张逸凯_171840708_report1
需积分: 0 86 浏览量
2022-08-03
11:51:01
上传
评论
收藏 1.58MB PDF 举报
为短文本推荐合适的emoji — 基于上下词及语境
的CBOW多层神经网络分类模型研究
张逸凯
1
1
南京大学, 学号: 171840708, 年级: 大二
Abstract Aiming at the feature sparsity and context dependence of short texts, this paper proposes a
short text classification method based on CBOW multilayer neural network. Using the sentiment
tendency of short text itself, this paper adopts TF-IDF-CF characterization method, on the other
hand, associates the short text context with the CBOW model to optimize the emoji of the chat
data. Experiments show that this hybrid model is higher in classification performance better
than the traditional Bayesian or SVM classification model. In the Kaggle competition private
leaderboard it achieved a higher accuracy of 0.17554.
Key words Short text classification CBOW model Improved weight TF-IDF-CF method Multi-layer
neural network classification Data Mining
摘要 针对短文本的特征稀疏性和上下文依赖性等特点, 本文提出一种基于CBOW多层神经网
络的短文本分类方 法. 利用短文本本身的情感倾向, 一方面采用改进权值的TF-IDF-CF特
征化方法, 另一方面关联短文本上下文语境运用CBOW模型对聊天数据进行最优emoji推
荐. 实验表明这种混合模型在分类性能上比传统的贝叶斯或支持向量机分类模型高出很
多, 在Kaggle竞赛private榜上取得0.17554的较高准确率.
1
剩余14页未读，继续阅读
评论0