Table 4.
Top ten most frequent web entities and labels detected within the visual content of Instagram posts with hashtags associated with Heat Not Burn tobacco products (n = 4629).
| Top web entities | n(%) | Mean confidence (min, max)a | Top labels detected | n (%) | Mean confidence (min, max)b |
|---|---|---|---|---|---|
| IQOS | 1078 (23.3%) | 1.4 (0.2, 5.6) | Product | 753 (16.3%) | 84.4% (80%, 94.1%) |
| Product | 1046 (22.6%) | 0.6 (0.2, 0.9) | Electronic device | 261 (5.6%) | 86.1% (80%, 92.3%) |
| Heat-not-burn-product | 886 (19.1%) | 0.7 (0.2, 0.9) | Technology | 249 (5.4%) | 85.7% (80%, 92.8%) |
| Product design | 859 (18.6%) | 0.7 (0.2, 0.9) | Fashion accessory | 105 (2.3%) | 85.3% (80.2%, 92%) |
| Design | 812 (17.5%) | 0.5 (0.1, 0.9) | Text | 114 (2.5%) | 90.8% (85.1%, 97.3%) |
| Tobacco | 634 (13.7%) | 0.6 (0.2, 0.7) | Purple | 108 (2.3%) | 87.1% (80%, 96%) |
| Electronic cigarette | 617 (13.3%) | 0.6 (0.2, 0.9) | Social group | 107 (2.3%) | 93.3% (89.2%, 97.8%) |
| Cigarette | 389 (8.4%) | 0.6 (0.2, 0.9) | Beauty | 107 (2.3%) | 88% (85.1%, 95.6%) |
| Marlboro | 360 (7.8%) | 0.6 (0.2, 0.9) | Blue | 102 (2.2%) | 96.8% (96.8%, 96.8%) |
| Font | 367 (7.9%) | 0.6 (0.1, 0.9) | Photography | 86 (1.9%) | 82.4% (80%, 99.2%) |
Note: Visual content including web entities (information about the post’s image based on other images on the internet) and identified labels (e.g. objects) were analyzed using Google Cloud Vision API in February 2018. The top ten web entities and labels detected were found using the “Tidytext” package in R.
Web entity confidence scores are unstandardized and do not represent a percentage of confidence.
Label confidence scores are standardized between 0 and 1 and represent percent confidence. A cut point was used to restrict the top labels to those at or above 80% confidence.