Temporal Knowledge Graph Recommender for Dynamic Relations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing knowledge graph-based recommender systems fail to account for dynamicity and the time aspect, leading to information overload and inability to reliably predict outcomes due to their inability to handle changing relations and time-dependent actions.
Innovation Solution
A method and system that utilize a temporal knowledge graph to predict future entities, links, and attributes, simulate actions across multiple time steps, and classify outcomes, providing a ranked list of recommended actions based on state of interest and model uncertainty.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Quantity of substance
If existing knowledge graph-based recommender systems process a big amount of relational data, then they can handle multi-modality and multi-dimensionality, but they cannot consider the time aspect and dynamicity of relations
Solution Approach 1:
The patent transforms static knowledge graphs into dynamic temporal knowledge graphs where relations and attributes are time-stamped. The system models dynamicity by allowing nodes and edges to appear, disappear, and change types over time, with each relation carrying temporal information about when it became active and when it ceased to be active. This enables the system to adapt to changing relations while processing large amounts of data.
Solution Approach 2:
The patent adds a temporal dimension to the existing knowledge graph structure. By introducing time as an additional dimension with timestamps for entities, relations, and attributes, the system transforms a 3-dimensional knowledge graph (entities, relations, attributes) into a 4-dimensional temporal knowledge graph that includes time. This allows the system to handle dynamic relations without sacrificing the ability to process large amounts of multi-modal and multi-dimensional data.
2Speed
If existing recommender systems recommend actions based on current state, then they can provide immediate recommendations, but they cannot predict when to best give recommendations or when future situations will arise
Solution Approach 1:
The patent performs preliminary actions by predicting future states and actions before they actually occur. The system uses the temporal knowledge graph to forecast future entities, relations, and attributes, and to identify potential future situations. By preparing recommendations in advance based on predicted future states, the system can provide timely recommendations at the optimal moment rather than reacting too late or too early.
Solution Approach 2:
The patent implements feedback mechanisms by continuously comparing predicted future states with actual observed states. The system uses time-stamped information to track how situations evolve and adjusts its predictions and recommendations based on this feedback. This allows the system to maintain accurate predictions of when future situations will arise and when recommendations should be provided, reducing information loss about timing.
3Adaptability or versatility
If humans process information in recommender systems, then they can understand complex relations, but they suffer from information overload and cannot reliably predict outcomes
Solution Approach 1:
The patent introduces an intermediary computational layer between the raw temporal knowledge graph data and human decision-making. The system automatically processes the complex temporal relations and multi-dimensional data using algorithms that can handle the full quantity of information without human cognitive limitations. This intermediary system performs the heavy lifting of analyzing temporal patterns and predicting outcomes, while presenting simplified, actionable recommendations to humans, thus resolving the information overload problem while preserving the ability to understand complex relations.
4Productivity
If AI-based recommender systems process data, then they can handle large volumes efficiently, but they cannot take into account relations between entities and multiple dimensions
Solution Approach 1:
The patent segments the complex task of processing temporal knowledge graphs into manageable components. The system divides the data into discrete time steps and processes each timestamp separately, breaking down the multi-dimensional relational data into structured triples (subject, predicate, object) with associated timestamps. This segmentation allows AI algorithms to efficiently process large volumes of data while maintaining the ability to capture relations and multiple dimensions through the organized temporal structure.
Solution Approach 2:
The patent changes the parameters of the knowledge graph representation by introducing temporal parameters (timestamps for entities, relations, and attributes). This transformation allows the system to maintain efficient AI-based processing while incorporating relational and multi-dimensional information. The temporal parameters provide a structured way to encode complex relations and dimensions that AI algorithms can process efficiently, resolving the contradiction between processing efficiency and adaptability.
Data Source
AI summary
A method for providing recommendations to users based on a state of interest is provided. The method includes organizing domain of interest information in an initial temporal knowledge graph, wherein t is a timestamp that refers to a present point in time. The method predicts, for at least one future point in time (t+x), future entities, future links between entities and/or future attributes of entities for the initial knowledge graph and produces at least one new knowledge graph based on the predictions, simulates situations resulting from the execution of a particular action or a combination of actions at certain points in time and predicting expected temporal knowledge graphs for the simulated situations, and classifies the knowledge graphs produced for the respective points in time and for the simulated situations based on the state of interest. A ranked list of recommended actions is provided based on the classification result.