Dialogue Path Knowledge Graph for Information Pushing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current product recommendation systems in electronic commerce, particularly dialogue recommendation systems, face challenges in effectively extracting user preferences from dialogue information and determining relevant product recommendations, leading to inefficiencies in information pushing and reduced user engagement.
Innovation Solution
A method and apparatus that extract preference attributes from user dialogue information, map them to a knowledge graph, generate a dialogue path based on time sequence, and predict a push strategy using a pre-trained model to determine and push relevant attribute or product information, improving coherence and relevance of pushed content.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If a dialogue recommendation system maps all user preferences to a vector space and takes all attributes as candidate attributes, then the system can comprehensively cover user preferences, but the dimensionality of candidate attributes becomes too high, reducing efficiency and making it difficult to determine relevant recommendations
Solution Approach 1:
The patent segments the high-dimensional candidate attribute space by constructing a dialogue path that orders attributes according to dialogue time sequence. This divides the overwhelming set of all attributes into a structured sequence along the dialogue path, making it manageable to process and select from relevant attributes at each step rather than facing all attributes simultaneously.
Solution Approach 2:
The patent extracts only the attributes that lie on the dialogue path from the complete set of candidate attributes. By taking out and focusing on this specific subset of attributes that are actually relevant to the current dialogue context, the system reduces dimensionality while maintaining comprehensiveness for the relevant scope.
2Reliability
If the system considers all attributes as candidate attributes for recommendation, then it can ensure completeness of recommendation options, but the training efficiency of the strategy prediction model decreases due to excessive action categories
Solution Approach 1:
The patent segments the action space by defining actions only along the dialogue path. Instead of considering all possible attributes as potential actions, the system divides the action space into a sequence of attributes that appear on the dialogue path, significantly reducing the number of action categories while maintaining reliability for the relevant recommendation scope.
Solution Approach 2:
The patent makes the candidate attribute set dynamic by determining it based on the current dialogue path rather than using a fixed set of all attributes. The candidate attributes change according to the dialogue context and progression, allowing the model to adapt to different dialogue states with a reduced and relevant action space.
3Adaptability or versatility
If the system pushes information based on all candidate attributes, then it can provide comprehensive product recommendations, but the pertinence and relevance of pushed information decreases due to information overload
Solution Approach 1:
The patent extracts only the attributes that are relevant to the current dialogue context by identifying them on the dialogue path. This extraction process filters out irrelevant attributes and focuses on the specific attributes that matter for the current recommendation task, thereby improving pertinence and relevance while maintaining comprehensive coverage of relevant options.
Solution Approach 2:
The patent applies local quality by making the candidate attribute set specific to each dialogue context rather than uniform across all scenarios. The candidate attributes are determined locally based on the dialogue path, ensuring that the pushed information is highly relevant to the specific user needs and dialogue state at each moment.
4Adaptability or versatility
If the system uses a large number of candidate attributes for information pushing, then it can cover more user preferences, but the coherence of pushed information decreases making it harder to maintain user engagement
Solution Approach 1:
The patent segments the attribute selection process by enforcing ordering along the dialogue path. This segmentation creates a coherent sequence of attribute discussions that follows the natural flow of dialogue, maintaining stability and coherence in the pushed information while still covering comprehensive user preferences through the structured progression along the path.
Data Source
AI summary
An information pushing method and apparatus. The method includes extracting preference attributes of a user from user dialogue information in the current dialogue scene; determining effective attribute nodes corresponding to the preference attributes in a preconstructed knowledge graph; arranging the effective attribute nodes according to a dialogue time sequence to generate a dialogue path; determining a candidate attribute set and a candidate commodity set on the basis of the dialogue path; using a pretrained strategy prediction model to predict the current pushing strategy on the basis of the current state vector; determining, on the basis of the current pushing strategy, an object to be pushed from the candidate attribute set or the candidate commodity set, and generating, on the basis of said object, information to be pushed; and pushing the information to be pushed.