Title: Vectorization of Users’ HistoricalRatings for Recommendation
Abstract: As a class of effectivevectorization methods, neural embedding's exploration on users' historicalratings in recommender systems is still very limited. In this talk, we firstlydiscuss state-of-the-art vectorization methods with ratings for recommendationtasks. We then present our recent work, Preference2Vec, which considers users'pairwise item preferences as vectorization units. It vectorizes the pairwisepreferences by maximizing the likelihood estimation of the conditionalprobability of each pairwise item preference given another one. Then thevectors of users and items can be generated easily. We comprehensively designthree experimental tasks on benchmark recommendation datasets to evaluate ourPreference2Vec method. The first experiment utilizes the movie genres asground-truths and directly assesses the quality of item (movies) vectors. Thesecond one evaluates the initialization independence of the user and itemvectors. The third experiment evaluates the recommendation performance based onour vectorization results.Our experimental results show significant improvementover state-of-art baselines.
Bio: Shuaiqiang Wang is now a Lecturer inInformation Management (equivalent to Assistant Professor in USA) at AllianceManchester Business School, the University of Manchester in United Kingdom. Hereceived Ph.D. and B.Sc. in Computer Science from Shandong University, China,in 2009 and 2004 respectively. During 2009, he visited Hong Kong BaptistUniversity as an exchange doctoral student. Before joined in the University ofManchester, he was an Assistant Professor at University of Jyväskylä in Finlandfrom 2014 to 2017, an Associate Professor at Shandong University of Finance andEconomics in China from 2011 to 2014, and a postdoctoral research associate atTexas State University in USA from 2010 to 2011. His researchinterests includerecommender systems, information retrieval and data mining. He has publishedmore than 40 papers in leading conferences like SIGIR, AAAI and CIKM, andjournals like TOIS, TKDE and TIST. He served as a PC member for anumber ofconferences like SIGIR, IJCAI and CIKM, and a reviewer for journalslike TOIS,AIJ, TWEB and TEVC. The detailed information can be found from his homepage http://personalpages.manchester.ac.uk/staff