Prompt Description

Okay, here are the complete Chinese prompts for the five core paper figures prepared for you. You can directly copy and paste them into AI-based image generation tools such as Boardmix AI. Figure 1: Overall Model Architecture Diagram Prompt: "Generate a scientific architecture diagram for an artificial intelligence academic paper, with a style similar to top journals, simple and modern. The main colors are indigo, teal, and slate gray. Layout: Horizontal flowchart, information flow from left to right. Left Input: Two parallel input streams: "Global Historical Information" (cube stack on the left) and "Local Historical Information" (cube stack on the right). There is a "Snapshot Sequence" network icon connecting the central path above. Central Core: The upper path is the "Multi-Hop Aggregation Module", with a hypergraph structure drawn inside (a central node connected to multiple nodes). The lower path is the "Local Periodicity-Aware Evolution Module", with a circular arrow icon. Both modules receive input from both sides. Right Output: The outputs of the two modules converge into the "Contrastive Learning Module" on the right. This module outputs two ellipses: "Entity Prediction Loss" and "Contrastive Learning Loss". There is a "Prediction Target" box at the top pointing to this module. Style: Use solid line arrows. The upper path is light blue, and the lower path is light green. All modules are rectangular and have aligned text." Figure 2: Semantic Initialization and Dynamic Hypergraph Construction Prompt: "Create a two-column (A and B) concept map. Column A (Semantic Initialization): The text '(Subject Entity, Relation, Object Entity)' on the left enters the 'BERT' box, and the output feature vector passes through the 'Linear Projection Layer' to generate 'Semantically Enhanced Initial Embeddings' (a set of nodes and edges). Column B (Dynamic Hypergraph Construction): Draw a timeline (t-2, t-1, t_query). At each timestamp, the standard graph on the left becomes a hypergraph on the right: a set of nodes is connected by colored block hyperedges (labeled 'Historical Facts'). An arrow points from 'Semantically Enhanced Initial Embeddings' to the nodes, and the 'Query Entity' at time t_query is highlighted. Style: Minimalist flat design." Figure 3: Local Path - Spatio-Temporal Hypergraph Convolution Prompt: "Design a detailed schematic diagram with the title 'Spatio-Temporal Hypergraph Convolution in Local Paths'. Main Module: 'Local Query View Hypergraph Convolution' rectangle. Inside: Divided into two parallel vertical streams: 1. Left Stream (Spatial Dimension): Draw a small hypergraph, with a 'Multi-Head Attention' icon (stacked squares) pointing to it, labeled 'Structure Aggregation'. 2. Right Stream (Temporal Dimension): Draw a clock icon and waveform, with an 'Enhanced Temporal Encoder' box pointing to it, labeled 'Time Aggregation with Exponential Decay'. The outputs of the two streams converge into the 'Dynamic Fusion Module' (circle with a + sign), and then pass through the 'Gated Fusion Module' (with a gate icon, and input 'Original Query Embedding'), outputting a 'Local Temporal Embedding' vector. Style: Blueprint style, light blue for the left stream, light orange for the right stream." Figure 4: Global Path - Multi-Hop Sampling and Reasoning Prompt: "Generate a multi-step diagram. Layout: Central node 'Query entity at t_query'"