A brain-inspired information processing algorithm and its application in text classification

摘要

Cognitive scientists believe that the human brain is constantly predicting the information it is going to receive, and this prediction ability is acquired based on the experiences gained from previous information it received over its lifetime. Inspired by this brain-behavior, we propose a new information processing algorithm with the building blocks named as boxes and routes. The function of boxes is to store the learned information and the function of routes is to represent the relationship between information. The novel algorithm features generality and objectivity. It imitates the mechanism of the human brain and has functions such as information learning, comparison, prediction, and forgetting. It also has advantages in dealing with continuous time series data by using routes and high-order boxes. The algorithm is self-adaptive and unsupervised. It does not need manual intervention information to train the model. It can learn useful information from undefined data and subsequently construct a hierarchical network corresponding to the characteristics of the input information, which can be used for classification or prediction. To prove the validity of this new algorithm, a classifier is constructed based on the hierarchical network to do text classification. We select a collection of Chinese literature from 30 litterateurs as samples to train the classifier. For 10 classes situation, the optimal average accuracy of classification reaches 79.5%, which outperforms other approaches commonly used in literatures, verifying the effectiveness of the proposed algorithm.