Information Extraction from Research Papers based on Conditional Random Field Model

Abstract

With the increasing use of CiteSeer academic search engines, the accuracy of such systems has become more and more important. The paper adopts the improved particle swarm optimization algorithm for training conditional random field model and applies it into the research papers’ title and citation retrieval. The improved particl swarm optimization algorithm  brings the particle swarm aggregation to prevent particle swarm from being plunged into local convergence too early, and uses the linear inertia factor and learning factor to update particle rate. It can control algorithm in infinite iteration by the iteration between particle relative position change rate. The results of which using the standard research papers’ heads and references to evaluate the trained conditional random field model shows that compared with traditionally conditional random field model and Hidden Markov Model, the conditional random field model ,optimized and trained by improved particle swarm, has been better ameliorated in the aspect of F1 mean error and word error rate. DOI: http://dx.doi.org/10.11591/telkomnika.v11i3.2188