Review Analysis

Number of Reviews

pb.configure(
    df = df_reviews
    , time_column = 'review_creation_dt'
    , metric = 'review_id'
    , metric_label = 'Share of Reviews'
    , metric_label_for_distribution = 'Number of Reviews'
    , agg_func = 'nunique'
    , norm_by='all'
    , axis_sort_order='descending'
    , text_auto='.1%'
    , update_fig={'xaxis': {'tickformat': '.0%'}}    
)

print(f'Total number of reviews: {df_reviews.review_id.nunique():,}')

Total number of reviews: 98,838

Let’s see at statistics and distribution of the metric.

pb.metric_info(freq='D')

Summary Statistics for "nunique_review_id_per_day" (Type: Integer)

Summary			Percentiles			Detailed Stats			Value Counts
Total	597 (100%)		Max	464		Mean	165.56		1	16 (3%)
Missing	---		99%	419.20		Trimmed Mean (10%)	159.12		0	14 (2%)
Distinct	303 (51%)		95%	374.20		Mode	1		5	8 (1%)
Non-Duplicate	143 (24%)		75%	269		Range	464		9	7 (1%)
Duplicates	294 (49%)		50%	154		IQR	220		17	7 (1%)
Dup. Values	160 (27%)		25%	49		Std	124.62		192	6 (1%)
Zeros	14 (2%)		5%	1.80		MAD	161.60		8	6 (1%)
Negative	---		1%	0		Kurt	-1.07		2	6 (1%)
Memory Usage	<1 Mb		Min	0		Skew	0.33		4	6 (1%)

Key Observations:

• Typical day: 1 review created

• 75% of days had ≤270 reviews

• Top 5% ≥375 reviews

Let’s look by different dimensions.

By Day of Week

pb.bar_groupby(y='review_creation_weekday', to_slide=True)

Key Observations:

• Fewest reviews on Mondays

• Sundays slightly more than Mondays but still low

• Possible review registration pattern

By Day Type

pb.bar_groupby(y='review_day_type')

Key Observations:

• 76% of reviews created on weekdays

• Matches fewer weekend days

By Review Score

pb.bar_groupby(y='review_score', to_slide=True)

Key Observations:

• Review score distribution:

  • 5 stars: 58%

  • 4 stars: 19%

  • 1 star: 12%

  • 3 stars: 8%

  • 2 stars: 3%

Review score

pb.configure(
    df = df_reviews
    , time_column = 'review_creation_dt'
    , metric = 'review_score'
    , metric_label = 'Average Review score'
    , agg_func = 'mean'
    , axis_sort_order='descending'
    , text_auto='.3s'
)

print(f'Average Review score: {df_reviews.review_score.mean():.2f}')

Average Review score: 4.07

Let’s see at statistics and distribution of the metric.

pb.metric_info(
    labels=dict(review_score='Review score')
    , title='Distribution of Review score'
    , xaxis_type='category'
)

Summary Statistics for "review_score" (Type: Integer)

Summary			Percentiles			Detailed Stats			Value Counts
Total	99.65k (100%)		Max	5		Mean	4.07		5	57.26k (57%)
Missing	---		99%	5		Trimmed Mean (10%)	4.34		4	19.15k (19%)
Distinct	5 (<1%)		95%	5		Mode	5		1	11.75k (12%)
Non-Duplicate	0 (<1%)		75%	5		Range	4		3	8.26k (8%)
Duplicates	99.64k (99%)		50%	5		IQR	1		2	3.22k (3%)
Dup. Values	5 (<1%)		25%	4		Std	1.36
Zeros	---		5%	1		MAD	0
Negative	---		1%	1		Kurt	0.43
Memory Usage	1		Min	1		Skew	-1.34

Key Observations:

• 58% of reviews had score 5.

