Document Summarization using Semantic Feature based on Cloud

Transcription

1 Advanced Science and echnology Letters, pp Document Summarization using Semantic Feature based on Cloud Yoo-Kang Ji 1, Yong-Il Kim 2, Sun Park 3 * 1 Dept. of Information & Communication Engineering, Dongshin Univ., Korea 2 onam University, South Korea 3 Mokpo National University, South Korea 1 neobacje@gmail.com 2 yikim@honam.ac.kr 3 sunpark@mokpo.ac.kr Abstract. his paper proposes a document summarization method using the extracted semantic feature which it is extracted by distributed parallel processing of NMF based cloud technique of adoop. he proposed method can well represent the inherent structure of documents using the semantic feature by the non-negative matrix factorization (NMF). In addition, it can summarize the big data document using adoop. he experimental results demonstrate that the proposed method can summarize the big data document which a single computer cannot summarize those. Keywords: document summarization, semantic features, distributed parallel processing, NMF 1 Introduction ith the fast growth of the Internet access by user, has increased the necessity of the information seeking. owever, it is difficult to find suitable information for net surfers from cyber space. Summary information can help to users, which the user can save time not only in deciding whether it is interesting or not but also in finding the information without having to read the full information. Document summarization is the process of reducing the sizes of documents while maintaining their basic outlines. hat is, it should distill the most important information (i.e., topics of document) from the document. he summarization method can involve either generic summaries or query-based summaries. A generic summary distills an overall sense of a document s contents, whereas a query-based summary distills only the contents of a document that is relevant to a user s query. It can also divide into single-document summarization or multi-document summarization according to the scope of the summary target. he purpose of multi-document summarization is to produce a single summary from a set of related documents, whereas single-document summarization is intended to summarize only one document [1]. raditional document summarization methods are restricted to summarize suitable information from the exploding cyber data (i.e., SNS, , message, blog, etc.), * Corresponding author ISSN: ASL Copyright 2013 SERSC

2 Advanced Science and echnology Letters since it have been studying for enhancing the summarization precision which it uses various statistical or natural language processing methods on single computer or server. In order to resolve the limitations of the traditional document summarizations, this paper study document summarization method which the information is summarized from a big document data. he proposed method uses the extracted semantic feature of document by distributed parallel processing of NMF based cloud technique of adoop [2] to summarize a big document data on the cyber space. he proposed method can well represent the inherent structure of documents using the semantic feature by the non-negative matrix factorization (NMF). In addition, it can summarize the big data document using adoop. he experimental results demonstrate that the proposed method can summarize the big data document which a single computer cannot summarize those. he rest of the paper is organized as follows: Section 2 describe the NMF algorithm in detail. In Section 3, adoop framework is introduced. In Section 4 explains the proposed methodlts. Finally, we conclude in Section 5. 2 Non-negative Matrix Factorization his section reviews NMF theory. In this paper, we define the matrix notation as follows: Let X *j be j th column vector of matrix X, X i * be i th row vector, and X ij be the element of i th row and j th column. NMF is to decompose a given m n matrix A into a non-negative semantic feature matrix and a non-negative semantic variable matrix as shown in Equation (1) [3]. A (1) where is a m r non-negative matrix and is a r n non-negative matrix. Usually r is chosen to be smaller than m or n, so that the total sizes of and are smaller than that of the original matrix A. he objective function is used minimizing the Euclidean distance between each ~ column of A and its approximation A =, which was proposed by Lee and Seung [10]. As an objective function, the Frobenius norm is used: Θ E (, ) A 2 F m n A ij i= 1 j= 1 l= 1 r il lj 2 (2) Updating and is kept until ΘE (, ) converges under the predefined threshold or exceeds the number of repetition. he update rules are as follows: 52 Copyright 2013 SERSC

3 Advanced Science and echnology Letters ( A), ( ) ( A ) ( ) (3) 3 adoop Framework he Apache adoop project develops open-source software for reliable, scalable, distributed computing. he adoop project includes the Apache adoop software library which is a framework that allows for the distributed processing of large data sets across clusters of computers using simple programming models. It is designed to scale up from single servers to thousands of machines, each offering local computation and storage. Rather than rely on hardware to deliver high-availability, the library itself is designed to detect and handle failures at the application layer, so delivering a highly-available service on top of a cluster of computers, each of which may be prone to failures [2]. 4 Proposed Document Clustering Method his paper proposes a document summarization method using semantic feature by NMF based adoop. he proposed method consists of two phases: summarization module, and distributed parallel processing module, as shown in Figure 1. In the subsections below, each phase is explained in full. Internet SNS User 1. Summarization module (a) preprocessing (b) summary algorithm results (c) adoop 2. Distributed parallel processing module Fig. 1. Document summarization method using adoop and semantic features Copyright 2013 SERSC 53

4 Advanced Science and echnology Letters In the summarization module consists of preprocessing and summary algorithm. In the preprocessing phase of Figure 1(a), Rijsbergen s stop words list is used to remove all stop words, and word stemming is removed using Porter s stemming algorithm [4, 5]. hen, the term document frequency matrix A is constructed from the document set. he term document frequency matrix is saved by distributed parallel processing to adoop framework in Figure 1(c). In the summary algorithm phase of Figure 1(b), Semantic features of document for summarizing are extracted by Liu s [6] NMF method based on distributed parallel processing on adoop MapReduce programming. he similarity between query and semantic feature vectors is calculated by cosine similarity. he semantic feature vector having the largest similarity value is selected. he semantic variable vector corresponding to the selected semantic feature vector is selected. he sentence corresponding to the largest value of semantic variable is extracted. hese steps are repeated until the predefined number of sentences to be summarized is reached. able 1 shows Liu s NMF method using MapReduce on adoop. able 1. Liu s NMF method using MapReduce [6] Stage ( A) ( ) ( A ) ( ) 1 X 1 = is computed to Map 2 Y 1 = is compute to Map 3 = X 1 is computed to Map Y 1 X 2 = A is computed to Map Y 2 = is computed to Map = X 2 is computed to Map Y 2 5 Conclusion raditional document summarization methods are restricted to summarize suitable information from the big document data on Internet (i.e., SNS, , message, blog, etc.), since it have been studying for enhancing the summarization precision which it uses various statistical or natural language processing methods on single computer or server. In order to resolve the limitations of the summarizations, this paper proposed document summarization method which the information is summarized from a big document data. he proposed method uses the extracted semantic feature of document by distributed parallel processing of NMF based adoop MapReduce [2] to summarize a big document data on Internet. he proposed method can well represent the inherent structure of documents using the semantic feature by the non-negative 54 Copyright 2013 SERSC

5 Advanced Science and echnology Letters matrix factorization (NMF). In addition, it can summarize the big data document using adoop. References 1. Mani, Automatic Summarization, John Benjamins Publishing Company, he Apache adoop project, (2013) 3. D. D. Lee,. S. Seung, Learning the parts of objects by non-negative matrix factorization, Nature, 401, pp , Oct. (1999) 4. B. Y. Ricardo, R. N. Berthier, Moden Information Retrieval: the concepts and technology behind search Second edition, ACM Press, (2011) 5.. B. Frankes, B. Y. Ricardo, Information Retrieval: Data Structure & Algorithms, Prentice-all, (1992) 6. C. Liu,. C. Yang, J. Fan, L.. e, Y. M. ang, "Distributed Nonnegative Matrix Factorization for eb-scale Dyadic Data Analysis on MapReduce," in Proceeding of the International orld ide eb Conferene Comittee, USA, pp.1-10, (2010) Copyright 2013 SERSC 55