--- title: 'ETDNet: Edge-Type Decoupling in Temporal Graphs' url: https://www.emergentmind.com/topics/edge-type-decoupled-network-etdnet type: topic --- # ETDNet: Edge-Type Decoupling in Temporal Graphs Searching arXiv for the ETDNet paper and closely related context papers mentioned in the provided data. arXiv Search Query: "Full-History Graphs with Edge-Type Decoupled Networks for Temporal Reasoning" Edge-Type Decoupled Network (ETDNet) is a temporal-graph neural architecture introduced alongside the full-history graph (FHG) representation for reasoning over evolving relational data in which prediction depends jointly on **who interacts with whom** and **how those interactions unfold over time**. Its motivating applications are driver-intention prediction in traffic and fraud detection in Bitcoin transaction networks. The defining premise is that temporal graphs should not be handled merely as a sequence of snapshots, because same-time interactions and cross-time continuity encode different semantics. ETDNet therefore operates on a single graph containing the entire observed history, and processes two disjoint edge families with separate modules: intra-time-step edges for structural context and inter-time-step edges for temporal propagation [2508.03251]. ## 1. Conceptual basis and problem setting ETDNet is designed for temporal reasoning on evolving relational data where plain sequence models are inadequate because each entity’s trajectory is influenced by other entities, while many temporal graph methods remain constrained by a **sequence of snapshots**. In a typical snapshot pipeline, one runs a GNN separately at each time step and then feeds snapshot embeddings into an RNN or Transformer. The ETDNet paper argues that this is limiting for two stated reasons: aggregating events into snapshots can blur fine-grained ordering and long-range temporal signals, and distant temporal dependencies may require many stacked graph or temporal layers, aggravating over-smoothing and over-squashing [2508.03251]. A second motivation is semantic heterogeneity. ETDNet treats instantaneous relational structure and temporal continuity as fundamentally different categories of edges. Same-time interactions encode local relational context, such as nearby vehicles or contemporaneous transaction flow. Cross-time edges encode persistence, causality, or propagation. The architecture’s central claim is therefore not merely that edges matter, but that **different edge semantics should not be forced through the same message function**. A common misconception is to conflate ETDNet with the full-history graph itself. The paper distinguishes them explicitly. The **full-history graph** is the graph construction that internalizes time into topology, whereas **ETDNet** is the neural encoder that learns on that graph. The representation and the network are separate contributions, although the paper’s empirical argument is that they are most effective when aligned. ## 2. Full-history graph formulation The full-history graph unfolds the entire observation window into a single graph. Time is defined on a discrete set $$ T=\{0,1,\dots,\tau_{\max}\}. $$ Entities are partitioned into dynamic entities \(U\) and static entities \(S\). For each dynamic entity \(u\in U\) and each time step \(t\in T_u\subseteq T\) where it exists, the graph contains a node \(u^t\) with feature vector $$ \mathbf{x}_{u,t}\in\mathbb{R}^{d}. $$ Each static entity \(s\in S\) is represented once with feature \(\mathbf{x}_s\). The intended node set is $$ \widetilde V = \{\,u^{t}\mid u\in U,\; t\in T_u\,\}\;\cup\; S. $$ The graph contains two disjoint edge families. Intra-time-step edges $$ D\subseteq \{(v^t,w^t)\mid t\in T\} $$ connect entities coexisting in the same time step. Inter-time-step