Identifying Significance of Product Features on Customer Satisfaction Recognizing Public Sentiment Polarity: Analysis of Smart Phone Industry Using Machine-Learning Approaches

Figures & data

Figure 1. Process of sentiment analysis and feature selection.

Table

Algorithm for sentiment scoring

Input For each comment: Compute sentiment_score Compute mean_sentiment_score Compute median_sentiment_score Compute stdv_mean Compute stdv_median IF stdv_mean < stdv_median Select mean_sentiment_score ELSE Select median_sentiment_score

Figure 2. Public opinion on Samsung S10 plus.

Figure 3. Comparison among five brands based on consumer sentiment.

Figure 4. PCA using two components.

Table 1. Smart phone models we experimented.

Brand	High range	Mid range	Low range
Samsung	Galaxy Note 9, Galaxy S8 plus, Galaxy S10 plus	Galaxy Note 4, Galaxy A8, Galaxy S 6	Galaxy Grand Prime, Galaxy S4, Galaxy J7
Apple	iPhone 8 plus, iPhone X, iPhone X S	iPhone 7, iPhone 7 plus, iPhone 8	iPhone 5, iPhone 5 S, iPhone 6
Huawei	Mate 10 pro, P9, P20 pro	Mate S3, Nova 3i, P20 lite	Ascend G510, Ascend 6, Mate 8
Xiaomi	Mi A1, Mi Mix 2, Pocophone F1	Mi A2, Mi A2 lite, Redmi note 6 pro	Redmi 6, Redmi 7, Redmi note 5
Oppo	F7, F11 pro, R17	F1, F3 plus, F9	A37, A 3 S, F1 S

Table 2. Some of the words that we added to vader lexicon.

omg (oh my god)	2.6	1.11355	[3, 4, 2, 4, 1, 2, 3, 1, 4, 2]
fn (fine)	0.8	0.6	[1, 0, 1, 2, 1, 1, 1, 1, 0, 0]
cc (carbon copy)	−0.4	0.66332	[0, 1, −1, 0, 0, 0, −1, −1, −1, −1]
dftba (don’t forget to be awesome)	0.5	1.0247	[0, 0, 3, 0, 0, 0, 2, 0, 0, 0]
fath (first and truest husband)	2.0	0.44721	[2, 3, 1, 2, 2, 2, 2, 2, 2, 2]
ftw (for the win)	0.5	0.80623	[0, 0, 1, 0, 0, 0, 0, 2, 2, 0]
gr8 (great)	3.0	0.63246	[3, 3, 2, 3, 3, 3, 4, 4, 3, 2]
ily (I love you)	2.1	0.3	[2, 2, 2, 2, 2, 2, 3, 2, 2, 2]
ns (not safe)	−1.4	0.66332	[−1, −1, −1, −1, −2, −1, −1, −3, −2, −1]
fomo (fear of missing out)	0.6	1.35647	[2, −2, 2, 1, 1, −2, 1, 1, 1, 1]
oh (overheard)	−0.5	0.92195	[0, 0, 0, 0, −3, 0, −1, 0, −1, 0]
gd (good)	1.9	0.9434	[2, 1, 1, 3, 2, 4, 2, 2, 1, 1]
wrst (worst)	−3.1	1.04403	[−4, −4, −3, −1, −3, −4, −2, −2, −4, −4]
bd (bad)	−2.5	0.67082	[−3, −2, −4, −3, −2, −2, −3, −2, −2, −2]
awsm (awesome)	2.6	1.11355	[3, 4, 2, 4, 1, 2, 3, 1, 4, 2]

Table 3. Public sentiment results of five brands.

Brand	Compound Sentiment Score	No. of Positive Reviews	No. of Negative Reviews	No. of Neutral Reviews
Samsung	0.38	6859	1892	249
Apple	0.35	7028	1739	233
Huawei	0.39	7168	1671	161
Oppo	0.31	7035	1793	172
Xiaomi	0.36	6853	1893	264

Table 4. Performance evaluation of machine learning algorithms with feature selection techniques.

Algorithm	Correctly Classified Instances (%)	Kappa statistic	Root mean squared error	TP Rate (Weighted Avg)	FP Rate (Weighted Avg)	Precision (Weighted Avg)	F-Measure (Weighted Avg)	ROC Area (Weighted Avg)
With Initial 31 attributes
Naïve Bayes	53.33	0.26	0.52	0.53	0.25	0.72	0.57	0.58
Multilayer Perceptron	40.00	−0.29	0.54	0.40	0.79	0.44	0.42	0.37
SVM	66.67	0.47	0.43	0.67	0.05	0.87	0.69	0.73
J48	40.00	0.20	0.55	0.40	0.09	0.81	0.42	0.70
Random Forest	40.00	0.04	0.47	0.40	0.44	0.55	0.43	0.54
With reduced 23 attributes found by UFS
Naïve Bayes	46.67	0.20	0.49	0.47	0.26	0.66	0.48	0.74
Multilayer Perceptron	46.67	−0.05	0.53	0.47	0.43	0.57	0.51	0.60
SVM	60.00	0.40	0.40	0.60	0.06	0.90	0.65	0.87
J48	33.33	0.15	0.55	0.33	0.10	0.79	0.32	0.64
Random Forest	53.33	0.15	0.45	0.53	0.42	0.63	0.56	0.63
With reduced 25 attributes found by RFE
Naïve Bayes	46.67	0.09	0.54	0.47	0.43	0.60	0.50	0.51
Multilayer Perceptron	33.33	−0.06	0.61	0.33	0.45	0.51	0.38	0.39
SVM	53.33	0.15	0.46	0.53	0.42	0.59	0.53	0.58
J48	40.00	0.20	0.55	0.40	0.09	0.81	0.42	0.70
Random Forest	46.67	0.20	0.44	0.47	0.26	0.67	0.48	0.74
With reduced 21 attributes found by PCA
Naïve Bayes	53.33	0.26	0.45	0.53	0.25	0.70	0.56	0.79
Multilayer Perceptron	60.00	0.21	0.46	0.60	0.24	0.70	0.64	0.75
SVM	73.33	0.55	0.38	0.73	0.04	0.88	0.75	0.88
J48	53.33	0.26	0.56	0.53	0.25	0.72	0.57	0.65
Random Forest	53.33	0.26	0.44	0.53	0.25	0.70	0.55	0.65

Table 5. Attribute variance value measured by PCA.

Attribute	Variance (absolute)	Attribute	Variance (absolute)
Battery (mAmp)	0.901725	Camera extra feature	0.001616
Video (pixel)	0.303088	OS version	0.001522
Front Video (pixel)	0.250623	CPU core	0.000911
Price	0.125263	Display size	0.000717
External Memory	0.085875	Brand	0.000713
Internal Memory	0.078258	Sensors	0.000646
Resolution (ppi)	0.047789	CPU (clock)	0.000340
Weight	0.022369	Dimension (width)	0.000218
Dimension (height)	0.010289	Screen protection	0.000168
Fast charging	0.006813	Selfie camera extra feature	0.000132
Dimension (length)	0.004406	Build material	0.000096
Video frame	0.004054	Battery Removal	0.000089
Selfie camera (pixel)	0.003908	Glass	0.000075
Front video frame	0.003839	Selfie camera (focus)	0.000072
Main camera (pixel)	0.003778	Main camera (focus)	0.000043
RAM	0.001925

Figure 5. Smart phone dataset explained variance.