| GNG | Growing Neural Gas | Self-organizing neural network that learns text data structure. Adaptively places nodes (concept representative points) according to data distribution and connects them with edges. Foundation technology for ThinkNavi’s concept structure models |
| MST | Minimum Spanning Tree | Edge network connecting GNG nodes with minimum total distance. Represents the most efficient connection structure between concepts |
| UMAP | Uniform Manifold Approximation and Projection | Non-linear dimension reduction algorithm that compresses high-dimensional data to low dimensions. Preserves local relationships between concepts while reducing to analyzable dimensions |
| Embedding | Embedding | Process of converting text to fixed-length numerical vectors. Semantically similar texts are placed close together in vector space. Uses OpenAI’s text-embedding-3 model |
| Node | Node | Concept representative point placed by GNG through learning. Occupies a specific position in data space with surrounding data (text) assigned to it |
| Cluster | Cluster | Groups of similar concepts (nodes) organized by theme. Classified using algorithms like Ward’s method or K-Means |
| CCM | Connected Concept Model | Multiple models built from different text columns of the same dataset (e.g., concept names and summaries), linked by row index. Enables multi-faceted analysis |
| Personal Model | Personal Model | User-specific concept structure model automatically built from kept chat messages. Functions as AI’s long-term memory |
| Mindware | Mindware | Concept structure model packaged as content. Viewable and purchasable in the store, with knowledge exploration through AI dialogue |
| Concierge | Concierge | AI chat widget automatically built from documents or websites. Embedded on external sites to auto-respond to visitor questions |
| Credit | Credit | Usage unit for AI operations within ThinkNavi. Required credit amounts are defined for each operation. 1 credit ≈ $0.032 |
| Dialektike | Dialektike | Socratic intellectual dialogue mode. AI acts as an equal dialogue partner, deepening thinking based on concept model knowledge. Also called Discourse Dojo |
| Chunk | Chunk | Text fragment. In concept extraction and model building, text is split into chunks before processing |
| Rehydration | Rehydration | Process of restoring saved model state to an engine session. Automatically performed when a session times out |
| Memory Observer | Memory Observer | Auxiliary AI that automatically determines “Does this question need past knowledge?” for chat questions. Triggers context injection from the Personal Model |
| QE | Quantization Error | Indicator of how accurately a node represents actual data. Lower values indicate higher representativeness |
| AHP | Analytic Hierarchy Process | Decision-making method based on multiple criteria. Derives weights through pairwise comparison and scores alternatives |
| SSE | Server-Sent Events | One-way streaming communication from server to client. Used to display chat responses in real-time |
