Service Switch Classifier for Cross-Domain Action Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing systems fail to accurately predict user intent and recommend relevant actions when users switch between different domains, such as from reading news to purchasing related items, due to limited consideration of actions across multiple domains, leading to imprecise predictions and irrelevant recommendations.
Innovation Solution
A service switch classifier is trained using graph embeddings and textual embeddings from past user action sequences across multiple domains to predict domain transitions and recommend actions, incorporating time spans, action types, and content lengths, and utilizing a voter mechanism to combine predictions for precise recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If existing systems predict user intent using limited single-domain actions, then the prediction process is simple and fast, but the prediction accuracy is low and recommendations are irrelevant
Solution Approach 1:
The patent merges action sequences from multiple domains into a unified representation. The service switch classifier integrates user actions across different domains (news, shopping, social networking, etc.) to predict domain transitions, combining heterogeneous data sources to improve prediction accuracy while managing system complexity through structured integration
2Measurement precision
If the system considers actions across multiple domains, then prediction accuracy improves, but computational resources are wasted on processing irrelevant data
Solution Approach 1:
The patent segments the prediction process into distinct components: action sequence generation, graph embedding construction, and service switch classification. This segmentation allows the system to process multi-domain actions efficiently by breaking down the complex task into manageable stages, improving accuracy while reducing computational waste through targeted processing
Solution Approach 2:
The patent introduces graph embeddings as an intermediary representation between raw user actions and final predictions. This intermediary layer transforms heterogeneous multi-domain actions into a unified graph structure that captures relationships between actions, enabling accurate predictions while reducing the computational burden of directly processing raw multi-domain data
3Adaptability or versatility
If the system provides generic recommendations without domain transition prediction, then the recommendation system is simple, but the recommendations are irrelevant to user intent
Solution Approach 1:
The patent implements a dynamic recommendation system where the service switch classifier adapts to user behavior patterns across domains. The system dynamically predicts domain transitions based on current user actions and adjusts recommendations accordingly, providing context-aware relevant recommendations while managing classifier complexity through structured learning from action sequences
Data Source
AI summary
One or more computing devices, systems, and/or methods for cross-domain action prediction are provided. Action sequence embeddings are generated based upon a textual embedding and a graph embedding utilizing past user action sequences corresponding to sequences of past actions performed by users across a plurality of domains. An autoencoder is trained to utilize the action sequence embeddings to project the action sequence embeddings to obtain intent space vectors. A service switch classifier is trained using the intent space vectors. In response to the service switch classifier predicting that a current user will switch from a current domain to a next domain, the current user is provided with a recommendation of an action corresponding to the next domain.