Conversational Recommendation Embeddings for Fewer User Queries
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Traditional conversational recommendation systems, such as chatbots, are inefficient in determining user preferences for product recommendations, leading to increased computing resource consumption and underperformance due to the need for multiple user queries to fulfill user intent.
Innovation Solution
A conversational recommendation system that embeds user preferences and item embeddings in a target domain, allowing for the selection of items based on proximity to positive preferences and distance from negative preferences, thereby reducing the need for multiple queries.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If traditional ML-based chatbots are used for recommendations, then they can provide conversational interactions, but they consume large computing resources and require multiple queries to fulfill user intent
Solution Approach 1:
The system segments the recommendation process into two distinct components: (1) a lightweight preference extraction module that captures user preferences from conversational input, and (2) an embedding-based recommendation engine that retrieves items based on preference embeddings. This segmentation eliminates the need for large ML models to process every recommendation query, significantly reducing computing resource consumption while maintaining conversational capability.
Solution Approach 2:
The patent introduces preference embeddings as an intermediary representation that bridges user preferences and item recommendations. Instead of using complex ML models to directly map conversations to recommendations, the system first extracts preferences and represents them as embeddings, which then serve as intermediaries for efficient item retrieval. This intermediary approach reduces the computational burden while improving recommendation accuracy.
2Measurement precision
If multiple queries are required to fulfill user intent, then the system can gather more preference information, but the interaction time and number of steps increase
Solution Approach 1:
The system performs preliminary preference extraction by analyzing user input and extracting preferences before the recommendation generation process. The preference extraction module proactively identifies and captures user preferences from conversational input, storing them as embeddings that can be immediately used for recommendation retrieval. This preliminary action eliminates the need for multiple back-and-forth queries to gather preference information, reducing interaction time while maintaining precision.
3Use of energy by moving object
If embedding-based recommendation is used, then computing resource consumption is reduced, but the system must effectively isolate and represent user preferences in a target domain
Solution Approach 1:
The patent extracts user preferences from conversational input by identifying and isolating preference-related information. The preference extraction module specifically targets and extracts preference expressions from user queries, separating them from other conversational elements. This extraction process transforms unstructured conversational data into structured preference representations that can be efficiently stored and queried in embedding space, reducing the complexity of preference representation.
Data Source
AI summary
Aspects of the disclosed technology include computer-implemented systems and methods for conversational recommendation systems, such as conversational chatbots that are configured to process user queries and generate responses. A recommendation system can receive a user query, provide a recommendation response, and solicit feedback from a user in a target domain. The system can display a first set of items and receive inputs indicative of preferences relative to the first set of items. The system can generate preference embeddings in an embedding space of the target domain based at least in part on the preferences and compare the preference embeddings with item embeddings in the embedding space. The system can select content items based at least in part on a distance between the preference embeddings and the item embeddings in the target embedding space and generate data for displaying the selected content items via the user interface.