For each company, we access the corresponding URL and, using scraping techniques, browse both the home page and the available links. During this process, the associated social networks are identified and extracted.

The evaluation was conducted by comparing the results obtained from scraping with those collected in the Survey on the Information Society in Companies (ESIE) regarding the variable of the presence of social networks on company websites. The evaluation of the project was carried out using the answers given in the surveys from two consecutive years (2022 and 2023).

Accuracy:

Measures the proportion of correct predictions (true positives and true negatives) out of the total number of predictions.

Accuracy = TP + TN / (TP + TN + FP + FN)

Accuracy stood at 83% in 2022.

Accuracy stood at 85% in 2023.

Sensitivity:

Clearly identified true positives.

Sensitivity = TP / (TP + FN)

Sensitivity stood at 78% in 2022.

Sensitivity stood at 81% in 2023.

Specificity:

Clearly identified true negatives.

Specificity = TN / (TN + FP)

Specificity stood at 95% in 2022.

Specificity stood at 96% in 2023.

Positive Predictive Value (PPV):

Proportion of positives predicted by the model that are actually positive.

Specificity = TN / (TN + FP)

The PPV was 97% in 2022.

The PPV was 98% in 2023.

When the model predicts that the website has social networks, there is a 98% probability that the prediction is correct.

Negative Predictive Value (NPV):

Proportion of negatives predicted by the model that are actually negative.

Negative Predictive Value (NPV):) = TN / (TN + FN)

The NPV was 64% in 2022.

The NPV was 66% in 2023.

1. Ability to identify true positives: The model shows sensitivity of 80%, indicating a high ability to detect positive cases.
2. High overall accuracy.
3. Excellent at identifying true negatives: Specificity of 95% is remarkably high.
4. Accurate confirmation of social networks: A PPV of 98% means that when the model predicts that a web page has social network integrations, there is a 98% probability that this prediction is correct.
5. Robust performance: The model stands out for its high accuracy, excellent specificity and exceptional positive predictive value.