper les pannes, soit d'adapter dynamiquement l'interface logicielle en fonction des besoins des utilisateurs. Tout d'abord, nous visons à identifier les origines des crashes du logiciel qui sont essentielles afin de pouvoir les corriger. Pour ce faire, nous proposons d'utiliser un test binomial afin de déterminer quel type de pattern est le plus approprié pour représenter les signatures de crash. L'amélioration de l'expérience utilisateur par la personnalisation et l'adaptation des systèmes aux besoins spécifiques de l'utilisateur exige une très bonne connaissance de la façon dont les utilisateurs utilisent le logiciel. Afin de mettre en évidence les tendances d'utilisation, nous proposons de regrouper les sessions similaires. Nous comparons trois types de représentation de session dans différents algorithmes de clustering. La deuxième contribution de cette thèse concerne le suivi dynamique de l'utilisation du logiciel. Nous proposons deux méthodes -- basées sur des représentations différentes des actions d'entrée -- pour répondre à deux problématiques industrielles distinctes : la prédiction de la prochaine action et la détection du risque de crash logiciel. Les deux méthodologies tirent parti de la structure récurrente des réseaux LSTM pour capturer les dépendances entre nos données séquentielles ainsi que leur capacité à traiter potentiellement différents types de représentations d'entrée pour les mêmes données. / Radiologists use medical imaging solutions on a daily basis for diagnosis. Improving user experience is a major line of the continuous effort to enhance the global quality and usability of software products. Monitoring applications enable to record the evolution of various software and system parameters during their use and in particular the successive actions performed by the users in the software interface. These interactions may be represented as sequences of actions. Based on this data, this work deals with two industrial topics: software crashes and software usability. Both topics imply on one hand understanding the patterns of use, and on the other developing prediction tools either to anticipate crashes or to dynamically adapt software interface according to users' needs. First, we aim at identifying crash root causes. It is essential in order to fix the original defects. For this purpose, we propose to use a binomial test to determine which type of patterns is the most appropriate to represent crash signatures. The improvement of software usability through customization and adaptation of systems to each user's specific needs requires a very good knowledge of how users use the software. In order to highlight the trends of use, we propose to group similar sessions into clusters. We compare 3 session representations as inputs of different clustering algorithms. The second contribution of our thesis concerns the dynamical monitoring of software use. We propose two methods -- based on different representations of input actions -- to address two distinct industrial issues: next action prediction and software crash risk detection. Both methodologies take advantage of the recurrent structure of LSTM neural networks to capture dependencies among our sequential data as well as their capacity to potentially handle different types of input representations for the same data.

過去對於探勘常見項目集的研究僅限於找出資料庫中交易紀錄的子集合，在這篇論文中，我們提出一個新的探勘主題：常見超集合探勘。常見超集合意指它包含資料庫中各筆紀錄的筆數多於最小門檻值，而原本用來探勘常見子集合的演算法並無法直接套用，因此我們以補集合的角度，提出了三個快速的演算法來解決這個新的問題。首先為Apriori-C：此為使用先廣後深搜尋的演算法，並且以掃描資料庫的方式來決定具有相同長度之候選超集合的支持度，第二個方法是Eclat-C：此為採用先深後廣搜尋的演算法，並且搭配交集法來計算倏選超集合的支持度，最後是DCT：此方法可利用過去常見子集合探勘的演算法來進行探勘，如此可以省下開發新系統的成本。 常見超集合的探勘可以應用在電子化的遠距學習系統，生物資訊及工作排程的問題上。尤其在線上學習系統，我們可以利用常見超集合來代表一群學生的學習行為，並且藉以預測學生的學習成就，使得老師可以及時發現學生的學習迷失等行為；此外，透過常見超集合的探勘，我們也可以為學生推薦個人化的課程，以達到因材施教的教學目標。 在實驗的部份，我們比較了各演算法的效率，並且分別改變實驗資料庫的下列四種變因：1) 交易資料的筆數、2) 每筆交易資料的平均長度、3) 資料庫中項目的總數和4) 最小門檻值。在最後的分析當中，可以清楚地看出我們提出的各種方法皆十分有效率並且具有可延伸性。 / The algorithms for the discovery of frequent itemset have been investigated widely. These frequent itemsets are subsets of database. In this thesis, we propose a novel mining task: mining frequent superset from the database of itemsets that is useful in bioinformatics, E-learning systems, jobshop scheduling, and so on. A frequent superset means that the number of transactions contained in it is not less than minimum support threshold. Intuitively, according to the Apriori algorithm, the level-wise discovering starts from 1-itemset, 2-itemset, and so forth. However, such steps cannot utilize the property of Apriori to reduce search space, because if an itemset is not frequent, its superset maybe frequent. In order to solve this problem, we propose three methods. The first is the Apriori-based approach, called Apriori-C. The second is the Eclat-based approach, called Eclat-C, which is a depth-first approach. The last is the proposed data complement technique (DCT) that we utilize original frequent itemset mining approach to discover frequent superset. The experimental studies compare the performance of the proposed three methods by considering the effect of the number of transactions, the average length of transactions, the number of different items, and minimum support. The analysis shows that the proposed algorithms are time efficient and scalable.

