Result
Result of Logistic Regression & Random Forest
| Product Category | Logistic Regression | Random Forest |
|---|---|---|
| Books | 0.85 / 0.84 | 0.83 / 0.83 |
| E-Books | 0.98 / 0.71 | 0.77 / 0.78 |
| Music | 0.85 / 0.83 | 0.72 / 0.72 |
| Digital Music | 0.82 / 0.74 | 0.67 / 0.68 |
| DVD | 0.80 / 0.78 | 0.68 / 0.68 |
| Digital Video | 0.78 / 0.72 | 0.66 / 0.65 |
| Software | 0.94 / 0.82 | 0.86 / 0.86 |
| Digital Software | 0.99 / 0.84 | 0.82 / 0.83 |
| Toys | 0.99 / 0.84 | 0.89 / 0.89 |
| Digital Video Games | 0.88 / 0.70 | 0.69 / 0.69 |
| Average | 0.89 / 0.72 | 0.83 / 0.76 |
Result of USE
Results of running model on Digital Video category dataset
| precision | recall | f1-score | support | |
|---|---|---|---|---|
| 0 | 0.78 | 0.79 | 0.78 | 6812 |
| 1 | 0.75 | 0.75 | 0.75 | 5924 |
| accuracy | 0.77 | 12736 | ||
| macro avg | 0.77 | 0.77 | 0.77 | 12736 |
| weighted avg | 0.77 | 0.77 | 0.77 | 12736 |
light_model = LightPipeline(pipeline2)
#Using a review that was stated Helpful on Amazon
text="The show is smart and awkwardly, yet deliciously, inappropriate. Miss, miss, miss Steve Carell but after a weak season 8, the Office has rebounded with season 9 and will end its run with high marks. Season 8 had its moments but the show seemed rudderless without Michael – Robert California and Nellie were just weird and Andy is no Michael. When it seemed all hope was lost, the show shifts to a more ensemble – no superstar- approach in season 9 which, with Michael gone, really works. With such wonderful characters in Dwight, Jim, Meridith, Stanley, Angela, Kevin, Oscar, Darrell and Phyllis it’s nice to have all the story lines going at once – Nellie fits in much better this year too. Andy and Erin are fine in the mix but are much better in doses than in being the main focus. A little of Andy goes a long way. That shift was a game changer in a good way."
light_model.annotate(text)['prediction'][0]
Result: 1
#Using a review that has not beed stated Helpful on Amazon YET
text="Liked it"
light_model.annotate(text)['prediction'][0]
Result: 0
text="I tossed it in the trash. It smelled so bad."
light_model.annotate(text)['prediction'][0]
Result: 0
Result of BERT
Results of running model on Digital Video category dataset
| precision | recall | f1-score | support | |
|---|---|---|---|---|
| 0 | 0.75 | 0.85 | 0.80 | 6812 |
| 1 | 0.79 | 0.68 | 0.73 | 5924 |
| accuracy | 0.77 | 12736 | ||
| macro avg | 0.77 | 0.76 | 0.76 | 12736 |
| weighted avg | 0.77 | 0.77 | 0.77 | 12736 |
light_model = LightPipeline(pipeline2)
#Using a review that was stated Helpful on Amazon
text="The show is smart and awkwardly, yet deliciously, inappropriate. Miss, miss, miss Steve Carell but after a weak season 8, the Office has rebounded with season 9 and will end its run with high marks. Season 8 had its moments but the show seemed rudderless without Michael – Robert California and Nellie were just weird and Andy is no Michael. When it seemed all hope was lost, the show shifts to a more ensemble – no superstar- approach in season 9 which, with Michael gone, really works. With such wonderful characters in Dwight, Jim, Meridith, Stanley, Angela, Kevin, Oscar, Darrell and Phyllis it’s nice to have all the story lines going at once – Nellie fits in much better this year too. Andy and Erin are fine in the mix but are much better in doses than in being the main focus. A little of Andy goes a long way. That shift was a game changer in a good way."
light_model.annotate(text)['prediction'][0]
Result: 1
#Using a review that has not beed stated Helpful on Amazon YET
text="Liked it"
light_model.annotate(text)['prediction'][0]
Result: 0
#Using a review that has not beed stated Helpful on Amazon YET
text="I tossed it in the trash. It smelled so bad."
light_model.annotate(text)['prediction'][0]
Result: 0
Business Recommendations
Sellers can identify helpful reviews in advance and use the insights from reviews to optimize their selling strategies by endorsing helpful reviews and use them as part of marketing campaigns.
Sellers can filter helpful reviews and analyze at a massive scale and derive key insights to improve products and understand target customers.
Amazon can identify unhelpful reviews, helping them better rank their reviews to improve overall customer satisfaction