TY - JOUR
T1 - Personal research idea recommendation using research trends and a hierarchical topic model
AU - Wang, Hei Chia
AU - Hsu, Tzu Ting
AU - Sari, Yunita
N1 - Funding Information:
The research is based on work supported by Taiwan Ministry of Science and Technology under Grant No. MOST 107-2410-H-006 040-MY3 and MOST 108-2511-H-006-009.
Publisher Copyright:
© 2019, Akadémiai Kiadó, Budapest, Hungary.
PY - 2019/12/1
Y1 - 2019/12/1
N2 - In the era of rapid technological advance, it is an important task for all researchers to keep up with trends when performing research. How to efficiently find suitable research topics while the number of papers is increasing rapidly is worthwhile to explore. To solve such problems, some researchers attempted to find research ideas by topic detection and tracking methods. However, these methods do not consider the users’ background knowledge and preferences, and they express a topic with general keywords, which does not effectively help researchers to develop new research ideas. Existing studies support that the title expresses the research idea the best. This study adapts this concept to propose an automatic title generation method that combines personalized recommendation methods and topic trend analysis methods to achieve this task. First, it uses hierarchical latent tree analysis to find the users’ interests for a topic structure and its representative keywords hidden in the existing research. Second, the interesting topic trends, popularity and user preferences in a hybrid recommendation method are considered. Finally, a natural language generation algorithm that is suitable for the titles of academic papers converts the original recommended-keywords into fluent title sentences that are designed for the users. Experiments have found that adding Google Trend indicators and personal factors can improve the performance of topic recommendations. The automatic title generation method using template-based and statistical information methods leads to excellent performances in both grammatical correctness and semantic expression. Moreover, for the users, the title is indeed more inspirational than the simple keywords for users to develop new research ideas.
AB - In the era of rapid technological advance, it is an important task for all researchers to keep up with trends when performing research. How to efficiently find suitable research topics while the number of papers is increasing rapidly is worthwhile to explore. To solve such problems, some researchers attempted to find research ideas by topic detection and tracking methods. However, these methods do not consider the users’ background knowledge and preferences, and they express a topic with general keywords, which does not effectively help researchers to develop new research ideas. Existing studies support that the title expresses the research idea the best. This study adapts this concept to propose an automatic title generation method that combines personalized recommendation methods and topic trend analysis methods to achieve this task. First, it uses hierarchical latent tree analysis to find the users’ interests for a topic structure and its representative keywords hidden in the existing research. Second, the interesting topic trends, popularity and user preferences in a hybrid recommendation method are considered. Finally, a natural language generation algorithm that is suitable for the titles of academic papers converts the original recommended-keywords into fluent title sentences that are designed for the users. Experiments have found that adding Google Trend indicators and personal factors can improve the performance of topic recommendations. The automatic title generation method using template-based and statistical information methods leads to excellent performances in both grammatical correctness and semantic expression. Moreover, for the users, the title is indeed more inspirational than the simple keywords for users to develop new research ideas.
UR - http://www.scopus.com/inward/record.url?scp=85074392607&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85074392607&partnerID=8YFLogxK
U2 - 10.1007/s11192-019-03258-x
DO - 10.1007/s11192-019-03258-x
M3 - Article
AN - SCOPUS:85074392607
SN - 0138-9130
VL - 121
SP - 1385
EP - 1406
JO - Scientometrics
JF - Scientometrics
IS - 3
ER -