Advertisement Image Classification—Visual ( <scp>RESNET</scp> ) Versus Textual ( <scp>BERT</scp> ) Features: An Experimental Study
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
ABSTRACT Newspapers serve as a vital source for various types of advertisements. Individuals eagerly await and search for advertisements relevant to them in newspapers. However, both printed newspapers and online newspapers lack the ability to provide category‐wise advertisement search options. As a result, searching a newspaper advertisement in a specific category becomes very time‐consuming and cumbersome due to sequential manual search across multiple newspapers. To address this problem in online newspapers, a classification model is needed that can classify advertisement images into predefined categories and hence allow users to perform category‐wise advertisement searches with much ease. This research introduces and compares two sets of classification models for advertisement images in online English newspapers in India. The first set utilizes visual features to train seven different classification models by fine‐tuning different layers of the Residual Network with 50 layers (ResNet50) pretrained model and achieves a maximum classification accuracy of 71.41%. The second set utilizes textual features to train 14 different classification models by fine‐tuning different layers of the pretrained Bidirectional Encoder Representations from Transformers (BERT) base model and achieves maximum classification accuracies in the range from 96.88% to 97.34%. This significant enhancement of more than 25% underscores the superiority of textual features over visual ones in understanding Indian online English newspaper advertisement images and holds promise for practical applications, including categorized advertisement searches.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.001 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.001 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from it