Bing Liu (born 1963) is a Chinese-American professor of computer science who specializes in data mining, machine learning, and natural language processing. In 2002, he became a scholar at University of Illinois at Chicago.[1] He holds a PhD from the University of Edinburgh (1988).[2][3] His PhD advisors were Austin Tate and Kenneth Williamson Currie, and his PhD thesis was titled Reinforcement Planning for Resource Allocation and Constraint Satisfaction.[4]
He developed a mathematical model that can reveal fake advertising.[5] Also, he teaches the course "Data Mining" during the Fall and Spring semesters at UIC. The course usually involves a project and various quiz/examinations as grading criteria.
He is best known for his research on sentiment analysis (also called opinion mining), fake/deceptive opinion detection, and using association rules for prediction. He also made important contributions to learning from positive and unlabeled examples (or PU learning), Web data extraction, and interestingness in data mining.
Two of his research papers published in KDD-1998 and KDD-2004 received KDD Test-of-Time awards in 2014 and 2015. In 2013, he was elected chair of SIGKDD, ACM Special Interest Group on Knowledge Discovery and Data Mining.
Association rule-based classification takes into account the relationships between each item in a dataset and the class into which one is trying to classify that item.[6] The basis is that there are two classes, a positive class and a negative class, into which one classifies items.[6] Some classification algorithms only check if a case/item is in the positive class, without understanding how much exactly the probability of it being in that class is.[6] Liu and his collaborators described a new association rule-based classification algorithm that takes into account the relationship between items and the positive and negative classes.[6] Each item is given a probability or scoring of being in the positive class or the negative class. It then ranks the items as per which ones would be most likely to be in the positive class.[6]
In a paper that Liu collaborated on, "Opinion Word Expansion and Target Extraction through Double Propagation", Qiu, Liu, Bu and Chen studied the relationship between opinion lexicons and opinion targets.[7] Opinion lexicons are word sets and opinion targets are topics on which there is an opinion.[7] The authors of that paper discuss how their algorithm uses a limited opinion word set with the topic and through double propagation, one is able to form a more detailed opinion word set on a set of sentences. Double propagation is the back and forth functional process between the word set and topic as the word set updates itself.[7] Some algorithms require set rules and thus are limited in what they can actually do and in what service they provide through updated opinion lists.[7] Their algorithm only requires an initial word set, which is updated through finding relations between the words in the set and the target word or vice versa.[7] The algorithm is done on a word population such as a set of sentences or a paragraph.[7]