Dynamic Interaction Graph Transformer for Sequential Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods struggle to comprehensively express the features of interaction objects in sequential interaction data, as they fail to accurately consider the dynamic changes and mutual influences between interaction parties in user interaction events.
Innovation Solution
A dynamic interaction graph is constructed based on chronological interaction sequences, where nodes represent interaction objects, and a Transformer-based neural network model processes node sequences to generate feature vectors that incorporate positional encoding and correlation between nodes, enabling the extraction of deep features from interaction objects.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional methods are used to process interaction events, then the processing is simple, but the feature expression of interaction objects is incomplete and fails to capture dynamic changes
Solution Approach 1:
The patent applies dynamics by constructing a dynamic interaction graph that evolves over time, where nodes represent interaction objects and edges represent interactions. The graph is updated incrementally as new interaction events occur, allowing the model to capture dynamic changes in object states and relationships. This resolves the contradiction by introducing temporal dynamics to improve feature expression without requiring complete reprocessing of historical data.
Solution Approach 2:
The patent segments the interaction history into discrete interaction events, each represented as a feature group with specific attributes. By dividing the continuous interaction stream into event-based units, the model can process and analyze individual interactions independently while maintaining the overall sequence, thereby improving feature precision without overwhelming computational complexity.
2Measurement precision
If comprehensive features of interaction parties are considered, then the feature accuracy improves, but the processing complexity increases significantly
Solution Approach 1:
The patent introduces a temporal dimension by organizing interaction events in chronological order and using positional encoding to represent time steps. This transforms the problem from analyzing complex multi-dimensional interaction features to analyzing sequences along the time dimension, which can be efficiently processed by Transformer architectures. The temporal dimension allows comprehensive feature consideration while maintaining manageable model complexity through sequential processing.
Solution Approach 2:
The patent uses an attention mechanism as an intermediary between the raw interaction features and the final feature representation. The attention mechanism selectively weights different interaction events and features based on their relevance, automatically filtering and prioritizing information. This intermediary layer enables comprehensive feature consideration while reducing complexity by focusing computational resources on the most important features and interactions.
3Reliability
If the dynamic changes of interaction parties are captured, then the feature representation improves, but the data processing complexity increases
Solution Approach 1:
The patent performs preliminary action by pre-processing interaction events into standardized feature groups with consistent schemas before building the dynamic interaction graph. Each interaction event is transformed into a structured format with defined attributes, and the graph construction follows a predetermined pattern where nodes and edges are created according to fixed rules. This preliminary structuring simplifies the overall processing complexity while ensuring reliable capture of dynamic changes.
4Measurement precision
If positional encoding is used to capture time sequence relationships, then the temporal feature extraction improves, but the computational overhead increases
Solution Approach 1:
The patent applies parameter changes by using learnable positional encoding vectors that are optimized during model training. Instead of using fixed or complex temporal features, the model learns optimal positional representations that capture temporal relationships efficiently. This approach improves temporal feature extraction accuracy while keeping computational overhead manageable through parameter optimization rather than algorithmic complexity.
Data Source
AI summary
This disclosure relates to processing sequential interaction data through machine learning. In one aspect, a method includes obtaining a dynamic interaction graph constructed based on a dynamic interaction sequence. The dynamic interaction sequence includes interaction feature groups corresponding to interaction events. Each interaction feature group includes a first object, a second object, and an interaction time of an interaction event that involved the first object and the second object. The dynamic interaction graph includes multiple nodes including, for each interaction feature group, a first node that represents the first object of the interaction feature group and a second node that represents the second object of the interaction feature group. A current sequence corresponding to a current node to be analyzed is determined. The current sequence is input into a Transformer-based neural network model. The neural network model determines a feature vector corresponding to the current node.