| W2V-ATT-LSTM: 一种针对线上线下消费模式的新型顾客满意度预测模型 |
| 点此下载全文 |
| 引用本文：张金,高瑞玲,谭文安,毛文逸,侯英琦.W2V-ATT-LSTM: 一种针对线上线下消费模式的新型顾客满意度预测模型[J].上海第二工业大学(中文版),2024,41(3):289-297 |
| 摘要点击次数: 248 |
| 全文下载次数: 45 |
|
| 基金项目:国家自然科学基金(61672022), 国家自然科学基金(61272036) 资助 |
|
| 中文摘要:随着移动互联网+的广泛发展, 各行各业的线上线下电子商务模式(online to offline, O2O) 也应运而生, 然而,同质化竞争和数据价值挖掘不足的问题限制了市场的持续向好发展。聚焦于O2O 模式下的顾客满意度研究, 提出了一种新型的中文文本预测模型, 命名为W2V-ATT-LSTM。该模型引入Attention 机制以提高对重要文本的感知能力, 进一步融合W2V 和LSTM 结构, 深度挖掘头部企业真实交易数据进行分析处理、特征选择和模型训练。通过LDA 模型进行主题挖掘, 深入了解消费者对产品或服务的感受, 为企业提供有针对性的改进建议。实验结果显示,W2V-ATT-LSTM 模型在公开数据集任务中的准确率(91.4%)、精确率(82.2%)、召回率(81.7%) 和F1 (81.4%) 等指标均优于KNN、贝叶斯、决策树、SVM 等传统机器学习算法; 在爬虫真实数据集任务中的准确率(94%)、精确率(90%)、召回率(89%) 和F1 (89%) 也优于W2V、LSTM、Bi-LSTM 和Bert; 在多个公开中文情感分析数据集上的优越性能也表明W2V-ATT-LSTM 对于理解和处理自然语言文本具有显著的实际应用价值。在当前竞争激烈的O2O市场, W2V-ATT-LSTM 模型能为顾客与商家提供可靠的决策参考, 有望帮助企业更好地理解顾客需求, 提升服务水平, 推动行业良性发展。 |
| 中文关键词:W2V-ATT-LSTM 模型 O2O 模式 顾客满意度 深度学习 |
| |
| W2V-ATT-LSTM: A Novel Customer Satisfaction Prediction Model for Online-to-Offline Consumption Patterns |
|
|
| Abstract:With the widespread development of the mobile Internet, online to offline (O2O) e-commerce models have emerged across various industries. However, challenges such as homogenized competition and inadequate data value extraction have limited the market,s continuous positive growth. Customer satisfaction research is focused within the O2O model and a novel Chinese text prediction model
named W2V-ATT-LSTM is introduced. This model incorporates an Attention mechanism to enhance the perception of important texts and integrates the W2V and LSTM structures to deeply analyze, process, select features, and train on real transaction data from leading enterprises. Using the LDA model for topic mining, it delves into consumers’ sentiments regarding products or services, providing
targeted suggestions for improvement to businesses. Experimental results indicate that the W2V-ATT-LSTM model surpasses traditional machine learning algorithms such as KNN, Bayesian, decision tree, SVM in accuracy (91.4%), precision (82.2%), recall (81.7%), and F1 score (81.4%) on public datasets. It also outperforms W2V, LSTM, Bi-LSTM, and Bert in accuracy (94%), precision (90%), recall (89%), and F1 score (89%) on real web scraping datasets. The superior performance on multiple public Chinese sentiment analysis datasets also demonstrates the significant practical application value of W2V-ATT-LSTM in understanding and processing natural language texts. In today,s highly competitive O2O market, the W2V-ATT-LSTM model can provide reliable decision-making references for both customers and merchants, potentially helping businesses better understand customer needs, improve service levels, and promote healthy industry development. |
| keywords:W2V-ATT-LSTM model online to offline model customer satisfaction deep learning |
| 查看全文 查看/发表评论 下载PDF阅读器 |
|
|
|