Let’s see at statistics and distribution of the metric per day.

pb.metric_info(freq='D')

Summary Statistics for "mean_review_score_per_day" (Type: Float)

Summary			Percentiles			Detailed Stats			Value Counts
Total	583 (98%)		Max	5		Mean	4.05		5	17 (3%)
Missing	14 (2%)		99%	5		Trimmed Mean (10%)	4.09		4	15 (3%)
Distinct	511 (86%)		95%	4.60		Mode	5		4.25	5 (<1%)
Non-Duplicate	479 (80%)		75%	4.26		Range	4		4.60	3 (<1%)
Duplicates	85 (14%)		50%	4.14		IQR	0.35		3.50	3 (<1%)
Dup. Values	32 (5%)		25%	3.92		Std	0.43		3	3 (<1%)
Zeros	---		5%	3.26		MAD	0.23		3.75	3 (<1%)
Negative	---		1%	2.70		Kurt	7.18		4.33	3 (<1%)
Memory Usage	<1 Mb		Min	1		Skew	-1.61		4.67	3 (<1%)

Key Observations:

• Daily average ratings:

  • Bottom 5% <3.26

  • Middle 50% 3.9-4.3

  • Top 5% >4.6

Let’s look by different dimensions.

By Day Type

pb.cat_compare(cat2='review_day_type'
            , visible_graphs=[2]
)
pb.bar_groupby(y='review_day_type').show()

Key Observations:

• Weekdays have slightly higher ratings

• More 5-star reviews weekdays

• More 1-star reviews weekends

By Day of Week

pb.cat_compare(cat2='review_creation_weekday'
            , visible_graphs=[2]
)
pb.bar_groupby(y='review_creation_weekday').show()

Key Observations:

• Sundays have lowest ratings

• Highest 1-star share on Sundays

• Lowest 5-star share on Sundays

Review Answer Time

pb.configure(
    df = df_reviews
    , time_column = 'review_creation_dt'
    , metric = 'review_answer_time_days'
    , metric_label = 'Average Review Answer Time, days'
    , metric_label_for_distribution = 'Review Answer Time, days'
    , agg_func = 'mean'
    , axis_sort_order='descending'
    , text_auto='.3s'
)

print(f'Average Review Answer Time: {df_reviews.review_answer_time_days.mean():.2f} days')

Average Review Answer Time: 3.13 days

Let’s see at statistics and distribution of the metric.

pb.metric_info(
    upper_quantile=0.95
    , hist_mode='dual_hist_trim'    
)

Summary Statistics for "review_answer_time_days" (Type: Float)

Summary			Percentiles			Detailed Stats			Value Counts
Total	99.65k (100%)		Max	518.70		Mean	3.13		1.03	7 (<1%)
Missing	---		99%	21.95		Trimmed Mean (10%)	2.08		1.05	7 (<1%)
Distinct	82.39k (83%)		95%	6.96		Mode	Multiple		0.96	6 (<1%)
Non-Duplicate	68.47k (69%)		75%	3.10		Range	518.61		0.97	6 (<1%)
Duplicates	17.26k (17%)		50%	1.67		IQR	2.10		1.05	6 (<1%)
Dup. Values	13.92k (14%)		25%	1.00		Std	9.75		0.96	6 (<1%)
Zeros	---		5%	0.62		MAD	1.29		0.97	6 (<1%)
Negative	---		1%	0.38		Kurt	792.26		0.98	6 (<1%)
Memory Usage	1		Min	0.09		Skew	23.69		3.47	6 (<1%)

Key Observations:

• Review response time bimodal: ~1 day and ~3.5 days

• 75% responded within 3.1 days

• Top 5% took ≥7 days

Let’s see at statistics and distribution of the metric per day.

pb.metric_info(freq='D')

Summary Statistics for "mean_review_answer_time_days_per_day" (Type: Float)

Summary			Percentiles			Detailed Stats			Value Counts
Total	583 (98%)		Max	22.62		Mean	3.18		1.85	1 (<1%)
Missing	14 (2%)		99%	8.11		Trimmed Mean (10%)	2.96		2.52	1 (<1%)
Distinct	583 (98%)		95%	5.85		Mode	Multiple		2.66	1 (<1%)
Non-Duplicate	583 (98%)		75%	3.48		Range	22.24		3.51	1 (<1%)
Duplicates	13 (2%)		50%	2.86		IQR	0.99		3.30	1 (<1%)
Dup. Values	0 (<1%)		25%	2.49		Std	1.68		2.05	1 (<1%)
Zeros	---		5%	1.67		MAD	0.66		1.91	1 (<1%)
Negative	---		1%	0.90		Kurt	50.02		3.64	1 (<1%)
Memory Usage	<1 Mb		Min	0.38		Skew	5.50		2.18	1 (<1%)

Key Observations:

• 5% of review days had average response time ≥5.85 days

Let’s look by different dimensions.

