| 研究課題/領域番号 |
15K00428
|
|---|
| 研究機関 | 会津大学 |
|---|
研究代表者 |
Paik Incheon 会津大学, コンピュータ理工学部, 教授 (70336478)
|
|---|
| 研究期間 (年度) |
2015-04-01 – 2018-03-31
|
| キーワード | Service Discovery / Social Service Network / Service Graph / Distributed Processing / Big Data / Hadoop / Map-Reduce / Deep Learning |
|---|
| 研究実績の概要 |
In the second year, we have done the following 3 topics: 1) Evaluation of service discovery performance on Global Social Service Network (GSSN) 2) Exploiting service discovery issues on GSSN, 3) Improving link quality.
First, we evaluated service discovery performance based on Map-Reduce GSSN(MR-GSSN). As the contents of the GSSN with 4 attributes is not adequate for real service discovery, we setup some basic conditioned and developed simulation method. The result for service discovery from MR-GSSN is almost same as that of GSSN. As computation performance had been increased by 30 times comparing to non-distributed environment, our approach can be recognized as a great success.
Second, in this research, we proposed Linked Social Service Network (LSSN) with multiple feature attributes based service discovery for Big Data Analytics. It is a combination of two advantages, which are precision and sociability of web services. According to the experimental results, we could find that the approaches are performing well.
Third, the calculation method of functionality of our existing approach was very simple, and it could not show good relationships between services. In this research, we used word embedding using deep learning to find related terms that are used for services. By applying this process to service data, it was possible to find and connect to similar services more efficiently. In the stage of this year, we have taken the first step to process terms in functional service similarity.
|
| 現在までの達成度 (区分) |
現在までの達成度 (区分)
2: おおむね順調に進展している
理由
Entire research progress is going smoothly. I explain the three aspects of the whole research in AY 2016 according to the last year’s plan.
1)Evaluation of service discovery performance on MR-GSSN: In this research, we defined a set of evaluation metrics for evaluation, and by the metrics, success rate about several conditions were calculated, and they showed very similar result with less than 1% error comparing with that of the GSSN. However, we could get improved computation performance with 30 times based on 18 Hadoop data nodes to that of one node computation.
2)Exploiting service discovery issues on GSSN: We tried to apply the GSSN to discover services in Big Data Analytics (BDA) domain. BDA Service publishing and composition are two most critical factors to provide comprehensive service. A new algorithm to map GSSN to BDA domain service has been developed and evaluated. Our experiment results showed that we have successfully completed the second stage of our architecture to automate the BDA.
3) Improving link quality: Basic background of this approach is to find real services for Social Network Service(SNS) or big data infrastructure using Deep Learning. A new method of service network construction using word embedding has been studied. It is the starting stage and result cannot be satisfied and we will improve this algorithm or change the current skip-gram and NCE algorithm into new one with more data.
|
| 今後の研究の推進方策 |
Next fiscal year is the last phase of the research. In the first year, Map-Reduce GSSN (MR-GSSN) algorithm and a distributed Hadoop infrastructure with 18 nodes had been constructed. Next in the second year, evaluation of service discovery performance on the MR-GSSN was carried out, and showed the same result of that of GSSN. And an application of the MR-GSSN to service discovery on Big Data Analytics domain has been investigated. Finally, a new approach to use our MR-GSSN to real field such as SNS services or BDA domain using deep learning.
In the next year (FY 2017), deeper investigation for the new approach to construct GSSN focusing functionality to be able to real application fields (APIs for Facebook, Twitter, Tensorflow, Hadoop, Spark, other machine learning utilities on the Web). The following will be studied.
1)Developing learning approach and algorithms to understand terms for specifications of input/output/precondition/effects and operation information of services using Deep Machine Learning methods. Result is to be well-indexed corpus using RDF or OWL ontology.
2)Developing algorithms and methods for calculating quality of links focusing functionality to reflex real situation of application fields mentioned before for GSSN.
3)Performance evaluation of application domain specified GSSN for discovery (further recommendation) aspect.
|
| 次年度使用額が生じた理由 |
There will be more usage for honorarium and travel fee for international conferences in the next year. The fee has been saved.
|
| 次年度使用額の使用計画 |
The fee will be used for student honorarium and business travel fee.
|