Die folgende Arbeit befasst sich mit der Alltagstauglichkeit moderner Massendatenverarbeitung, insbesondere mit dem Problem der Assoziationsregelanalyse. Vorhandene Datenmengen wachsen stark an, aber deren Auswertung ist für ungeübte Anwender schwierig. Daher verzichten Unternehmen auf Informationen, welche prinzipiell vorhanden sind. Assoziationsregeln zeigen in diesen Daten Abhängigkeiten zwischen den Elementen eines Datenbestandes, beispielsweise zwischen verkauften Produkten. Diese Regeln können mit Interessantheitsmaßen versehen werden, welche dem Anwender das Erkennen wichtiger Zusammenhänge ermöglichen. Es werden Ansätze gezeigt, dem Nutzer die Auswertung der Daten zu erleichtern. Das betrifft sowohl die robuste Arbeitsweise der Verfahren als auch die einfache Auswertung der Regeln. Die vorgestellten Algorithmen passen sich dabei an die zu verarbeitenden Daten an, was sie von anderen Verfahren unterscheidet. Assoziationsregelsuchen benötigen die Extraktion häufiger Kombinationen (EHK). Hierfür werden Möglichkeiten gezeigt, Lösungsansätze auf die Eigenschaften moderne System anzupassen. Als Ansatz werden Verfahren zur Berechnung der häufigsten $N$ Kombinationen erläutert, welche anders als bekannte Ansätze leicht konfigurierbar sind. Moderne Systeme rechnen zudem oft verteilt. Diese Rechnerverbünde können große Datenmengen parallel verarbeiten, benötigen jedoch die Vereinigung lokaler Ergebnisse. Für verteilte Top-N-EHK auf realistischen Partitionierungen werden hierfür Ansätze mit verschiedenen Eigenschaften präsentiert. Aus den häufigen Kombinationen werden Assoziationsregeln gebildet, deren Aufbereitung ebenfalls einfach durchführbar sein soll. In der Literatur wurden viele Maße vorgestellt. Je nach den Anforderungen entsprechen sie je einer subjektiven Bewertung, allerdings nicht zwingend der des Anwenders. Hierfür wird untersucht, wie mehrere Interessantheitsmaßen zu einem globalen Maß vereinigt werden können. Dies findet Regeln, welche mehrfach wichtig erschienen. Der Nutzer kann mit den Vorschlägen sein Suchziel eingrenzen. Ein zweiter Ansatz gruppiert Regeln. Dies erfolgt über die Häufigkeiten der Regelelemente, welche die Grundlage von Interessantheitsmaßen bilden. Die Regeln einer solchen Gruppe sind daher bezüglich vieler Interessantheitsmaßen ähnlich und können gemeinsam ausgewertet werden. Dies reduziert den manuellen Aufwand des Nutzers. Diese Arbeit zeigt Möglichkeiten, Assoziationsregelsuchen auf einen breiten Benutzerkreis zu erweitern und neue Anwender zu erreichen. Die Assoziationsregelsuche wird dabei derart vereinfacht, dass sie statt als Spezialanwendung als leicht nutzbares Werkzeug zur Datenanalyse verwendet werden kann. / The importance of data mining is widely acknowledged today. Mining for association rules and frequent patterns is a central activity in data mining. Three main strategies are available for such mining: APRIORI , FP-tree-based approaches like FP-GROWTH, and algorithms based on vertical data structures and depth-first mining strategies like ECLAT and CHARM. Unfortunately, most of these algorithms are only moderately suitable for many “real-world” scenarios because their usability and the special characteristics of the data are two aspects of practical association rule mining that require further work. All mining strategies for frequent patterns use a parameter called minimum support to define a minimum occurrence frequency for searched patterns. This parameter cuts down the number of patterns searched to improve the relevance of the results. In complex business scenarios, it can be difficult and expensive to define a suitable value for the minimum