Jaeseok Myung

SPARQL basic graph pattern processing with iterative MapReduce

There have been a number of approaches to adopt the RDF data model and the MapReduce framework for a data warehouse, as the data model is suitable for data integration and the data processing framework is good for large-scale fault-tolerant data analyses. Nevertheless, most approaches consider the data model and the framework separately. It has been difficult to create synergy because there have been only a few algorithms which connects the data model and the framework. In this paper, we offer a general and efficient MapReduce algorithm for SPARQL Basic Graph Pattern which is a set of triple patterns to be joined. In a MapReduce world, it is known that the join operation requires computationally expensive MapReduce iterations. For this reason, we minimize the number of iterations with the followings. First, we adopt traditional multi-way join into MapReduce instead of multiple individual joins. Second, by analyzing a given query, we select a good join-key to avoid unnecessary iterations. As a result, the algorithm shows good performance and scalability in terms of time and data size.

A general maturity model and reference architecture for saas service

In today’s dynamic IT environment with increased global competition, enterprises must achieve greater business agility and decrease the TCO (Total Cost of Ownership) of their system for service. As the need for innovative software circulation process emerges, SaaS (Software as a Service) is introduced for integrating software service framework. But most of current ASP (Application Service Provider) Players have difficulties to migrate their systems to SaaS Platform for the lack of maturity model and process. In this paper, we have surveyed several cases of SaaS service, and we identified the common key functions of SaaS service. We contend that the practical maturity model is a key enabler for achieving migration to innovative SaaS service platform. To assist in building our SaaS maturity model, we defined two important axes of maturity model and introduced the detailed components of each phase with the reference architecture which contains the essential activities according to the common functions of SaaS service.

The Method for a Summarization of Product Reviews Using the User's Opinion

As the number of transactions in E-market places is growing, more and more product information and product reviews are posted on the Internet. Because customers want to purchase good products, product reviews became most important information. But, because of the massive volume of reviews, customers can’t read all reviews. In order to solve this problem, a lot of research is being carried out in Opinion Mining. Through the Opinion Mining, we can know about contents of whole product reviews. Traditionally research on Natural Language Processing was applied to the Opinion Mining area in early stage. Recently, the computational statistics are applied to handle massive volume of reviews. In this research, we suggest a method for summarization of product reviews using the user’s opinion, feature occurrences, and the rate of review in order to improve the performance of existing methods. With this method, we can handle massive volumes of reviews in a short time efficiently. We guarantee the correctness of the review summary by finding out the semantic meaning of reviews. Besides, we show these advantages through some experiments.

Handling data skew in join algorithms using MapReduce

We introduce a skew handling algorithm, called multi-dimensional range partitioning.The proposed algorithm is more efficient than traditional MapReduce-based join algorithms.The proposed algorithm is scalable regardless of the size of input data. One of the major obstacles hindering effective join processing on MapReduce is data skew. Since MapReduces basic hash-based partitioning method cannot solve the problem properly, two alternatives have been proposed: range-based and randomized methods. However, they still remain some drawbacks: the range-based method does not handle join product skew, and the randomized method performs worse than the basic hash-based partitioning when input relations are not skewed. In this paper, we present a new skew handling method, called multi-dimensional range partitioning (MDRP). The proposed method overcomes the limitations of traditional algorithms in two ways: 1) the number of output records expected at each machine is considered, which leads to better handling of join product skew, and 2) a small number of input records are sampled before the actual join begins so that an efficient execution plan considering the degree of data skew can be created. As a result, in a scalar skew experiment, the proposed join algorithm is about 6.76 times faster than the range-based algorithm when join product skew exists and about 5.14 times than the randomized algorithm when input relations are not skewed. Moreover, through the worst-case analysis, we show that the input and the output imbalances are less than or equal to 2. The proposed algorithm does not require any modification to the original MapReduce environment and is applicable to complex join operations such as theta-joins and multi-way joins.

