基于深度神经网络的微博图书名识别研究  

A Deep Neural Network for Book Title Identification in Microblog

作　　者：朱娜娜[1] 景东[2] 薛涵[3] 

机构地区：[1]哈尔滨学院图书馆,哈尔滨150001 [2]哈尔滨工业大学机电工程学院,哈尔滨150001 [3]哈尔滨工程大学图书馆,哈尔滨150001

出　　处：《图书情报工作》2016年第4期102-106,141,共6页Library and Information Service

基　　金：国家社会科学基金项目"社交媒体突发公共事件的协同应急机制研究"(项目编号:14CXW045);教育部人文社会科学研究项目"微博突发公共事件传播路径的实时分析及其趋势预测"(项目编号:13YJC860013)研究成果之一

摘　　要：[目的 /意义]微博作为一种新兴的社交媒体平台,被互联网用户广泛关注。微博数据中包含着大量的用户信息、用户行为及用户生成内容,基于微博内容自动识别图书名有利于分析用户阅读兴趣、收集用户对图书的评价和挖掘图书相关知识。[方法 /过程]基于微博的数据特点,提出一种基于深度神经网络的表示学习方法,利用微博中候选图书名的上下文连续向量化表示,实现微博内容中的图书名自动识别。[结果 /结论]实验结果表明,所提出的方法显著优于传统基于特征工程的有指导机器学习方法,并达到91.92%的精确率。[ Purpose/significance] As a blooming social media platform, microblog has received extensive attention by the web users. The microblog data include massive user profile, user behavior and user generated content. Automatic identification of book title in microblog contributes to analysis of user interests and data mining of books. [ Method/ process ] Based on the features of the microblog data, in this paper, we proposed a deep neural network approach to identify the book title in the microblog which is posted by users. [ Result/conclusion ] The experimental results show that the proposed approach significantly outperforms the traditional supervised learning approaches which are based on the feature engineering and the accuracy reaches 91.92%.

关 键 词：图书名识别 神经网络 深度学习 微博 

分 类 号：G206[文化科学—传播学]