support because it depends strongly on the particular datasets. Users are often unable to set this parameter for unknown datasets, and unsuitable minimum-support values can extract millions of frequent patterns and generate enormous runtimes. For this reason, it is not feasible to permit ad-hoc data mining by unskilled users. Such users do not have the knowledge and time to define suitable parameters by trial-and-error procedures. Discussions with users of SAP software have revealed great interest in the results of association-rule mining techniques, but most of these users are unable or unwilling to set very technical parameters. Given such user constraints, several studies have addressed the problem of replacing the minimum-support parameter with more intuitive top-n strategies. We have developed an adaptive mining algorithm to give untrained SAP users a tool to analyze their data easily without the need for elaborate data preparation and parameter determination. Previously implemented approaches of distributed frequent-pattern mining were expensive and time-consuming tasks for specialists. In contrast, we propose a method to accelerate and simplify the mining process by using top-n strategies and relaxing some requirements on the results, such as completeness. Unlike such data approximation techniques as sampling, our algorithm always returns exact frequency counts. The only drawback is that the result set may fail to include some of the patterns up to a specific frequency threshold. Another aspect of real-world datasets is the fact that they are often partitioned for shared-nothing architectures, following business-specific parameters like location, fiscal year, or branch office. Users may also want to conduct mining operations spanning data from different partners, even if the local data from the respective partners cannot be integrated at a single location for data security reasons or due to their large volume. Almost every data mining solution is constrained by the need to hide complexity. As far as possible, the solution should offer a simple user interface that hides technical aspects like data distribution and data preparation. Given that BW Accelerator users have such simplicity and distribution requirements, we have developed an adaptive mining algorithm to give unskilled users a tool to analyze their data easily, without the need for complex data preparation or consolidation. For example, Business Intelligence scenarios often partition large data volumes by fiscal year to enable efficient optimizations for the data used in actual workloads. For most mining queries, more than one data partition is of interest, and therefore, distribution handling that leaves the data unaffected is necessary. The algorithms presented in this paper have been developed to work with data stored in SAP BW. A salient feature of SAP BW Accelerator is that it is implemented as a distributed landscape that sits on top of a large number of shared-nothing blade servers. Its main task is to execute OLAP queries that require fast aggregation of many millions of rows of data. Therefore, the distribution of data over the dedicated storage is optimized for such workloads. Data mining scenarios use the same data from storage, but reporting takes precedence over data mining, and hence, the data cannot be redistributed without massive costs. Distribution by special data semantics or user-defined selections can produce many partitions and very different partition sizes. The handling of such real-world distributions for frequent-pattern mining is an important task, but it conflicts with the requirement of balanced partition.

