This paper presents NoSQL Over SQL Block as a Value Database (NOSD), a system that speeds up data retrieval time and availability in very large relational databases. NOSD proposes a Block as a Value model (BaaV). Unlike a relational database model where a relation is$R(K,\ A_{1},\ A_{2},\ \ldots A_{n})$, with a key attribute$K$and a set of attributes of the relation:$A_{1}, A_{2}, \ldots A_{n}$, BaaV represents a relation$R(K, r_{1}, r_{2}, \ldots r_{n})$with a key attribute$K$and a set of$n$relations called blocks. Each$r$contains a set of its own attributes denoted as$r(k,\ a_{1},\ a_{2},\ldots a_{n})$with a key attribute$k$and a set of$n$attributes. The relations$r_{1}, r_{2}, \ldots r_{n}$in$R$are related through foreign key relationships to a super relation$R$with primary key$K$. The BaaV model is then denoted in a keyed block format$R\{K,\ B\}$, where$K$is a key to a block of values$B$of partial relations implemented on NoSQL databases and replicating existing large relational database systems. As opposed to conventional systems such as Zidian, Google's Spanner, SparkSQL and Simple Buttom-Up (SBU) which implement SQL over NoSQL and replicate data into different nodes, NOSD implements NoSQL over SQL and uses Lucene functionality on NoSQL to enhance data retrieval costs. Experimenting with our proposed model, we demonstrated the performance of NOSD under the following conditions to prove its novelty (a) scan free queries, and (b) bounded queries on NoSQL databases. We showed that NOSD (a) performs excellently than ordinary relational databases (b) guarantees no scans for no scan queries (c) allows parallelization in query execution, and (d) can be deployed into existing SQL databases with guaranteed horizontal scalability, data retention and accurate autonomous data replication. Using existing benchmark systems, we demonstrated that NOSD outperforms existing SQL databases, SQL over NoSQL systems and is novel in ensuring that existing large SQL database systems utilize the functionalities of NoSQL databases without data loss.$A_{1}, A_{2}, \ldots A_{n}$

In Collaborative Filtering methods, tailored recommendations cannot be obtained when the user-item matrix is sparse (i.e., has low user-item interactions such as item ratings or purchases). Conventional recommendation systems (ChoiRec12, HPCRec18, HSPRec19) utilizing mining techniques such as clustering, frequent and sequential pattern mining along with click and purchase similarity measures for item recommendation cannot perform well when the user-item interactions are less, as the number of items keep increasing rapidly. Additionally, they have not explored the integration of semantic information of products extracted from customers' purchase histories into the item matrix and the pattern mining process. To address this problem, this paper proposes (SEMSRec) which integrates semantic information of E-commerce products extracted from purchase histories into all phases of recommendation process (pre-processing, pattern mining and recommendation). This is achieved by i) learning semantic similarities between items from customers' purchase histories using Prod2vec model, ii) leveraging this information to mine semantically rich sequential purchase patterns and, iii) enriching the item matrix with semantic and sequential product purchase information before applying item based collaborative filtering. Thus, SEMSRec can provide Top-K personalized recommendations based on semantic similarities between items without the need for users' ratings on items. Experimental results on publically available E-commerce data set show that SEMSRec provides more relevant recommendations over other existing methods.

Discovering potentially useful and previously unknown information or knowledge from heterogeneous web contents such as "list all laptop prices from Walmart and Staples between 2013 and 2015 including make, type, screen size, CPU power, year of make", would require the difficult task of finding the schema of web documents from different web pages, performing web content data integration, building their virtual or physical data warehouse integration before web content extraction and mining from the database. Wrappers that extract target information from web pages can be manual, semi-supervised or automatic systems. Automatic systems such as the WebOMiner system, use some data extraction techniques based on parsing the web page html source code into a document object model (DOM) tree, then traverse the DOM for pattern discovery. Some limitations of these existing systems include using complicated matching techniques such as tree matching, Finite state automata, not yielding accurate results for complex queries such as historical and derived.

This paper describes the algorithms (called OTEP DW auto) for automatically updating the integrated games data warehouse and cognitive profile data sources for purposes of identifying child’s cognitive skill level. The techniques decribed in this paper represent an extension to the data integration engine adopted by an online product called “Think2Learn” developed by OTEP Inc.(Online Training & Education Portal). OTEP focuses on using the Internet, natural playing environment of online computer games to give parents automated opportunity to screen and follow their children’s cognitive development. Current data integration efforts of the system when new games (such as speech games) are added or new cognitive skills matrix are added would require manual re-coding of the system which is a costly and time-consuming process. The cognitive skills matrix maps cognitive skills level of games player such as “basic reading level is good” to their games performance in comparison to the norms of other players. The proposed OTEP DW auto is capable of building the OTEP data warehouse schema automatically, extracting, cleaning and propagating data from data sources. It also provides a dynamic graphical user interface GUI-based interface for answering tens of frequently asked questions.

