In recent years, e-commerce development has been in full swing. From the online retail platform headed by Taobao to the vertical e-retailers such as fresh food e-retailers, China's e-commerce market is booming sustainably. The Daily Active User (DAU) of Taobao mobile app reached a peak of 260 million users in June 2018.
Since last year, the Alibaba has encouraged its customers to write reviews on its e-commerce platform. The more words and images the customers provide in their reviews, the higher scores they will get. While having more reviews increases the likelihood of product sales, it also brings difficulties in preventing and controlling malicious reviews. On an average, the Alibaba e-commerce platform generates 20 million reviews every day. The review content not only reflects the sellers' reputation but also affects decisions made by customers before they place orders. Moreover, reviews are the most direct way for customers to submit feedback and interact with sellers.
Brand makers can analyze the reviews to find out and improve the most attractive selling points of commodities and aspects. Subsequently, they can learn the current consumption philosophy, predict the fashion trend, and increase the purchase conversion rate. Assume that a seller receives hundreds of reviews every day. On average, a total duration of two customer service personnel's working hours gets utilized in manually identifying good and poor reviews and recording the essential information reported by the customer. This duration increases in the case of any marketing activity or sales promotion.
To cope with these issues, the Alibaba Cloud natural language processing system (NLP) provides a function specifically for review quality and authenticity analysis. Based on the NLP technology and corpus in the e-commerce industry, this function can automatically analyze historical reviews and new reviews generated every day to convert texts into structured attribute fields. Thus, it efficiently identifies and follows up with the net negative reviews and analyzes the most influential factors that affect the purchase conversion rate.
The major use of Alibaba Cloud review analysis function is to analyze the customer reviews, as well as the colloquial and short text contents similar to microblogs. The review analysis function extracts essential attributes of commodities, builds the label system that conforms to the industrial features, and helps customers quickly learn the experiences of historical customers. This function provides complete processing of the Alibaba e-commerce platform and ensures sequential output of reviews, which are helpful in the decision-making process of customers.
The function also supports emotional analysis on evaluation-type or microblog-type content. It structures user reviews and classifies these reviews into three categories: commodity, service, and logistics, based on which emotional analysis is performed to help sellers carry out the after-sales service.
The function supports the emotional analysis of commodity attributes. It extracts key attributes and provides the scores to let sellers learn the customers' impression of products sold by them. Hence, it gives a reference for making decisions on the subsequent sales directions. For clarity, let's look at an example review.
"The material is comfortable, and the style is good except that the size is relatively small. Fortunately, I saw the reviews of other customers and chose a size larger than my normal size. This size is just suitable for me. I am quite satisfied. A full score! The service is good, and the shipment is fast."
The following shows how the review analysis function judges whether this review is positive or negative.
{
"text": "The material is comfortable, and the style is good except that the size is relatively small. Fortunately, I look through the reviews of other customers and choose a size larger than my normal size. This size is just suitable for me. I am quite satisfied. A full score! The service is good, and the shipment is fast.",
"cate": "clothing"
}
Text is the label for the review and contains thee review text to be analyzed. Cate is the label for industry category. The industry category is selected from a list of existing categories, which includes "furniture", "bedding", "bags", and "make up".
Gradually, other industries may be added into the system according to the needs of the business. Pay attention to the document update or consult the customer service personnel.
Alibaba's NLP will analyze the review text and cate tags and generate the following response:
{
"result": {
"aspectItem":
[
{
"aspectCategory":"Logistics",
"aspectIndex":"55 57",
"aspectTerm":"Delivery",
"opinionTerm":"The delivery speed is fast",
"aspectPolarity":"Positive"
},
{
"aspectCategory":"Fabric/material",
"aspectIndex":"0 2",
"aspectTerm":"Material",
"opinionTerm":"Comfortable",
"aspectPolarity":"Positive"
},
{
"aspectCategory":"Seller's service",
"aspectIndex":"51 53",
"aspectTerm":"Service",
"opinionTerm":"The service is good",
"aspectPolarity":"Positive"
},
{
"aspectCategory":"Type/style",
"aspectIndex":"5 7",
"aspectTerm":"Style",
"opinionTerm":"Good",
"aspectPolarity":"Positive"
},
{
"aspectCategory":"Size",
"aspectIndex":"11 13",
"aspectTerm":"Size",
"opinionTerm":"Relatively small",
"aspectPolarity":"Negative"
}
],
"textPolarity": "Positive",
"textIntensity": "0.4851228125853578"
}
}
The table below describes the list of parameters used in the response.
Parameter name | Value |
textPolarity | Emotional polarity of the entire text. The values include Positive, Neutral, and Negative. If the input of the text field is invalid, "-100" is returned. |
textIntensity | Emotional intensity of the entire text. The value ranges from -1 to 1. The higher the value is, the more favorable the result is. The smaller the value is, the more negative the outcome is. A value close to 0 indicates neutral. |
aspectItem | Attribute list. Each element is a JSON field. |
aspectCategory | Attribute category |
aspectIndex | Start position and end position of the attribute term |
aspectTerm | Attribute term |
opinionTerm | Emotion term |
aspectPolarity | Attribute polarity (Positive, Neutral, and Negative) |
After sorting the data as shown in the response, human reviewers can easily evaluate the reviews received by customers. What's more, because they are consistently tagged, the entire process can be sped up through automation.
Read similar blogs and learn more about Alibaba Cloud's products and solutions at www.alibabacloud.com/blog.
Alibaba Cloud Empowers China's Open Source Cloud Computing Communities
2,599 posts | 758 followers
FollowAlibaba Cloud Project Hub - March 19, 2024
Nick Patrocky - February 17, 2024
Alibaba Cloud Project Hub - April 17, 2023
Lana - April 14, 2023
Alibaba Clouder - July 22, 2020
Alibaba Clouder - January 23, 2020
2,599 posts | 758 followers
FollowConduct large-scale data warehousing with MaxCompute
Learn MoreMore Posts by Alibaba Clouder