By Day of Week

pb.histogram(color='review_creation_weekday').show()
pb.bar_groupby(y='review_creation_weekday').show()

Key Observations:

• Slowest responses to Friday reviews

• Fastest responses to Monday reviews

Comment Message Lenght

pb.configure(
    df=df_reviews
    , time_column='review_creation_dt'
    , metric='review_comment_message_len'
    , metric_label='Median Review Comment Message Lenght'
    , metric_label_for_distribution='Review Comment Message Lenght'
    , agg_func='median'
    , text_auto='.3s'
)

print(f'Median Review comment message lenght: {df_reviews.review_comment_message_len.median():.2f}')

Median Review comment message lenght: 54.00

Let’s see at statistics and distribution of the metric.

pb.metric_info()

Summary Statistics for "review_comment_message_len" (Type: Integer)

Summary			Percentiles			Detailed Stats			Value Counts
Total	41.46k (42%)		Max	347		Mean	71.16		9	1.36k (1%)
Missing	58.18k (58%)		99%	219		Trimmed Mean (10%)	63.62		13	573 (<1%)
Distinct	255 (<1%)		95%	194		Mode	9		17	557 (<1%)
Non-Duplicate	6 (<1%)		75%	100		Range	347		3	520 (<1%)
Duplicates	99.39k (99%)		50%	54		IQR	72		29	498 (<1%)
Dup. Values	249 (<1%)		25%	28		Std	56.60		31	495 (<1%)
Zeros	114 (<1%)		5%	9		MAD	47.44		28	488 (<1%)
Negative	---		1%	3		Kurt	0.16		11	485 (<1%)
Memory Usage	1		Min	0		Skew	1.02		33	478 (<1%)

Key Observations:

• 75% of reviews have messages ≤100 characters

Let’s look by different dimensions.

By Review Score

pb.bar_groupby(y='review_score', to_slide=True)

Key Observations:

• Lower ratings correlate with longer messages

• Negative reviews tend to be more detailed

NPS

For calculating NPS, we will divide customers into the following groups:

• Promoters: customers who gave a rating of 5

• Passive: customers who gave a rating of 4

• Detractors: customers who gave a rating of 1-3

Let’s look at how NPS changed by month.

tmp_df_res = (
    df_reviews.pivot_table(index=pd.Grouper(key='review_creation_dt', freq='D'), columns='review_score', values='review_id', aggfunc='nunique')
)
tmp_df_res['total_responses'] = tmp_df_res.sum(axis=1)
tmp_df_res['promoters'] = tmp_df_res[5]
tmp_df_res['detractors'] = tmp_df_res[1] + tmp_df_res[2] + tmp_df_res[3]
tmp_df_res['nps'] = (tmp_df_res['promoters'] - tmp_df_res['detractors']) * 100 / tmp_df_res['total_responses']
tmp_df_res.reset_index(inplace=True)

Let’s see at statistics and distribution of the metric per day.

tmp_df_res['nps'].explore.info(
    labels=dict(nps='NPS per Day')
    , title='Distribution of NPS per Day'
)

Summary Statistics for "nps" (Type: Float)

Summary			Percentiles			Detailed Stats			Value Counts
Total	476 (82%)		Max	70.97		Mean	31.54		0	7 (1%)
Missing	107 (18%)		99%	57.75		Trimmed Mean (10%)	33.80		25	3 (<1%)
Distinct	449 (77%)		95%	51.30		Mode	0		30	3 (<1%)
Non-Duplicate	429 (74%)		75%	42.87		Range	141.97		36.43	2 (<1%)
Duplicates	133 (23%)		50%	36.16		IQR	18.59		34.57	2 (<1%)
Dup. Values	20 (3%)		25%	24.27		Std	17.46		40	2 (<1%)
Zeros	7 (1%)		5%	-2.59		MAD	12.81		45.07	2 (<1%)
Negative	28 (5%)		1%	-27.43		Kurt	3.84		37.08	2 (<1%)
Memory Usage	<1 Mb		Min	-71		Skew	-1.56		32.46	2 (<1%)