edges $$ \mathcal H \subseteq \{(v^t,w^{t+1})\mid t<\tau_{\max}\} $$ connect temporally successive events. The canonical case is a self-link \((u^t,u^{t+1})\), but the paper explicitly generalizes temporal edges to cross-entity hand-offs such as \((p^t,q^{t+1})\). The resulting graph is $$ G=(\widetilde V,E),\qquad E=D\cup\mathcal H. $$ This construction internalizes time into graph topology itself. Paths along \(\mathcal H\) are acyclic and ordered forward in time, while \(D\) remains local to a time step and may contain cycles. A plausible implication is that temporal causality becomes explicit at the graph level rather than being imposed only by recurrent or Transformer-style ordering outside the graph. | Element | Formal definition | Semantics | |---|---|---| | Dynamic node | \(u^t\) | Entity occurrence at time \(t\) | | Static node | \(s\) | Time-invariant context | | Intra-time-step edge | \(D\) | Same-frame structure | | Inter-time-step edge | \(\mathcal H\) | Temporal continuity or propagation | The distinction between self-temporal edges and cross-entity temporal hand-offs is especially important. In traffic, \(\mathcal H\) is used as a self-chain across frames. In Elliptic++, \(\mathcal H\) connects a transaction to later transactions that spend its outputs. ETDNet is therefore not restricted to per-entity temporal traces; it can encode directed temporal flow between distinct entities. ## 3. Architecture and update mechanism ETDNet stacks \(L\) identical layers. The initial embedding is $$ h_x^{(0)}=\mathbf{x}_x. $$ At each layer, one branch performs structural aggregation over \(D\), another performs temporal aggregation over \(\mathcal H\), and a fusion layer combines both messages with the previous embedding. The architecture is called **edge-type decoupled** because the two edge families are processed with separate modules and separate parameters rather than inside a single aggregation rule [2508.03251]. | Component | Operates on | Function | |---|---|---| | Step Attention (SA) | \(D\) | Structural aggregation within a time step | | History Attention (HA) | \(\mathcal H\) | Temporal attention over reachable history | | Fusion Layer (FL) | Previous state + SA + HA | Residual learned combination | ### Step Attention For a dynamic node \(u^t\), the intra-step neighborhood is $$ \mathcal N_D(u^t)=\{\,v^t\mid (v^t,u^t)\in D\}. $$ SA uses \(K_s\) stacked sublayers, each with \(H_s\) attention heads, so a single ETD layer can expand the same-timestep receptive field to \(K_s\) hops. For head \(r\) and sublayer \(k\), the attention score is $$ e_{uv}^{(r,k)} = \mathrm{LeakyReLU}\!\Bigl( \mathbf{a}_{r,k}^{\!\top} \bigl[\, \mathbf{W}_{Q,r,k}\,h^{(l,k-1)}_{u^{t}} \;\Vert\; \mathbf{W}_{K,r,k}\,h^{(l,k-1)}_{v^{t}} \,\bigr]\Bigr), $$ with normalization $$ \alpha_{uv}^{(r,k)} = \frac{\exp(e_{uv}^{(r,k)})}{ \sum_{w^{t}\in\mathcal{N}_{D}(u^{t})}\exp(e_{uw}^{(r,k)})}, $$ and head-specific message $$ m_{u^t}^{(r,k)} = \sum_{v^t\in\mathcal N_D(u^t)} \alpha_{uv}^{(r,k)}\mathbf W_{V,r,k}h_{v^t}^{(l,k-1)}. $$ After concatenating heads, SA applies output projection, residual addition, and LayerNorm: $$ h^{(l,k)}_{u^t}= \mathrm{LayerNorm}\!\Bigl( \mathbf O_k[m_{u^t}^{(1,k)}\Vert\cdots\Vert m_{u^t}^{(H_s,k)}] +h_{u^t}^{(l,k-1)} \Bigr). $$ After \(K_s\) sublayers, the structural message is $$ m^{D}_{u^t}:=h^{(l,K_s)}_{u^t}. $$ The module is GAT-like in form, but its scope is restricted to the intra-time-step edge family \(D\), and its internal \(K_s\)-stack is intended to capture richer same-frame structure before temporal fusion. ### History Attention For the current node \(u^t\), HA attends over a bounded predecessor set $$ \mathcal{P}_{B}(u^{t}) =\Bigl\{\,w^{\tau}\;\Big|\; w^{\tau}\!\in\!\operatorname{Hop}_{\le B}(\mathcal{H},u^{t}),\; \tau