Method for classifying problems on basis of PGM (probabilistic graph models)

Abstract

The invention discloses a method for classifying problems on the basis of PGM (probabilistic graph models). The method includes a modeling phase and an inferring phase. The first phase includes manually classifying training datasets; substituting the classified labeled datasets into the probabilistic graph models; constructing network structures of directed acrylic nets; computing prior probability and conditional probability of various observation nodes to obtain conditional probability distribution of models. The second phase includes carrying out Bayesian inference on the basis of Gibbs algorithms according to existing network structures and the CPD (conditional probability distribution) to obtain categories of the problems. Compared with existing algorithms for classifying problems, the method has the advantages that the probabilistic graph models are built, the models are trained by the aid of training data, the problems are classified by the aid of trained models, and accordingly the method not only has the characteristic of high interpretability of processes for classifying problems on the basis of rules, but also has merits of independence from expert knowledge and automatic learning of processes for classifying problems on the basis of machine learning.

Description

technical field [0001] The invention relates to a classification method, in particular to a problem classification method based on PGM, and belongs to the field of computer software. Background technique [0002] With the rapid development of information technology, the form of information retrieval has developed from the original keyword retrieval to the retrieval based on the form of question and answer. Natural language is used as input, and according to certain rules, possible answers to questions raised by users are extracted from large-scale document collections. The question answering system specifically involves three parts: question classification, information retrieval, and answer extraction. Question classification is responsible for limiting the answer space and selecting answers. Strategy; information retrieval searches for possible results in the document collection according to the keywords in the question; answer extraction is based on the limitation of quest...

AI Technical Summary

Problems solved by technology

[0004] The rule-based problem classification method uses expert knowledge to preset a set of rules for each problem type. When the problem to be solved meets these rules, it is determined that the problem belongs to this category. This method has high accuracy, pertinence, and explanatory Strong, but the main disadvantage is that the manual establishment and maintenance of the rule base is a lot of work, relying on expert knowledge, and poor flexibility

Images