Font Size: a A A

Chronological Citation Recommendation Technology Based On Information-need Shifting

Posted on:2016-07-10Degree:DoctorType:Dissertation
Country:ChinaCandidate:Z R JiangFull Text:PDF
GTID:1108330470970007Subject:Management Science and Engineering
Abstract/Summary:PDF Full Text Request
While the volume of publications has increased dramatically, there is an urgent need to assist researchers in locating the high-quality candidate-cited papers from a research repository. Citation recommendation methods are proposed for addressing this problem and attracting increasing attention in the fields of information retrieval and knowledge engineering. However, traditional scholarly recommendation approaches ignore the chronological nature of the citation recommendation task, which will lead to the misunderstanding of user information-need and result in the poor recommendation performance. This study proposes a novel method "Chronological Citation Recommendation (CCR)," which focuses on the dynamic change of the information-need in citation recommendation task.Chronological citation recommendation models the information-need shifting (InS) with three-level modeling. First level modeling achieves the goal of dynamic time-related information extraction and construction: â‘ for text information, this study employs dynamic topic modeling with supervised learning (DTM-SL) approaches to characterize the content "time-varying" dynamics in a topic level; â‘¡for citation relations, this study extracts the time-decay citation relation information. Second level modeling achieves the goal of heterogeneous information integration and dynamic recommendation feature construction, there are two solutions: â‘ dynamic topic/citation influence model (DTCIM), which is an iteration model that can generate time-series citation recommendation results in different time slices; â‘¡graph mining based on dynamic scientific information heterogeneous graph (DSIHG). Meta-path-based random walk algorithm is used to achieve the intergration of heterogeneous information and construction of the time-related dynamic feature. Third level modeling achieves the goal of dynamic recommendation feature intergration and feature weight optimization:this study uses information-need variant modeling based on Learning-to-Rank (InVM-L2R) approaches to train different models in different time slices, which model the information-need variants, optimize the feature weights and capture the time-evolving trajectory of feature weight. The learned models will generate the final "Chronological Citation Recommendation" rankings.Experiments on the sixty-year-spanning ACM corpus show that comparing with traditional scholarly recommendation approaches, chronological citation recommendation can significantly enhance the citation recommendation performance and effectively help users to get the required academic literature. CCR are currently used in an international research project, and it will be applied in the course environment of the American universities. CCR considers InS in citation recommendation research, which is an important theoretical innovation in research field; compared with the traditional citation recommendation methods, CCR has a significant advantage, which is an important approach innovation. For exploratory research, interdisciplinary research and academic review research, CCR also has the important theoretical and practical significance.
Keywords/Search Tags:Information-need Shifting, Citation Recommendation, Dynamic Topic Model, Supervised Learning, Scientific Heterogeneous Graph, Graph Mining, Learning-to-Rank
PDF Full Text Request
Related items