Scholarlink

<p>In the busy life and schedule of an individual, online shopping gaining popularity. A large variety of brands, price ranges, and discounts are also attracting people to online purchasing. But, the quality of a product or a new brand is always a challenge for a customer. The customer is completely dependent on the product, product reviews, and brand rating provided by the shopping site. This product and brand rating is the analytical reflection of the reviews collected from customers and experts. A customer submits the reviews in textual form to express his thoughts, opinions, and feelings about the benefit or loss that he faced with that deal. On a shopping or emarket site, thousands are reviews are submitted by the customer and it is impossible to analyze each review separately. Such sites can use a customer satisfaction analysis algorithm to precisely measure the supplied evaluations. Using this customer satisfaction analysis, we may find out if the product is well-received by the buyer. To efficiently quantify product reviews, this research proposes a unique HAF-wBiLSTM (Handcrafted and Automated Features-driven weighted-BiLSTM) model. Combining handmade features with BoW features generates a large featureset in the proposed deep learning model. Feature criticality weight, frequency feature weights, filtered extraction of 1 and 2&#xa0;g, and other features are included in this set. Weighted and bidirectional long short-term memory (LSTM) models analyze this augmented collection of information. We check the correctness and dependability of the suggested model. This analytical evaluation is carried out in comparison to CNN, LSTM, Tree-LSTM, FL-LSTM, and SVM models. The accuracy increases asserted by the suggested model were as follows: 15.28% vs SVM, 9.46% versus LSTM, 4.4% versus BiLSTM, 12.42% versus CNN, 9.77% versus Tree-LSTM, 17.24% versus RNN, and 2.18% versus LSTM + FL. Similarly, when compared to state-of-the-art approaches, the findings confirmed that the suggested HAF-wBiLSTM model had better reliability and performance.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Design of a contextual and dependent features-based HAF-wBiLSTM model for predicting customer satisfaction

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年2月13日

查看全文
  • Priya Darshini,
  • Hardayal Singh Shekhawat

摘要

In the busy life and schedule of an individual, online shopping gaining popularity. A large variety of brands, price ranges, and discounts are also attracting people to online purchasing. But, the quality of a product or a new brand is always a challenge for a customer. The customer is completely dependent on the product, product reviews, and brand rating provided by the shopping site. This product and brand rating is the analytical reflection of the reviews collected from customers and experts. A customer submits the reviews in textual form to express his thoughts, opinions, and feelings about the benefit or loss that he faced with that deal. On a shopping or emarket site, thousands are reviews are submitted by the customer and it is impossible to analyze each review separately. Such sites can use a customer satisfaction analysis algorithm to precisely measure the supplied evaluations. Using this customer satisfaction analysis, we may find out if the product is well-received by the buyer. To efficiently quantify product reviews, this research proposes a unique HAF-wBiLSTM (Handcrafted and Automated Features-driven weighted-BiLSTM) model. Combining handmade features with BoW features generates a large featureset in the proposed deep learning model. Feature criticality weight, frequency feature weights, filtered extraction of 1 and 2 g, and other features are included in this set. Weighted and bidirectional long short-term memory (LSTM) models analyze this augmented collection of information. We check the correctness and dependability of the suggested model. This analytical evaluation is carried out in comparison to CNN, LSTM, Tree-LSTM, FL-LSTM, and SVM models. The accuracy increases asserted by the suggested model were as follows: 15.28% vs SVM, 9.46% versus LSTM, 4.4% versus BiLSTM, 12.42% versus CNN, 9.77% versus Tree-LSTM, 17.24% versus RNN, and 2.18% versus LSTM + FL. Similarly, when compared to state-of-the-art approaches, the findings confirmed that the suggested HAF-wBiLSTM model had better reliability and performance.