Recommendation in Scholarly Big Data Yuning Mao February

Recommendation in Scholarly Big Data Yuning Mao February 28, 2021

Page. 2 Academic Literature Recommendation Overlook ▪ 80+ approaches ▪ 21% were not evaluated ▪ Among the evaluated approaches ▪ 19% were not evaluated against a baseline ▪ Of the user studies performed, 60% had 15 or fewer participants or did not report on the number of participants ▪ Information on runtime and coverage was rarely provided

Page. 3 Recommendation in Acemap ▪ What is Acemap? Generally speaking, it’s an academic search system.

Page. 4 Recommendation in Acemap ▪ What is Acemap? generally speaking, it’s an academic search system. ▪ Challenges of RS in Acemap: Huge Sparsity Cold Start

Page. 5 #User vs #Item

Page. 6 “Coldest”Start ▪ New user problem

Page. 7 “Coldest”Start ▪ New item problem

Page. 8 Our Solution ▪ Collaborative Topic Modeling (KDD’ 11 best student paper award, used by New York Times!) ▪ LDA + CF ▪ LDA uses text to analyze content ▪ CF uses user preferences to assist

Page. 9 Latent Dirichlet Allocation Blei, David M. , Andrew Y. Ng, and Michael I. Jordan. "Latent dirichlet allocation. " the Journal of machine Learning research 3 (2003): 9931022.

Page. 10 Collaborative Filtering

Page. 11 Model Training CTR LDA

Page. 12 Datasets ▪ Citeulike a bibliography sharing website with tons of active users ▪ Microsoft Academic Graph (MAG) a heterogeneous graph containing scientific publication records, citation relationships between those publications, as well as authors, institutions, journals and conference "venues" and fields of study

Page. 13 Data Selection For Citeulike ▪ 5000+ users ▪ 16000+ papers with (crawled) abstract ▪ 200, 000+ user-item interactions For MAG ▪ 6, 900, 000+ with abstract ▪ 28, 000+ with abstract #citation > 50 since 2005 ▪ 390, 000+ with abstract #citation > 10 since 2000 ▪ 251 users with 46 user-item interactions…

Page. 14 Data Processing ▪ mysql. connector (huge efforts to data preprocessing) ▪ NLTK (stopword, word form, isdigit, is. English, tfidf…) ▪ LDA (from the original author) -> Light. LDA (from Microsoft research) ▪ User-item pair simulation according to Field of Study (Fo. S)

Page. 15 Experiments ▪ For citeulike

Page. 16 Experiments ▪ For real-world system Acemap (using MAG)

Page. 17 References ▪ Beel, Joeran, et al. "Research paper recommender system evaluation: a quantitative literature survey. " Proceedings of the International Workshop on Reproducibility and Replication in Recommender Systems Evaluation. ACM, 2013. ▪ Blei, David M. , Andrew Y. Ng, and Michael I. Jordan. "Latent dirichlet allocation. " the Journal of machine Learning research 3 (2003): 993 -1022. ▪ https: //en. wikipedia. org/wiki/Collaborative_filtering ▪ Wang, Chong, and David M. Blei. "Collaborative topic modeling for recommending scientific articles. " Proceedings of the 17 th ACM SIGKDD international conference on Knowledge discovery and data mining. ACM, 2011. ▪ http: //www. citeulike. org/ ▪ http: //research. microsoft. com/en-us/projects/mag/ ▪ https: //github. com/blei-lab/lda-c ▪ http: //research. microsoft. com/en-us/groups/ai/lightlda. aspx

Thanks for listening ！