The Instance Dynamic Generalization Based Coreference Resolution and Its Application

Coreference resolution is the core task of natural language processing, it is helpful for discourse analysis, automatic summarization, information extraction, information retrieval, information filtering and machine translation. In this paper, we improved the dynamic generalization of case-based method in a number of issues.Then we developed a variety of solutions for the generation of coreference chain from binary classification results. Based on the two parts of work before, we applied the instance-based dynamic generalization coreference mechanism in other natural language processing tasks, to help enhance the treatment effect of these problems.This paper focuses on three problems: the instance-based dynamic generalization coreference mechanism related issues; the generation of coreference chain from binary classification results; and the application of the instance-based dynamic generalization coreference mechanism in other natural language processing tasks.The work related with Instance-based dynamic generalization coreference mechanism focused on Mention identifying and accurate generalization points. We presented two kinds of methods based on classifier and sequence tagging to identify mentions in order to reduce levels of error propagation. In addition, to improve the quality of generic points, we introduced a new kind of generalization points named“accuracy generation points”. Therefore, this paper studies these two issues and put forward appropriate solutions in a variety of experiments on the corpus show the effectiveness of our work.Previous paper to solve the binary classification learning algorithm, the next thing is to deal with the generation scheme of coreference chain. This article discusses the three kinds of generation algorithms: the method based on lexical information; the method based on classification confidence level; Ranking based approach. The three options work differently for different occasions, the results achieved are also different. These three programs we carried out a detailed analysis and comparison to prove the effectiveness of our work, but the generation process is always limited by the processing of binary classification accuracy and is difficult to achieve a fundamental breakthrough.Based on the binary classification algorithm and the coreference chain generation method, we applied Instance-based dynamic generalization coreference mechanism to the global entity relation extraction tasks. This task is different from the traditional sentence-level relationship extraction and extracts relationships between all the entities in the document, not limited in the same sentence. Proved by the experiments in the dataset from field of music, the use of coreference resolutiomn is introduced the document level information, we can greatly improve the global entity relation extraction results.