基于注意力机制的在线用户痛点信息挖掘

Online User Pain Point Information Mining Based on Attention Mechanism

[目的/意义]在用户痛点信息破碎化的背景下,及时准确了解客户对特定产品和服务的要求和体验诉求,把握用户对产品和服务主要关注度和问题所在,对企业克服产品和服务缺陷,改善用户产品体验,满足用户客观需求具有重要的现实意义。[方法/过程]利用条件随机场CRF序列标注和GloVe词向量编码,注意力机制和VGG-19构建基于文本、图像及双特征融合的用户痛点提取和分析模型,建立用户痛点量化与计算,实现基于注意力机制的在线用户痛点信息挖掘。[结果/结论]对OPPO Find X5 Pro手机在线评论用户痛点信息进行挖掘,该模型有效提取了文本、语音和图像中的用户情感特征,建立了用户情感词三个类别的分类和量化,得到用户痛点计算得分,构建了OPPO Find X5 Pro手机用户痛点5个等级划分。该模型计算准确度比现有典型模型平均高13.49%。

[Purpose/significance]Under the background of users’pain point information fragmentation,it is of great practical significance to timely and accurately understand customers’requirements and experience demands for specific products and services.This can grasp the main concerns and problems of users on products and services,and enterprises can overcome product and service defects,improve user product experience,and meet the objective needs of users.[Method/process]Conditional random field(CRF)sequence annotation,word vector coding with Global Vectors for Word Representation(GloVe),attention mechanism and VGG-19 are used to construct user pain point extraction and analysis model based on text,image and image-text fusion,establish quantitation and calculation of user pain points,and realize of online user pain point information mining based on attention mechanism.[Result/conclusion]OPPO Find X5 Pro mobile phone online comments users’pain point information is mined.This model effectively extracts the user emotional features in text,language and image,establishes the classification and quantification of three categories of user emotional words,and obtains the user pain point calculation score.Construct OPPO Find X5 Pro mobile phone user pain point five levels.The calculation accuracy of this model is 13.49%higher than that of existing typical models.