Evidence Document Recognition Chinese Experts

Expert Evidence Document Recognition (EEDR) is a hot area of current research. The concept of EEDR can be defined as follows:Given the name and organization of an expert, certain categories of expert evidence document, such as Expert Homepage, Expert Academic Resources page, Expert Wiki page, Expert Blog page, should be labeled to candidate expert evidence document accurately and quickly. This paper, focuses on the issue of EEDR, has done a series of studies and discussion in Expert Homepage Recognition (EHR), Expert Evidence Document Chinese Name Entity Recognition (CNER), and Undirected Graph Model for expert evidence document recognition and other technologies. The main conclusions are as follows:1. Proposes an Expert Homepage Recognition (EHR) method based on Markov logic networks. Experts Homepage is one of the most important resources in expert evidence document, and, EHR is also the foundation of expert evidence document recognition. Most of existing methods for EHR does not take into account the relationships between candidate expert homepages. This paper, analysis the word, link type, link recall features of individual candidate expert homepages and the relationship between pages firstly, and then integrates individual candidate expert homepages features and relationships between pages into Markov Logic Networks, takes advantage of effective methods of learning and inference to conduct EHR.2. Proposes an Expert Evidence Document Chinese Name Entity Recognition (CNER) method integrated of long distance dependence features. Accurately identify the named entity in expert evidence documents is the basis of judging the content relationships between documents which could directly affect the accuracy of Expert Evidence Document Recognition. For the issue that existing method for Chinese Name Entity Recognition usually fail to consider the long distance features, this paper, comprehensive utilize candidate entity independent features (word, part of speech, adjacent words), short-range dependent features (label of adjacent word), as well as long-range dependent features, propose an expert Evidence Document Chinese Name Entity Recognition method based on Markov Logic Networks which utilize first order formula to represent these three kind of features and integrate these three kind of features into Markov network by Markov Logic networks to conduct Chinese Name Entity Recognition.3. Proposes a method for expert evidence document recognition based on Undirected Graph Model. Most of existing methods for EEDR focus on the study of EHR, without considering other types of expert evidence document, which cannot meet the demand of the expert resources for Expert retrieval system; moreover, Expert evidence documents often contain a wealth of relationships. This paper, firstly, comprehensively utilize candidate expert evidence documents independent features (word, link, structured data) and the relationships between pages (links and content relationship); secondly, integrate these features into undirected graphic model, create undirected graphic model for Expert Evidence Documents Recognition, then, use gradient descent method to learn the weights of features, use Gibbs sampling methods for expert evidence document recognition.4. Make use of the above research results, extract the features of individual candidate expert evidence documents independent features and the relationships between pages, construct expert evidence document recognition experiment platform and achieve the expert evidence document recognition prototype system.