Knowledge elements are the basic unit of knowledge. Extracting knowledge elements from literatures enables granular knowledge organization and retrieval. This paper proposes a rule‐based framework for extracting method Knowledge Elements (KEs) in scientific literature aiming to improve the accuracy and complete extraction of method KEs. The method is divided into two stages: semi‐automated extraction of initial description rules of method KEs and automated derivation of additional description rules. In a preliminary evaluation on 415 papers, the precision and recall for the method KEs extraction are 0.77 and 0.84, respectively, indicating the effectiveness of our method.
This paper describes our approach to the CL-SciSumm 2020 shared task toward the problem of identifying reference span of the citing article in the referred article. In Task 1a, we apply and compare different methods in combination with similarity scores to identify spans of the reference text for the given citance. In Task 1b, we use a logistic regression to classifying the discourse facets.
Abstract. Social-based recommendation and collaborative filtering-based recommendation have their own characteristics. Considering that traditional collaborative filtering only makes use of users' behavior data but ignores users' social relationships, a recommendation algorithm combined with social and collaborative filtering was proposed in this paper. Traditional item-based collaborative filtering algorithm was improved first, and then the hybrid recommendation algorithm was constructed by considering the complementarity of users' behavior data and social relationships, which can relieve the existing problems of collaborative filtering such as data sparse and cold start and is proved to improve the accuracy of the recommendation though experiment. IntroductionTraditional personalized recommendation technology is mainly calculating the users' preferences for items by mining users' behavior data and using user's existing preferences for items to predict items which user may like, then the items is recommended to target user. The current mainstream recommendation algorithm is collaborative filtering algorithm. Although collaborative filtering algorithm has been widely used, but there are still two major issues of data sparse and cold start. In order to solve these problems, many scholars have put forward solutions. Yu-fang Zhang et al [1] proposed a two-step filling matrix method combined with conditional probability algorithm, which can partly relieve data sparse and improve the effectiveness of the recommendation. Yang Yang et al [2] proposed a recommendation algorithm based on matrix factorization and user nearest neighbor model, which can effectively improve the accuracy of the prediction score. Shu-chao Ma [3] proposed a graph search algorithm based on two layer graph model, which can solve the problem of data sparsity and scalability. Sotirios P.Chatzis [4] proposed a dynamic Bayesian probability matrix decomposition model, which can reflect that users' behaviors change with time and is proved to have some certain superiority through experiment.The above methods all can effectively relieve the data sparsity and partly improve recommender efficiency of traditional collaborative filtering. With the rise of the social Ecommerce model, personalized recommendation combined with social network becomes a new research direction of the recommendation system. Previous studies have shown that recommendation combined with social information can effectively improve the effect of personalized recommendation [5][6]. Collaborative filtering-based recommendation reflect the continuation of the users' own interests or the preferences of the people with similar interests, while social-based recommendation can simulate the real society, which reflect the trust of friends. Therefore, combining collaborative filtering-based recommendation with social-based recommendation becomes a hot spot in current research. Ma et al [7] proposed a factor model based on matrix factorization to alleviate the data sparsity and poor prediction ac...
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.