Search

Nowadays, academic materials such as articles, patents, lecture notes, and observation records often use both texts and images (i.e., dual-modal data) to illustrate scientific issues. Measuring the similarity of such dual-modal academic data largely depends on dual-modal features, which is far from satisfying in practice. To learn dual-modal feature representation, most current approaches mine interactions between texts and images on top of their fusion networks. This work proposes a multi-fusion deep learning framework that learns semantically richer dual-modal representations. The framework designs multiple fusion points in the feature space of various levels, and gradually integrates the fusion information from the low-level to the high-level. In addition, we develop a multi-channel decoding network with alternate fine-tuning strategies to mine modal-specific features and cross-modal correlations thoroughly. To our knowledge, this is the first work to bring forward deep learning functions for dual-modal academic data. It reduces the semantic and statistical attribute differences between two modalities, thereby learning robust representations. A large number of experiments conducted on real-world data sets show that our method has significant performance compared with state-of-the-art approaches.

With the rapid increase of publishable research articles and manuscripts, the pressure to find reviewers often overwhelms the journal editors. This paper incorporates the major entity level metrics found in the heterogeneous publication networks into a pattern mining process in order to recommend academic reviewers and potential research collaborators. In essence, the paper integrates authors' h-index and papers' citation count and proposes a quantification to account for the author diversity into one formula duped impact to measure the real influence of a scientific paper. Thereafter, this paper formulates two kinds of target patterns and mines them harnessing the high-utility itemset mining (HUIM) framework. The first pattern, researcher-general topic patterns (RGP), is a pattern that includes only researchers; whereas, the researcher-specific topic patterns (RSP) is comprised of combinations of researchers and keywords that summarize their niche of expertise. The HUI algorithms of Two Phase, IHUP, UP-Growth, FHM, FHN, HUINIV-Mine, D2HUP, and EFIM were compared on two real-world citation datasets related to Deep Learning and HUIM, in addition to the open source mushroom dataset. The EFIM algorithm showed good performance in terms of run time and memory usage. Consequently, it was then used to mine the patterns within the proposed framework. The discovered patterns of RGP and RSP showed high coverage, proving the efficiency of the proposed framework.

In the context of big scholarly data, various metrics and indicators have been widely applied to evaluate the impact of scholars from different perspectives, such as publication counts, citations, h -index, and their variants. However, these indicators have limited capacity in characterizing prospective impacts or achievements of scholars. To solve this problem, we propose the Academic Potential Index (API) to quantify scholar's academic potential. Furthermore, an algorithm is devised to calculate the value of API. It should be noted that API is a dynamic index throughout scholar's academic career. By applying API to rank scholars, we can identify scholars who show their academic potentials during the early academic careers. With extensive experiments conducted based on the Microsoft Academic Graph dataset, it can be found that the proposed index evaluates scholars' academic potentials effectively and captures the variation tendency of their academic impacts. Besides, we also apply this index to identify rising stars in academia. Experimental results show that the proposed API can achieve superior performance in identifying potential scholars compared with three baseline methods.

The prediction of scholars' scientific impact plays a significant role in accelerating the advancement of science, such as providing basis for the noble prizes, predicting the future influential scholars or research trends, offering tenures for researchers, and selecting promising candidates for research funding. Therefore, the study on scientific impact is of great significance and has drawn increasing interests. However, most current literature on predicting the impact of scholars neglect several vital facts, which are the time evolvement of academic networks, the distinct dynamics of different scholars' impact, and the mutual influence among different scholarly entities. Inspired by the above-mentioned facts, we propose the PePSI solution for personalized prediction of scholars' scientific impact. Our method primarily classifies scholars into different types according to their citation dynamics. For different scholars, we apply modified random walk algorithms to predict their impact in heterogeneous temporal academic networks with different time functions to capture the time-varying feature of academic networks. Experimental results on real data set demonstrate the effectiveness of PePSI in predicting top scholars and the overall impact of scholars with a rather short-term academic information as compared with the state-of-the-art prediction methods.

Computer sciences and related disciplines evolve around developing, evaluating, and applying algorithms. Typically, an algorithm is not developed from scratch, but uses and builds upon existing ones, which often are proposed and published in scholarly articles. The ability to capture this evolution relationship among these algorithms in scientific literature would not only allow us to understand how a particular algorithm is composed, but also shed light on large-scale analysis of algorithmic evolution through different temporal spans and thematic scales. We propose to capture such evolution relationship between two algorithms by investigating the knowledge represented in citation contexts, where authors explain how cited algorithms are used in their works. A set of heterogeneous ensemble machine-learning methods is proposed, where the combination of two base classifiers trained with heterogeneous feature types is used to automatically identify the algorithm usage relationship. The proposed heterogeneous ensemble methods achieve the best average F1 of 0.749 and 0.905 for fine-grained and binary algorithm citation function classification, respectively. The success of this study will allow us to generate a large-scale algorithm citation network from a collection of scholarly documents representing multiple time spans, venues, and fields of study. Such a network will be used as an instrument not only to answer critical questions in algorithm search, such as identifying the most influential and generalizable algorithms, but also to study the evolution of algorithmic development and trends over time. [ABSTRACT FROM AUTHOR]