Matrix chain multiplication via multi-way join algorithms in MapReduce

In this paper, we translate the multiplication of several matrices into a multi-way join operation among several relations. Matrix multiplication is widely used for many graph algorithms, such as those that calculate the transitive closure. These algorithms benefit from the multi-way join operation because this operation reduces the number of binary multiplications. Our implementation is based on the MapReduce framework, allowing us to provide scalable computation for large matrices. Although several papers have investigated matrix multiplication using MapReduce, this paper takes a different perspective. First, we expand the problem from binary multiplication to n-ary multiplication. For this reason, we apply the concept of parallelism, not only to an individual operation but also to the entire equation. Second, we represent a matrix as a relation consisting of (row, col, val) records and translate a multiplication into a join operation in database systems. This facilitates the efficient storage of sparse matrices, which are very common in real-world graph data, and the easy manipulation of matrices. Although this work is still in progress, we conducted a number of experiments to verify the idea. We also discuss current limitations and future works.

A MapReduce-based filtering algorithm for vector similarity join

Vector Similarity Join is a fundamental operation that is utilized in data cleaning and analysis. Since most objects can be represented as feature vectors, finding similar pairs of objects is quite an important task. However, Vector Similarity Join is a heavy computational job, because its complexity is proportional to the square of the number of vectors. In order to diminish its computational load, many filtering techniques have been proposed so far. In addition to that, algorithms for distributed systems also have been researched to manage large datasets. But, the state-of-the-art studies also suffer from voluminous computations. In this paper, we propose a MapReduce algorithm that efficiently executes Vector Similarity Join. In the first stage of our algorithm, we use prefix filtering to reduce the number of candidate pairs. The second stage calculates similarities from candidate pairs of the first stage. We present candidates quantity prediction formulas to demonstrate the effectiveness of our algorithm. Experimental results show that our algorithm outperforms state-of-the-art MapReduce algorithms.

Exploiting inter-operation parallelism for matrix chain multiplication using MapReduce

In this paper, we address the matrix chain multiplication problem, i.e., the multiplication of several matrices. Although several studies have investigated the problem, our approach has some different points. First, we propose MapReduce algorithms that allow us to provide scalable computation for large matrices. Second, we transform the matrix chain multiplication problem from sequential multiplications of two matrices into a single multiplication of several matrices. Since matrix multiplication is associative, this approach helps to improve the performance of the algorithms. To implement the idea, we adopt multi-way join algorithms in MapReduce that have been studied in recent years. In our experiments, we show that the proposed algorithms are fast and scalable, compared to several baseline algorithms.

A Holistic Approach to Product Review Summarization

As online shopping is becoming commonplace, more and more product information and product reviews are posted on the Internet. Because customers cannot see and feel the products directly, product reviews are becoming an essential source of qualitative information. As a result, the volume of reviews is increasing drastically and review summarization (or opinion mining) is becoming an important tool for a practical utilization. Presented in this paper is a holistic approach to product review summarization which improves on existing methods. In extracting product features and opinion words, we present a flexible extraction tool that can be customized according to the characteristics of source documents. For sentiment analysis and feature scoring, we utilize the rate scores, which are overall ratings of the product by users. Feature dependent sentiment polarities of opinion words are also utilized. Experiments show that these considerations improve the accuracy of product review summarization.

A proximity-based fallback model for hybrid web recommender systems

Although there are numerous websites that provide recommendation services for various items such as movies, music, and books, most of studies on recommender systems only focus on one specific item type. As recommender sites expand to cover several types of items, though, it is important to build a hybrid web recommender system that can handle multiple types of items. The switch hybrid recommender model provides a solution to this problem by choosing an appropriate recommender system according to given selection criteria, thereby facilitating cross-domain recommendations supported by individual recommender systems. This paper seeks to answer the question of how to deal with situations where no appropriate recommender system exists to deal with a required type of item. In such cases, the switch model cannot generate recommendation results, leading to the need for a fallback model that can satisfy most users most of the time. Our fallback model exploits a graph-based proximity search, ranking every entity on the graph according to a given proximity measure. We study how to incorporate the fallback model into the switch model, and propose a general architecture and simple algorithms for implementing these ideas. Finally, we present the results of our research result and discuss remaining challenges and possibilities for future research.