Click the link to use the template.
Customer satisfaction is a top priority for e-commerce companies. Major e-commerce platforms offer user review features, making it highly valuable to extract customer satisfaction insights from these reviews promptly. This case will demonstrate how to use Bayeslab to effectively mine information from product user reviews.
Data Preparation
The data used in this case comes from a collection of customer reviews for Amazon products, with each record including the name or alias of the reviewer, the country, the number of reviews submitted, the date the review was posted, the rating given, a short title summarizing the review, the full text of the review, and the date the customer used the product.
Let's deal with the data first, click on the data table on the left to see the data in the data preview popup on the right.
prompt:
Remove the rows with empty data from the Review Text column and the result is written to the table.
Analysis of user comments
Explore Data Analysis （EDA）
Analysis of the Rating
prompt:
Plotting a bar graph based on the number of different Rating
AI Chart Interpretation
Observations
Distribution of Ratings: The chart shows a bar graph of customer ratings from 1 to 5. The majority of reviews are extremely negative, with a rating of 1, totaling 13,123. This is significantly higher than any other rating.
Positive vs. Negative Ratings: The second most common rating is 5, with 4,528 counts, suggesting a polarized set of customer experiences. Ratings of 3 and 4 are less frequent, with 885 and 1,292 respectively, indicating less moderate experiences.
Unbalanced Feedback: There are relatively few ratings of 2 (1,227), suggesting a gap in slightly negative experiences. Most customer feedback tends to be either very positive or very negative.
Since 1-star ratings are the most numerous, this may indicate that there are significant problems with the product or service that need to be improved to increase customer satisfaction.
Country distribution of users
prompt:
Counting the number of different Countries and displaying them on a bar graph: only the top 10 countries are shown.
AI Chart Interpretation
The chart displays the number of customer reviews from the top 10 countries. The United States (US) leads significantly with 9,286 reviews, followed by Great Britain (GB) with 7,294 reviews. There is a notable drop in review counts after these two countries, with Canada (CA) having 708 reviews and India (IN) with 629. Other countries like Ireland (IE) and Denmark (DK) have similar lower counts, ranging between 150 and 242 reviews.
As can be seen from the graph, the number of comments is very unevenly distributed, with the US and the UK having far more comments than the other countries.
High Frequency Keyword Analysis
Let's analyze which keywords appear most frequently in these reviews to dig into which issues customers are most concerned about.
prompt:
Count the words that appear more frequently in Review Text and map the word cloud.
The largest words in the word cloud are “amazon”, “customer”, “service”, ” delivery”, ‘product’, ‘order’, etc. These words appear most frequently and reflect the main topics discussed in user reviews.
Customer Review Sentiment Analysis
Sentiment analysis of customer reviews refers to the categorization of customer reviews into positive, negative, and neutral, and we can discover customer satisfaction through sentiment analysis.
prompt:
Sentiment classification (pos/neg/neutral) based on the content of Review Text, a new column is added to write the sentiment classification result. Results are written back to the table
Summary of Emotional Categories
prompt:
Count the number of different Sentiment, draw a bar chart, use different colors to display, descending order
AI Chart Interpretation
The chart shows the sentiment counts of customer reviews, displayed as a bar chart with different colors for each sentiment category. Here are some observations:
Distribution: Positive reviews significantly exceed the others, with a count of 10,621. Negative reviews follow with 8,340, and neutral reviews are the least frequent at 2,094.
Trend: The data indicates a higher tendency towards positive sentiment among the reviews, but negative sentiments also form a substantial portion, almost catching up to positives.