A Review Helpfulness Modeling Mechanism for Online E-commerce: Multi-Channel CNN End‑to‑End Approach

Figures & data

Figure 1. A proposed CNN-TRI Framework.

Table 1. An example review of the amazon book dataset.

Attribute	Value
Reviewer ID	A2PK3NTC9RMEF4
Product ID	1894063171
Total number of votes	20
Number of helpful votes	18
Star rating	2
Review time	03, 11, 2009
Review title	Disappointing
Review text	There are eleven stories in this “grimoire;” stories where Holmes encounters crimes and/or events beyond the usual scope of the rational detective. Some are good, some are not …

Figure 2. Review helpfulness vote distribution.

Figure 3. Review length distribution.

Table 2. Review length difference.

Review	Length
100% (Maximum)	6,154
90%	358
Multiples	17.19

Table 3. Results on Affect Word Embedding Dimension on Performance.

Word Embedding Dimension	Accuracy ± SD	Precision ± SD	Recall ± SD	F1-Score ± SD
100	0.692 ± 0.002	0.679 ± 0.028	0.731 ± 0.013	0.704 ± 0.012
200	0.683 ± 0.005	0.678 ± 0.013	0.701 ± 0.034	0.688 ± 0.011
300	0.719 ± 0.800	0.709 ± 0.010	0.743 ± 0.027	0.725 ± 0.012
400	0.699 ± 0.009	0.686 ± 0.016	0.737 ± 0.027	0.710 ± 0.009
500	0.697 ± 0.016	0.676 ± 0.018	0.759 ± 0.017	0.715 ± 0.011

Note: SD (Standard Deviation)

Table 4. Results on Affect Dropout Rate on Performance.

Dropout Rate	Accuracy ± SD	Precision ± SD	Recall ± SD	F1-Score ± SD
0.1	0.696 ± 0.009	0.703 ± 0.017	0.682 ± 0.049	0.691 ± 0.019
0.2	0.714 ± 0.009	0.706 ± 0.019	0.737 ± 0.037	0.720 ± 0.011
0.3	0.698 ± 0.013	0.683 ± 0.014	0.741 ± 0.017	0.710 ± 0.012
0.4	0.720 ± 0.004	0.712 ± 0.017	0.741 ± 0.034	0.725 ± 0.008
0.5	0.715 ± 0.005	0.708 ± 0.024	0.739 ± 0.053	0.721 ± 0.013
0.6	0.716 ± 0.016	0.736 ± 0.025	0.681 ± 0.077	0.704 ± 0.035
0.7	0.711 ± 0.027	0.704 ± 0.026	0.727 ± 0.039	0.715 ± 0.029
0.8	0.701 ± 0.010	0.699 ± 0.012	0.705 ± 0.736	0.702 ± 0.009
0.9	0.702 ± 0.004	0.709 ± 0.007	0.686 ± 0.016	0.697 ± 0.006

Note: SD (Standard Deviation)

Table 5. Results on Affect Vocabulary Size on Performance.

Vocabulary Size	Accuracy ± SD	Precision ± SD	Recall ± SD	F1-Score ± SD
90131 (Maximum)	0.710 ± 0.021	0.714 ± 0.020	0.701 ± 0.026	0.707 ± 0.022
90000	0.716 ± 0.018	0.712 ± 0.021	0.726 ± 0.013	0.719 ± 0.015
80000	0.712 ± 0.022	0.702 ± 0.022	0.736 ± 0.028	0.719 ± 0.022
70000	0.706 ± 0.011	0.707 ± 0.017	0.707 ± 0.026	0.707 ± 0.012
60000	0.715 ± 0.008	0.719 ± 0.013	0.708 ± 0.025	0.713 ± 0.011
50000	0.720 ± 0.007	0.717 ± 0.010	0.729 ± 0.037	0.722 ± 0.014
40000	0.726 ± 0.008	0.711 ± 0.017	0.758 ± 0.028	0.733 ± 0.007
30000	0.718 ± 0.015	0.714 ± 0.008	0.725 ± 0.038	0.719 ± 0.021
20000	0.710 ± 0.013	0.708 ± 0.016	0.715 ± 0.030	0.711 ± 0.016
10000	0.721 ± 0.015	0.728 ± 0.014	0.706 ± 0.039	0.716 ± 0.024

Note: SD (Standard Deviation)

Figure 4. Comparison of CNN-TRI and Benchmarks.

Table 6. Performance Comparison with Benchmarks.

Model	Accuracy	Percentage Change	F1-Score	Percentage Change
TRI	0.705	+2.912%	0.704	+4.05%
CM	0.693	+4.632%	0.699	+4.906%
CNN	0.675	+7.553%	0.683	+7.354%
SVM	0.629	+15.346%	0.644	+13.735%
NB	0.578	+25.608%	0.437	+57.75%