Prof. Jung Jin Lee學術專題演講107/02/27

  • 2018-02-08
  • 楊 文敏

主講人:Prof. Jung Jin Lee (Dept. of Statistics and Actuarial Science, Soongsil University, Korea)
      目:A Maximum Entropy Approach for Modelling Term Dependencies in Text Retrieval
      間:民國107227 (星期二) 下午410 
             We propose an approach for modelling to find an optimal ordered list of document for a given set of documents and a user query. Documents and queries are represented as binary term incidence vectors. The Maximum Entropy Principle is applied to model term dependencies for document ranking according to the Probability Ranking Principle (PRP). The PRP states that document should be ranked by the decreasing probability of relevance to the user request by considering term dependence. The performance of the PRP is compared with the so-called Binary Independence Model (BIM) which is widely used for practical reason. Document ranking experiments are presented on data sets from the Microsoft LETOR collection.
Keywords: maximum entropy principle, probability ranking principle, text retrieval