Research And Implementation Of Fine-grained Text Topic Detection Technology

With the rapid development and application of network technology,users' demand for information discrimination is also increasing.The theme discrimination is no longer satisfied with the users' demand for information on the text.In many practical applications,it is necessary to classify the text at multiple levels,that is,to determine whether a text belongs to a large category or not.It is further determined whether it belongs to a smaller subclass until it is subdivided into the finest grained category.Fine-grained text topic detection is a technique for judging whether a new text data belongs to a particular topic and then deciding which subtopic to belong to.Based on this,people can not only pick out the topic information you want to pay attention to from a lot of network information,but also get the subtopic information of this topic information and the proportion of each subtopic.This paper uses two methods to implement this technique.One is to use the method of multiple classification,the other is to use the method of multi-layer classification.Fine-graded text topic detection of multiple classification using multiple classification have three parts,including the specific subject judgment of the new text part using the method of one class classification,subtopic creation of a specific topic part using the method of clustering and subtopic identification of new text part using the method of classification.This paper selects a clustering method through a lot of experiments,adjusts the cluster according to the task,and experiments show it's effectiveness.The Fine-graded text topic detection of multi-layer classification bulid a new model for multi-layer classification.The new model based on the classification model Hierarchical softmax,instead of using a network to conduct a classification,but combines multiple classifications into a Hierarchical softmax model,using a network for multi-layer classification.This model is called Multi-layer classification model based on Hierarchical softmax.The model not only simplifies the process of Multi-layer classification,but also have a good performance.