2016 Fiscal Year Research-status Report
A Study on Social Context Summarization
| Project/Area Number |
15K16048
|
|---|
| Research Institution | Japan Advanced Institute of Science and Technology |
|---|
Principal Investigator |
NGUYEN MinhLe 北陸先端科学技術大学院大学, 先端科学技術研究科, 准教授 (30509401)
|
|---|
| Project Period (FY) |
2015-04-01 – 2018-03-31
|
| Keywords | comment extraction / sentence extraction / text summarization / contextual summarization / word representation / deep learning |
|---|
| Outline of Annual Research Achievements |
In this fiscal year, we have achieved good results on social context summarization. Our work published in EICIR 2016 is extended and published in the journal. We developed SoRTESum to take the advantages of social information such as document content reflection to extract summary sentences and social messages. SoRTESum was extensively evaluated on two datasets which shows the state of the art performance. Another work is to present a summary model (SoCRFSum), which is formulated as a sequence labeling problem (CRFs), which exploits the support of external information to model sentences and comments. SoCRFSum was validated on a dataset collected from Yahoo News and show interesting results. We also report a summarization method named SoSVMRank, which integrates the social information of a Web document to generate a high-quality summarization.
Beside that, we have developed two corpora for social contextual summarization in both English and Vietnamese.
Along with research on deploying deep learning for text summarization, we would like to adapt our work by exploiting deep learning techniques. We propose a novel method for mapping a concept to a vector representation using Wiktionary sense definition. We have published an interesting work for turning SVM parameters using Genetic program and it is published in an international journal. This method can compare another models and it can help to turn parameters in SVM learning. This can be widely applied for many machine learning tasks including its applications in text summarization.
|
| Current Status of Research Progress |
Current Status of Research Progress
1: Research has progressed more than it was originally planned.
Reason
In this year, we have successfully exploited social context summarization system and showed that our model attained the state of the art performance. We have reported our works on two journals and top conferences.
|
| Strategy for Future Research Activity |
In the future work, we will focus on multiple document in social contextual summarization and abstractive text summarization. We also would like to exploit deep learning models for social context summarization.
We will also consider the task of sentence compression when considering social context.
|
| Remarks |
A summarization sentence summarization system
|
