Topic Analysis Method And Its Application On Publication Management System

With the development of technology,the number of academic literatures increases by leaps and bounds,and new research topics are constantly emerging.How to classify,manage and analysis academic literatures effectively has great significance for both the researchers and the development of science and technology.Among the traditional research work on academic literatures,most focus on the aspects,such as theme analysis,social network analysis,while few is on the self-interest of the researchers.This paper started with the researchers,researched and analyzed the academic literatures by topic models and ensemble methods,finally delivered an academic literature management system for academic persons.The main contributions can be summarized as follows:First,pointing at the status that existing research works rarely consider the class-imbalance problem of academic literatures,propose an ensemble method to classify academic literatures with class-imbalance dataset.This method integrates topic model with ensemble learning method,rebuilds the training dataset by sampling with replacement and improves the performance of classify task of academic literatures by integrate several weak classifiers.We verified this method by experiments on real dataset.Second,pointing at the status that researchers must cost lots of time and energy to discover the hot points of academic conference or the trend of development,propose a method to learn the research hot points of the academic conferences and the trend of development by regarding the hot points as distribution of topic-words and observing the change of the research hot points and the trend of development with the relative entropy.We verified this method by the analysis for the top international conference NIPS of machine learning.Third,pointing the status that existing literature management system always show the wrong message,propose a method to obtain high confidence information automatically by extracting the basic information(title,author,etc.)from the academic literatures by machine learning and verifying to the network information got by the web crawler,finally to effective the research of researchers.Fourth,to complete the above-mentioned work and integrate the base function of literature management system,design and implement the academic literature management system-PubMS,to provide management,query functions for the academic literatures and fund projects.Furthermore,the system helps the researching with its study assistant function such as academic literatures recommendation and so on.This system has already been used in our lab.