How to Build a User‑Perception‑Driven Self‑Awareness System for Network Operations

Overview

The self‑perception system places real‑time user perception at the core of network operation. By continuously collecting user feedback, service usage, and network performance indicators, the system builds a unified user portrait and drives automated, personalized optimization actions.

Four Key Stages

1. Perception Manifestation

Raw experience data from mobile and broadband services are fused through an OB (business‑object) domain integration layer. AI‑enabled analytics convert textual feedback, QoS metrics, and service logs into quantifiable events, tags, and scores. The output is a comprehensive user portrait that captures user value, usage preferences, and behavior characteristics, enabling the identification of factors that most affect perceived quality.

2. Perception Demand

The manifested portrait is analysed to extract explicit user demands. Demands are classified into two groups:

Network‑related demand : traffic patterns, cell‑level performance, application‑specific latency, etc.

Service‑related demand : subscription plan suitability, ARPU, device type, complaint history.

Statistical correlation and root‑cause analysis link these demands to specific quality‑of‑experience (QoE) degradations, forming the basis for targeted optimization strategies.

3. Perception Response

For each dissatisfied user segment, the OB‑fused big‑data evaluation model generates corrective actions. The model evaluates:

User value score

Preference weight

Behavioral risk factors

Based on these scores, the system either adjusts existing optimization tasks (e.g., parameter tuning, resource re‑allocation) or creates new tasks for users lacking coverage. Task priority is dynamically recomputed using impact scope and severity thresholds.

4. Perception Feedback

After strategy execution, key performance indicators (KPIs such as drop rate, latency, throughput, and post‑action satisfaction scores) are monitored. A closed‑loop verification mechanism compares pre‑ and post‑action metrics, feeds the delta back to the strategy scheduler, and triggers automatic re‑prioritisation or refinement of the optimization plan.

Implementation Architecture

The system relies on a “dual‑full” user portrait that merges mobile‑network and broadband data streams:

Data ingestion : real‑time logs, CDRs, network probes, and user‑generated feedback are streamed into a unified data lake.

OB domain fusion : cross‑domain identifiers (e.g., user ID) link business objects from the mobile (B) and broadband (O) domains, enabling end‑to‑end traceability.

Feature engineering : statistical and deep‑learning models extract features such as session quality, application usage, and complaint sentiment.

Clustering : unsupervised algorithms (e.g., K‑means, DBSCAN) group users with similar dissatisfaction patterns, supporting a “one‑case‑one‑solution” approach.

Evaluation model : a weighted scoring function Score = w1*Value + w2*Preference + w3*Behavior ranks users for remediation priority.

Task engine : a rule‑based scheduler creates, updates, and retires optimization tasks, exposing APIs for downstream OSS/BSS systems.

Feedback loop : KPI dashboards capture before/after metrics; statistical tests (e.g., paired t‑test) validate significance of improvements.

Figures illustrate the perception‑driven lifecycle, demand‑closure process, response‑closure workflow, and feedback application scenarios.