Which is More Influential, “Who” or “When” for a User to Rate in Online Review Site?

Thời gian: 14:00 đến 16:00 Ngày 16/07/2018

Địa điểm: B4-705, VIASM

Báo cáo viên: Hiroshi Motoda (Osaka University)

Tóm tắt:

I start with often discussed difficult problems in social networkanalysis and how we challenge these problems without technicaldetails. These include computation of influence degree in informationdiffusion in social network, learning diffusion probability,identifying influential people, super mediators and critical links,and detecting changes. Then, I focus on one of the problems which isidentifying influential users in online review systems.

At its heart the act of reviewing is very subjective, but in realitymany factors influence user's decision. This can be called socialinfluence bias. We pick two factors, ``Who'' and ``When'' and discusswhich factor is more influential when a user posts his/her own rateafter reading the past review scores in an online review system. Weshow that a simple model can learn the factor metric quite efficientlyfrom a vast amount of data that is available in many online reviewsystems and clarify that there is no universal solution and theinfluential factor depends on each dataset. We use a weightedmultinomial generative model that takes account of each user'sinfluence over other users. We consider two kinds of users: real andvirtual, in accordance with the two factors, and assign an influencemetric to each. In the former each user has its own metric, but in thelatter the metric is assigned to the order of review posting actions(rating).  Both metrics are learnable quite efficiently with a fewtens of iterations by log-likelihood maximization. Goodness of metricis evaluated by the generalization capability. The proposed method wasevaluated and confirmed effective by five review datasets. Differentdatasets give different results. Some dataset clearly indicates thatuser influence is more dominant than the order influence while theresults are the other way around for some other dataset, and yet otherdataset indicates that both factors are not relevant. The third oneindicates that the decision is very subjective, i.e., independent ofothers' review. We tried to characterize the datasets, but were onlypartially successful. For datasets where user influence is dominant,we often observe that high metric users have strong positivecorrelations with three more basic metrics: 1) the number of reviews auser made, 2) the number of the user's followers who rate the sameitem, 3) the fraction of the user's followers who gave the similarrate, but this is not always true. We also observe that the majorityof users is normal (average) and there are two small groups of users,each with high metric value and low metric value. Early adopters arenot necessarily influential.

Bio:

Hiroshi Motoda is Professor Emeritus of Osaka University, GuestProfessor of the Institute of Scientific and Industrial Research(ISIR) of Osaka University. He was a scientific advisor at AFOSR/AOARD(Asian Office of Aerospace Research and Development, Air Force Officeof Scientific Research, US Air Force Research Laboratory) since 2006till 2018, and worked as an international program officer. Beforethat, he was a professor in the division of Intelligent SystemsScience at the Institute of Scientific and Industrial Research ofOsaka University since 1996 until March, 2006. Before joining theuniversity, he had been with Hitachi since 1967. At Hitachi heparticipated in research on nuclear reactor core management,diagnosis, control and design until 1985, and then moved to the fieldof machine learning, knowledge acquisition, qualitative reasoning anddiagrammatic reasoning. After joining the university, he extended hisresearch to scientific knowledge discovery and data mining. At AOARDhe worked on social network analysis while managing several basicresearch projects on Computational Intelligence. He received his Bs,MSc and PhD degrees, all in nuclear engineering from the University ofTokyo. He is now an honorary member of the steering committee of AsianConference on Machine Learning (ACML), an honorary member of thesteering committee of Pacific Rim International Conference ofArtificial Intelligence (PRICAI), a life long member of the steeringcommittee of Pacific Asian Conference of Knowledge Discovery and DataMining (PAKDD), and a steering committee member of IEEE InternationalConference on Data Science and Advanced Analytics (DSAA). He receivedthe best paper awards twice from Atomic Energy Society of Japan andthree times from Japanese Society for Artificial Intelligence (JSAI),the outstanding achievement awards from JSAI, the distinguishedcontribution award from PAKDD and PRICAI, and the outstandingcontribution award from Web Intelligence Consortium. He wrote/editedfour books on feature selection/extraction/construction. His book``Fundamentals of Data Mining'' was awarded the 2007 Okawa PublishingPrize. He is a fellow of JSAI.