Select Page

Weighted Normalised Matrix

The weighted and normalised matrix enables direct, fair comparison between all search categories, removing bias from unequal sample sizes and object density across the dataset. In this analysis, each image is assigned a weight inversely proportional to the number of images in its category. This ensures that categories with more images do not dominate the results. When an image contains multiple detected objects, its weight is divided equally among those objects, so object-rich images do not disproportionately influence group totals.

After all object-level weights are summed, the matrix is column-normalised so each analytic group (for example, a Maslow level or object group) sums to exactly 1. This approach produces proportional results that are directly comparable across groups, making it possible to identify the distinctive thematic content of each category on an equal footing.

Group Assignment & Aggregation

Each object is mapped to a single group:

  • Either one of six Maslow psychological levels,
  • Or one of ten DA object groups (such as Humans, Food, Technology, etc.).

For each search category and each group, all weighted object contributions are summed across the entire dataset, producing a (category × group) matrix for both Maslow and DA frameworks.

Column-Wise Normalisation

To enable direct proportional comparison across groups, the sum of weighted contributions for each group (Maslow or DA) is normalised so that the total for each column is exactly 1.0:

 

  • This is not an extra weighting step; it is a uniform scaling that preserves the relative proportions within each group.
  • The resulting values represent the proportion of total weighted content in each group attributable to each search theme.

Matrix Structure

Analysis Dimensions Column Totals Each Cell Represents
Maslow 20 categories × 6 levels Each level sums to 1 The proportion of weighted psychological content for each level from each category
DA Object Groups 20 categories × 10 groups Each group sums to 1 The proportion of weighted object presence for each DA group from each category

Key Contributions by Level

DA Object Group Top 3 Contributors (Proportions) Maslow Level Top 3 Contributors (Proportions)
Humans Education (16.5%), doomscrolling (14.7%), technology_addiction (14.7%) Level 1: Physiological Needs instinct (7.1%), social_media_addiction (6.4%), doomscrolling (6.3%)
Technology & Digital Media technology_addiction (18.1%), Connection (17.4%), Creativity (17.2%) Level 2: Safety Needs safety (15.7%)*, instinct (9.6%), danger (9.3%)
Natural Landscapes & Phenomena nature (31.0%), danger (17.9%), sadness (15.8%) Level 3: Love and Belonging social_media_addiction (8.3%), technology_addiction (7.7%), doomscrolling (7.0%)
Built Structures & Indoor/Urban Spaces Education (22.3%), play (14.1%), isolation (13.5%) Level 4: Esteem Needs education (12.4%), standardisation (9.4%), technology_addiction (7.5%)
Manual Tools & Functional Objects isolation (15.2%), disconnection (14.9%), safety (14.8%) Level 5: Self-Actualization creativity (6.7%), sadness (6.3%), human_instinct (5.9%)
Animals & Non-Human Organisms danger (50.8%), human_instinct (29.9%), instinct (17.1%) Level 6: Self-Transcendence standardisation (8.8%), nature (8.2%), artificial (7.4%)
Plants & Botanical Elements nature (39.4%), danger (26.4%), conformity (14.0%)
Vehicles & Transportation freedom (31.2%), safety (23.4%), danger (23.0%)
Art, Sports & Recreational Objects play (32.9%), sadness (24.1%), conformity (15.8%)
Food, Beverages & Tableware happiness (19.2%), social_media_addiction (18.0%), danger (17.0%)

*Safety’s high contribution to Safety Needs reflects a methodological artefact, as the term functions both as a search category and a Maslow class.

Conclusion:

This weighted, normalised approach provides a structurally fair view of the relative thematic presence of each category, enabling robust, bias-free comparison across both psychological and object-based frameworks. Results reflect proportional analytic presence under equalised conditions and are not intended to represent real-world frequency. For prevalence or dominance, refer to raw or unweighted data.

←→