基于词典-TextCNN-Word2Vec组合模型的在线评价细粒度情感分析  

A Fine-grained Sentiment Analysis of Online Reviews Based on the Dictionary-TextCNN-Word2Vec Combined Model

作　　者：惠调艳[1] 王智 何振华 秦春秀[1] Hui Tiaoyan;Wang Zhi;He Zhenhua;Qin Chunxiu(School of Economics and Management,Xidian University,Shanxi Xi’an 710126)

机构地区：[1]西安电子科技大学经济与管理学院,陕西西安710126

出　　处：《情报理论与实践》2025年第2期168-177,共10页Information Studies:Theory & Application

基　　金：国家自然科学基金青年科学基金项目“融合多模态UGC数据情感信息的旅游需求预测新方法研究”(项目编号:72201201);国家社会科学基金重点项目“场景驱动的我国关键核心领域文献资源精细组织与精准服务模式研究”(项目编号:22ATQ002)的成果。

摘　　要：[目的/意义]线上购物逐渐成为消费主流,在线情感评价成为消费者购买、厂商产品改进的重要决策依据。[方法/过程]深度挖掘商品显性和隐性属性特征,提出了融合词典-TextCNN-Word2Vec的在线评价细粒度情感分析模型。首先,利用Protégé软件和Pellet推理机推理等,构建了涵盖外观、硬件、软件、价格、质量、物流和服务7大主题维度的领域本体模型,并建立属性特征词典和情感词典;其次,针对三类在线评价,分别应用基于词典的显性属性情感分析模型、基于TextCNN的显性特征情感分类模型、基于Word2Vec的隐性特征情感分析模型,计算属性特征词的情感值;最后,通过词频加权法和熵权法,自下而上计算各层级主题属性的情感值,实现了多层次细粒度的情感挖掘。[结果/结论]综合基于词典、TextCNN和Word2Vec情感属性映射的三种模型的在线情感分析,显著提高了商品属性特征和情感分析的准确性,商品显性和隐性属性特征的总提取率高达93.77%,商品特征情感分析的加权平均准确率为86.78%。该组合模型为数字经济时代商品多属性特征的细粒度在线情感评价提供了创新研究方法。[Purpose/significance]Online shopping is increasingly becoming the mainstream mode of consumption,with online sentiment reviews serving as a crucial basis for consumers’purchasing decisions and manufacturers’product improvements.[Method/process]The explicit and implicit attribute features of products are deeply mined,and an online fine-grained sentiment analysis model is proposed,integrating a dictionary-based approach,TextCNN,and Word2Vec.First,using the Protégésoftware and Pellet reasoner,a domain ontology model covering seven thematic dimensions—appearance,hardware,software,price,quality,logistics,and service—was constructed,along with the establishment of an attribute feature dictionary and a sentiment dictionary.Second,for three types of online reviews,a dictionary-based explicit attribute sentiment analysis model,a TextCNNbased explicit feature sentiment classification model,and a Word2Vec-based implicit feature sentiment analysis model were applied to calculate the sentiment values of attribute feature words.Finally,using a frequency-weighted method and the entropy weight method,the sentiment values of theme attributes at different levels were calculated from the bottom up,achieving multi-level finegrained sentiment mining.[Result/conclusion]The integrated sentiment analysis,combining dictionary-based methods,TextCNN,and Word2Vec sentiment attribute mapping models,significantly enhanced the accuracy of product attribute feature extraction and sentiment analysis.The overall extraction rate of explicit and implicit product attributes reached 93.77%,with a weighted average accuracy of 86.78%for sentiment analysis of product features.This combined model offers an innovative approach for finegrained online sentiment evaluation of multi-attribute product features in the digital economy era.

关 键 词：细粒度情感分析 情感词典 TextCNN Word2Vec 

分 类 号：TP391.1[自动化与计算机技术—计算机应用技术]