Nous présentons dans cette thèse un système de correction automatique d'annotations (tags) fournies par des utilisateurs qui téléversent des vidéos sur des sites de partage de documents multimédia sur Internet. La plupart des systèmes d'annotation automatique existants se servent principalement de l'information textuelle fournie en plus de la vidéo par les utilisateurs et apprennent un grand nombre de "classifieurs" pour étiqueter une nouvelle vidéo. Cependant, les annotations fournies par les utilisateurs sont souvent incomplètes et incorrectes. En effet, un utilisateur peut vouloir augmenter artificiellement le nombre de "vues" d'une vidéo en rajoutant des tags non pertinents. Dans cette thèse, nous limitons l'utilisation de cette information textuelle contestable et nous n'apprenons pas de modèle pour propager des annotations entre vidéos. Nous proposons de comparer directement le contenu visuel des vidéos par différents ensembles d'attributs comme les sacs de mots visuels basés sur des descripteurs SIFT ou des motifs fréquents construits à partir de ces sacs. Nous proposons ensuite une stratégie originale de correction des annotations basées sur la fréquence des annotations des vidéos visuellement proches de la vidéo que nous cherchons à corriger. Nous avons également proposé des stratégies d'évaluation et des jeux de données pour évaluer notre approche. Nos expériences montrent que notre système peut effectivement améliorer la qualité des annotations fournies et que les motifs fréquents construits à partir des sacs de motifs fréquents sont des attributs visuels pertinents / This thesis presents a new system for video auto tagging which aims at correcting the tags provided by users for videos uploaded on the Internet. Most existing auto-tagging systems rely mainly on the textual information and learn a great number of classifiers (on per possible tag) to tag new videos. However, the existing user-provided video annotations are often incorrect and incomplete. Indeed, users uploading videos might often want to rapidly increase their video’s number-of-view by tagging them with popular tags which are irrelevant to the video. They can also forget an obvious tag which might greatly help an indexing process. In this thesis, we limit the use this questionable textual information and do not build a supervised model to perform the tag propagation. We propose to compare directly the visual content of the videos described by different sets of features such as SIFT-based Bag-Of-visual-Words or frequent patterns built from them. We then propose an original tag correction strategy based on the frequency of the tags in the visual neighborhood of the videos. We have also introduced a number of strategies and datasets to evaluate our system. The experiments show that our method can effectively improve the existing tags and that frequent patterns build from Bag-Of-visual-Words are useful to construct accurate visual features

Die folgende Arbeit befasst sich mit der Alltagstauglichkeit moderner Massendatenverarbeitung, insbesondere mit dem Problem der Assoziationsregelanalyse. Vorhandene Datenmengen wachsen stark an, aber deren Auswertung ist für ungeübte Anwender schwierig. Daher verzichten Unternehmen auf Informationen, welche prinzipiell vorhanden sind. Assoziationsregeln zeigen in diesen Daten Abhängigkeiten zwischen den Elementen eines Datenbestandes, beispielsweise zwischen verkauften Produkten. Diese Regeln können mit Interessantheitsmaßen versehen werden, welche dem Anwender das Erkennen wichtiger Zusammenhänge ermöglichen. Es werden Ansätze gezeigt, dem Nutzer die Auswertung der Daten zu erleichtern. Das betrifft sowohl die robuste Arbeitsweise der Verfahren als auch die einfache Auswertung der Regeln. Die vorgestellten Algorithmen passen sich dabei an die zu verarbeitenden Daten an, was sie von anderen Verfahren unterscheidet. Assoziationsregelsuchen benötigen die Extraktion häufiger Kombinationen (EHK). Hierfür werden Möglichkeiten gezeigt, Lösungsansätze auf die Eigenschaften moderne System anzupassen. Als Ansatz werden Verfahren zur Berechnung der häufigsten $N$ Kombinationen erläutert, welche anders als bekannte Ansätze leicht konfigurierbar sind. Moderne Systeme rechnen zudem oft verteilt. Diese Rechnerverbünde können große Datenmengen parallel verarbeiten, benötigen jedoch die Vereinigung lokaler Ergebnisse. Für verteilte Top-N-EHK auf realistischen Partitionierungen werden hierfür Ansätze mit verschiedenen Eigenschaften präsentiert. Aus den häufigen Kombinationen werden Assoziationsregeln gebildet, deren Aufbereitung ebenfalls einfach durchführbar sein soll. In der Literatur wurden viele Maße vorgestellt. Je nach den Anforderungen entsprechen sie je einer subjektiven Bewertung, allerdings nicht zwingend der des Anwenders. Hierfür wird untersucht, wie mehrere Interessantheitsmaßen zu einem globalen Maß vereinigt werden können. Dies findet Regeln, welche mehrfach wichtig erschienen. Der Nutzer kann mit den Vorschlägen sein Suchziel eingrenzen. Ein zweiter Ansatz gruppiert Regeln. Dies erfolgt über die Häufigkeiten der Regelelemente, welche die Grundlage von