This paper proposes NeuDetect, which applies a classification rule mining Neural Network technique to wireless network packets captured through hardware sensors for purposes of real time detection of anomalous packets. To address the problem of high false alarm rate confronted by current wireless intrusion detection systems, this paper presents a method of applying artificial neural networks mining classification technique to wireless network intrusion detection system. The proposed system, NeuDetect, solution approach is to find normal and anomalous patterns on pre-processed wireless packet records by comparing them with training data using Back-propagation algorithm.

This paper proposes the integration of semantic information drawn from a web application's domain knowledge into all phases of the web usage mining process (preprocessing, pattern discovery, and recommendation/prediction). The goal is to have an intelligent semantics-aware web usage mining framework. This is accomplished by using semantic information in the sequential pattern mining algorithm to prune the search space and partially relieve the algorithm from support counting. In addition, semantic information is used in the prediction phase with low order Markov models, for less space complexity and accurate prediction, that will help ambiguous predictions problem. Experimental results show that semantics-aware sequential pattern mining algorithms can perform 4 times faster than regular non-semantics-aware algorithms with only 26% of the memory requirement.

Position Coded Pre-order Linked Web Access Pattern (PLWAP) mining algorithm is one of the existing efficient web sequential pattern mining algorithms, which stores the frequent sequences of the entire sequential database in a compressed tree form with position coded nodes. However, for very long sequences exceeding thirty two nodes, the number of bits an integer position code can hold, the PLWAP algorithm's performance begins to degrade because it employs linked lists to store conjunctions of long position codes and the linked list traversals slow down the algorithm both during tree construction and mining. PLWAP algorithm also uses each and every node in the frequent 1-item event queue to test for that event inclusion in the suffix tree root set during mining.

This paper proposes WIDS, a wireless intrusion detection system, which applies data mining clustering technique to wireless network data captured through hardware sensors for purposes of real time detection of anomalous behavior in wireless packets. Using hardware sensors to capture network packets enables detection of attacks before they reach access points and ensures all packets transmitted in the networks are analyzed for a more complete attack detection. The proposed mining based technique for wireless network intrusion detection contributes by reducing the need for training data, reducing false positives and increasing the effectiveness of attack detection on networks with few (one to twenty) connections. The proposed WIDS design approach involves real time pre-processing of sensor data using a density-based, Local Sparsity Coefficient (LSC) outlier detection algorithm to assign anomaly scores to the connection records. Connection records with low anomaly scores are used as initial starting cluster centre positions for building clusters. The algorithm continuously derives minimum deviation as the maximum of distances between all pairs of cluster centre positions. New records which have their distances from the closest cluster more than the minimum deviation, are tagged as anomaly and moved to alert cluster. One major result of this paper is detection of MAC spoofing attacks by tracking sequence numbers, which ensures duplicate or spoofed (stolen) MAC addresses are not used in the network.

Data stream applications like sensor network data, click stream data, have data arriving continuously at high speed rates and require online mining process capable of delivering current and near accurate results on demand without full access to all historical stored data. Frequent sequential mining is the process of discovering frequent sequential patterns in data sequences as found in applications like Web log access sequences. Mining frequent sequential patterns on data stream applications contend with many challenges such as limited memory for unlimited data, inability of algorithms to scan infinitely flowing original dataset more than once and to deliver current and accurate result on demand. Existing work on mining frequent patterns on data streams are mostly for non-sequential patterns. This paper proposes SSM-algorithm (sequential stream mining-algorithm), that uses three types of data structures (D-List, PLWAP tree and FSP-tree) to handle the complexities of mining frequent sequential patterns in data streams. It summarizes frequency counts of items with the D-list, continuously builds PLWAP tree and mines frequent sequential patterns of batches of stream records, maintains mined frequent sequential patterns incrementally with FSP tree. The proposed algorithm can be deployed to analyze e-commerce data where the primary source of data is click stream data.

A horizontal fragment of a database class in an object-oriented database system contains subsets of its instance objects (or class extents) reflecting the way applications are accessing database objects. Allocating well-defined fragments of classes to distributed sites has the advantage of minimizing transmission costs of data to remote sites as well as minimizing retrieval time of data needed locally. A re-fragmentation of the system is needed when application access and schema information have undergone sufficient changes. We provide a technique for measuring the performance of object horizontal fragments placed at distributed sites. This work provides a platform for dynamic object horizontal fragmentation and for comparing object horizontal fragmentation schemes.

Careful selection of aggregate views and some of their most used indexes to materialize in a data warehouse reduces the warehouse query response time as well as warehouse maintenance cost under some storage space constraint. Data warehouses collect and store large amounts of integrated enterprise data from a number of independent data sources over a long period of time. Warehouse data are used for online analytical processing to assist management in making quick and competitive business decisions. Precomputing and storing summary tables (materialized views) reduces the amount of time needed to recompute these views across several source tables in order to answer complex warehouse queries. A data cube is an elegant way for representing aggregate information in a Warehouse and is an n-dimensional view with 2/sup n/ subviews. The paper presents a uniform technique for selecting the subviews of the data cube and their indexes to materialize in order to produce the best resultant benefit to the system in terms of query response time and maintenance cost while satisfying some storage space constraint.