Key Observations:

• Only ~5% of days had good NPS (>50)

• 5% had negative NPS

• Indicates customer dissatisfaction spikes

Comment Title

Let’s look at the word cloud from review titles.

df_reviews.viz.wordcloud('review_comment_title')

Key Observations:

• Most review titles use positive language

Let’s look at the top words by frequency.

fig = df_reviews.analysis.word_frequency(
    'review_comment_title'
    , text_auto=True
    , title='Top 10 Most Frequent Words in Review Title'
)
pb.to_slide(fig)
fig.show()

Key Observations:

• Most common title words: “recomend”, “excellent”

Let’s analyze the sentiment of the text.

df_reviews.analysis.sentiment('review_comment_title')

Key Observations:

• ~10% of titles are negative

• Sentiment IQR above 0 (neutral/positive bias)

Comment Message

Let’s look at the word cloud and the top words by frequency from the review messages.

df_reviews.viz.wordcloud('review_comment_message')
fig = df_reviews.analysis.word_frequency(
    'review_comment_message'
    , text_auto=True
    , title='Top 10 Most Frequent Words in Review Message'
)
pb.to_slide(fig)
fig.show()

Key Observations:

• Many words relate to delivery

• Most common review word: “product”

Let’s analyze the sentiment of the text.

df_reviews.analysis.sentiment('review_comment_message')

Key Observations:

• ~15% of messages are negative

• Overall sentiment leans positive

Impact of Rating on Review Text

Score 1

Let’s look at the word cloud, top 20 words by frequency, and the emotional tone of the text for a rating of 1.

df_reviews[lambda x: x.review_score==1].viz.wordcloud('review_comment_message')
df_reviews[lambda x: x.review_score==1].analysis.word_frequency('review_comment_message').show()
df_reviews[lambda x: x.review_score==1].analysis.sentiment('review_comment_message')

Key Observations:

• 1-star reviews:

  • Contain negative words

  • Clearly negative sentiment (IQR <0)

---

Score 2

Let’s look at the word cloud, top 20 words by frequency, and the emotional tone of the text for a rating of 2.

df_reviews[lambda x: x.review_score==2].viz.wordcloud('review_comment_message')
df_reviews[lambda x: x.review_score==2].analysis.word_frequency('review_comment_message').show()
df_reviews[lambda x: x.review_score==2].analysis.sentiment('review_comment_message')

Key Observations:

• 2-star reviews:

  • Contain negative words

  • Mostly negative sentiment

---

Score 3

Let’s look at the word cloud, top 20 words by frequency, and the emotional tone of the text for a rating of 3.

df_reviews[lambda x: x.review_score==3].viz.wordcloud('review_comment_message')
df_reviews[lambda x: x.review_score==3].analysis.word_frequency('review_comment_message').show()
df_reviews[lambda x: x.review_score==3].analysis.sentiment('review_comment_message')

Key Observations:

• 3-star reviews:

  • Fewer negative words

  • Leans positive overall

---

Score 4

Let’s look at the word cloud, top 20 words by frequency, and the emotional tone of the text for a rating of 4.

df_reviews[lambda x: x.review_score==4].viz.wordcloud('review_comment_message')
df_reviews[lambda x: x.review_score==4].analysis.word_frequency('review_comment_message').show()
df_reviews[lambda x: x.review_score==4].analysis.sentiment('review_comment_message')

Key Observations:

• 4-star reviews:

  • Many positive words

  • Clearly positive sentiment

---

Score 5

Let’s look at the word cloud, top 20 words by frequency, and the emotional tone of the text for a rating of 5.

df_reviews[lambda x: x.review_score==5].viz.wordcloud('review_comment_message')
df_reviews[lambda x: x.review_score==5].analysis.word_frequency('review_comment_message').show()
df_reviews[lambda x: x.review_score==5].analysis.sentiment('review_comment_message')

Key Observations:

• 5-star reviews:

  • Dominated by positive words

  • Strongly positive sentiment