Interessantheitsmaßen bilden. Die Regeln einer solchen Gruppe sind daher bezüglich vieler Interessantheitsmaßen ähnlich und können gemeinsam ausgewertet werden. Dies reduziert den manuellen Aufwand des Nutzers. Diese Arbeit zeigt Möglichkeiten, Assoziationsregelsuchen auf einen breiten Benutzerkreis zu erweitern und neue Anwender zu erreichen. Die Assoziationsregelsuche wird dabei derart vereinfacht, dass sie statt als Spezialanwendung als leicht nutzbares Werkzeug zur Datenanalyse verwendet werden kann. / The importance of data mining is widely acknowledged today. Mining for association rules and frequent patterns is a central activity in data mining. Three main strategies are available for such mining: APRIORI , FP-tree-based approaches like FP-GROWTH, and algorithms based on vertical data structures and depth-first mining strategies like ECLAT and CHARM. Unfortunately, most of these algorithms are only moderately suitable for many “real-world” scenarios because their usability and the special characteristics of the data are two aspects of practical association rule mining that require further work. All mining strategies for frequent patterns use a parameter called minimum support to define a minimum occurrence frequency for searched patterns. This parameter cuts down the number of patterns searched to improve the relevance of the results. In complex business scenarios, it can be difficult and expensive to define a suitable value for the minimum support because it depends strongly on the particular datasets. Users are often unable to set this parameter for unknown datasets, and unsuitable minimum-support values can extract millions of frequent patterns and generate enormous runtimes. For this reason, it is not feasible to permit ad-hoc data mining by unskilled users. Such users do not have the knowledge and time to define suitable parameters by trial-and-error procedures. Discussions with users of SAP software have revealed great interest in the results of association-rule mining techniques, but most of these users are unable or unwilling to set very technical parameters. Given such user constraints, several studies have addressed the problem of replacing the minimum-support parameter with more intuitive top-n strategies. We have developed an adaptive mining algorithm to give untrained SAP users a tool to analyze their data easily without the need for elaborate data preparation and parameter determination. Previously implemented approaches of distributed frequent-pattern mining were expensive and time-consuming tasks for specialists. In contrast, we propose a method to accelerate and simplify the mining process by using top-n strategies and relaxing some requirements on the results, such as completeness. Unlike such data approximation techniques as sampling, our algorithm always returns exact frequency counts. The only drawback is that the result set may fail to include some of the patterns up to a specific frequency threshold. Another aspect of real-world datasets is the fact that they are often partitioned for shared-nothing architectures, following business-specific parameters like location, fiscal year, or branch office. Users may also want to conduct mining operations spanning data from different partners, even if the local data from the respective partners cannot be integrated at a single location for data security reasons or due to their large volume. Almost every data mining solution is constrained by the need to hide complexity. As far as possible, the solution should offer a simple user interface that hides technical aspects like data distribution and data preparation. Given that BW Accelerator users have such simplicity and distribution requirements, we have developed an adaptive mining algorithm to give unskilled users a tool to analyze their data easily, without the need for complex data preparation or consolidation. For example, Business Intelligence scenarios often partition large data volumes by fiscal year to enable efficient optimizations for the data used in actual workloads. For most mining queries, more than one data partition is of interest, and therefore, distribution handling that leaves the data unaffected is necessary. The algorithms presented in this paper have been developed to work with data stored in SAP BW. A salient feature of SAP BW Accelerator is that it is implemented as a distributed landscape that sits on top of a large number of shared-nothing blade servers. Its main task is to execute OLAP queries that require fast aggregation of many millions of rows of data. Therefore, the distribution of data over the dedicated storage is optimized for such workloads. Data mining scenarios use the same data from storage, but reporting takes precedence over data mining, and hence, the data cannot be redistributed without massive costs. Distribution by special data semantics or user-defined selections can produce many partitions and very different partition sizes. The handling of such real-world distributions for frequent-pattern mining is an important task, but it conflicts with the requirement of balanced partition.

No description available.