Qloo creates highly structured data science on consumer taste and entity attributes across the largest categories of entertainment, culture, and consumer products.
To process the data in real-time, Qloo uses proprietary machine learning algorithms that utilize leading statistical methodologies, rooted in the latest research in the emerging field of Neuroaesthetics — these include: deep learning methods, Bayesian statistics, neural networks, and proprietary NLP algorithms. We optimize for correlations that are meaningful and uniquely correlated (known as “anomalous co-occurrences”). Proprietary recursive training models are used to validate insights, properly partitioning all training data to avoid overfitting. Qloo connects over a dozen data domains in search of uniquely relevant insights to drive true excellence in programmatic cultural intelligence.
Our Deep Learning data science works across three distinct layers:
Qloo maintains a database of cultural entities that includes a detailed ontology of all attributes pertaining to each entity. The goal here is to understand the fundamental DNA of each cultural entity, (whether book, movie, product or hotel) by mapping and structuring, and scoring the most important attributes.
Below is a breakdown of entity counts by category, Qloo’s AI currently services more than 150 million primary entities along with all their associated attributes). These primary entities are available turn-key as input signals into the Qloo API.
|Category||Total Number Of Entities|
|Music: Artists, Albums, Songs||31,500,000+|
|Film: Movies, Actors, Directors||2,500,000+|
|TV: TV Shows, Actors||850,000+|
|Dining: Restaurants, Cafes||2,700,000+|
|Nightlife: Bars, Clubs||900,000+|
|Fashion: Brands, Designers, Stores||550,000+|
|Category||Total Number Of Entities|
|Books: Book Titles, Authors||18,500,000+|
|Travel: Destinations, Hotels, Attractions||4,700,000+|
|People: Public Figures, Influencers|| 175,000+|
|Media: Apps, Podcasts, Games|| 1,100,000+|
| Products: Brands, Products|| 105,000,000+|
|Sports: Athletes, Teams||25,000+|
For every entity, Qloo maps a rigorous ontology of its most salient attributes, and cross references this to all other entities within the domain and even across domains.
For instance, for two entity types, (a music artist and a restaurant) the below table shows the top level attributes that Qloo might maintain for each entity.
|Parent Genre||Biographical Details|
|Record Label||Chart Statistics|
|Business Attributes||Dress Code|
|Hours of Operation||Price Levels|
| Other Locations||Noise Levels|
|All Menu Items||Ambience|
|Dietary Details|| Similar To|
Qloo API processes anonymized batch signals across culture. Qloo API is fundamentally GDPR-compliant as a data processor.
Input signal into Qloo’s API can be any of the 150+ million entities pre-mapped, and can also be attributes.
Examples of raw signals include:
Desired response is delivered in less than 50 milliseconds on average along with any delimiting fields, appended attributes and parameters
Qloo then retrains models for subsequent calls based on batch anonymized input signals. Primary machine learning involves proprietary neural nets.
Qloo develops custom routes for enterprises to address specific value-adds. These models will typically combine primary taste signals with Layer 1 attributes to create models that address unique use-cases.
Some examples of models that have been developed for clients include the following routes:
|Content Themes||Social Issues||Generational Affinity|
|Mainstream/Novelty||Travel Preferences||Lifestyle Preferences|
| Introversion/Extraversion||Price Sensitivity|| Genre Preferences|
|Vanity||Openness||Propensity to Consume|
|Political Leanings||Irony/Sincerity||Media Preferences|
Popular evaluation metrics in recommender systems explained by Giorgos Papachristoudis, PHD – Chief Data Scientist