Contextual rule-based feature engineering for author-paper identification

Abstract

We present the ideas and methodologies that we used to address the KDD Cup 2013 challenge on author-paper identification. We firstly formulate the problem as a personalized ranking task and then propose to solve the task through a supervised learning framework. The key point is to eliminate those incorrectly assigned papers of a given author based on existing records. We choose Gradient Boosted Tree as our main classifier. Through our exploration we conclude that the most critical factor to achieve our results is the effective feature engineering. In this paper, we formulate this process as a unified framework that constructs features based on contextual information and combines machine learning techniques with human intelligence. Besides this, we suggest several strategies to parse authors' names, which improve the prediction results significantly. Divide-conquer based model building as well as the model averaging techniques also benefit the prediction precision.

Microsoft Academic Graph: When experts are not enough

“…Being able to identify the real-world person based on the name appearing on a publication is a highly desirable feature but also a technically challenging problem. Author name disambiguation has drawn intensive research interests (Cock et al, 2013;Kanani, McCallum, & Pal, 2007;Li et al, 2015;Liu, Lei, Liu, Wang, & Han, 2013;Roy et al, 2013;Wick, Kobren, & McCallum, 2013;Zhang, Xinhua, Huang, & Yang, 2019;Zhang, Zhang, Yao, & Tang, 2018;Zhong et al, 2013), yet the state-of-the-art techniques, using only information in the publication data such as coauthorship, affiliations, and topics, typically do not yield high enough accuracy, especially for Asian or popular Western names. The reward of using these machine learning techniques is not high enough, so most systems have just used a simple name key (e.g., the author's last name prepended with first or middle initials, as in Google Scholar) to associate author names with publication clusters.…”

Task-Guided and Path-Augmented Heterogeneous Network Embedding for Author Identification

“…• Supervised feature-based baselines. As widely used in similar author identification/disambiguation problems [12,13,34,9,33], this thread of methods first extract features for each pair of training data, and then applies supervised learning algorithm to learn some ranking/classification functions. Following them, we extract 20+ related features for each pair of paper and author in the training set (details can be found in appendix).…”

“…The problem of author identification has been briefly studied before [11]. And we also notice KDD Cup 2013 has similar author identification/disambiguation problem [12,13,34,9,33], where participants are asked to predict which paper is truly written by some author. However, different from the KDD Cup, our setting is different from them in the sense that (1) existing authors are unknown in our double-blind setting, and (2) we consider the reference of the paper, which is one of the most important sources of information.…”

“…Unlike traditional supervised learning, dense vectorized representations [16,15] are not directly available in networked data [26]. Hence, many traditional methods under network settings heavily rely on problem specific feature engineering [12,13,34,9,33].…”

“…Unlike traditional supervised learning, dense vectorized representations [16,15] are not directly available in networked data [26]. Hence, many traditional methods under network settings heavily rely on problem specific feature engineering [12,13,34,9,33].Although feature engineering can incorporate prior knowledge of the problem and network structure, usually it is time-consuming, problem specific (thus not transferable), and the extracted features may be too simple for complicated data sets [3]. Several network embedding methods [17,26,25] have been proposed to automatically learn feature representations for networked data.…”