An experimental academic project
An open-source recommender framework that turns sensor-captured interactions and knowledge graph context into explainable rules and ranked content.
01 / How it worked
InBeat treated recommendation as a chain of evidence: capture a user signal, connect it to semantic context, turn it into learned rules, then use those rules for ranking.
01 · Sensor input
Research prototypes used Microsoft Kinect to observe non-verbal cues while a person watched video. Gaze and body-derived signals could serve as implicit feedback without requiring a rating, click, or remote-control action.
02 · Knowledge graphs
Linked data connected the watched item to entities, types, and broader concepts. Evidence about one item could therefore contribute to a topic or category—not merely an isolated URL.
03 · Rule learning
Aggregation rules translated interaction evidence into interest. Association-rule learning discovered recurring relationships between semantically annotated content and user behavior; Apriori-based pruning kept user models compact and fast.
04 · Recommendation
The recommender applied learned preferences through interchangeable algorithms and APIs to score candidate content. Separate capture, learning, and recommendation modules kept the framework reusable.
02 / Architecture
Each module had a distinct role, allowing experiments to exchange sensor inputs, preference learners, and recommendation algorithms.
The General Analytics INterceptor received interaction events, enriched them with context, and converted raw signals into a usable measure of interest.
The Preference Learning module used association-rule learning and pruning to turn semantically described interactions into interpretable user models.
The Recommender System applied learned preferences through pluggable algorithms and APIs to score and order candidate content.
03 / Research record
Research across recommender systems, semantic web, multimedia interaction, and non-invasive interest measurement.
Archive note
InBeat is preserved as an experimental academic project. The framework, research trail, and demonstration remain available for study; the former public application is no longer maintained.