Personalized Query Autocomplete Using Purchase-History Reranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing query auto-completion systems on e-commerce sites lack personalization based on user purchase history, failing to optimize search query suggestions effectively.
Innovation Solution
A mechanism that reranks search query candidates using a natural language processing model trained on user purchase history, incorporating context from the user's current search session to provide personalized suggestions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If query auto-completion is applied to an EC site without considering purchase history, then the system is simpler to operate, but the personalization and relevance of search query suggestions deteriorate
Solution Approach 1:
The system performs preliminary action by training the natural language processing model in advance using user purchase history data. This pre-training enables the model to capture user preferences and purchasing tendencies before actual search queries are made, allowing for personalized suggestions without adding operational complexity during user interaction.
Solution Approach 2:
The system implements self-service by automatically utilizing user purchase history data to train and update the natural language processing model. The model serves itself by continuously learning from user behavior patterns, enabling personalized query suggestions without requiring manual configuration or complex user input beyond the initial purchase history data.
2Reliability
If a sorting model is trained based on past prefixes and selected queries, then search query candidates can be reranked according to past search tendencies, but the system cannot effectively incorporate purchase-related information for more personalized reranking
Solution Approach 1:
The system applies parameter changes by transitioning from a traditional sorting model that processes only query prefixes to a natural language processing model that processes both query candidates and contextual information. This parameter change enables the model to incorporate purchase history, search context, and other relevant factors, significantly improving adaptability to various data sources while maintaining high relevance in search query recommendations.
3Manufacturing precision
If purchase history information is incorporated into the reranking mechanism, then personalized search query suggestions improve, but the device complexity increases
Solution Approach 1:
The system introduces an intermediary approach by using a pre-trained natural language processing model as a mediator between raw purchase history data and search query suggestions. This intermediary model processes and transforms complex purchase history information into meaningful representations that can be directly used for personalized reranking, simplifying the overall system architecture while maintaining high personalization capability.
Data Source
Figure 1
Figure 2
Figure 3~4A
AI summary
An information processing apparatus acquires a prefix input on an EC (E-Commerce) site by a user, acquires a context representing a feature relating to a search on the EC site by the user, generates a plurality of search query candidates, based on the prefix, and reranks the plurality of search query candidates, using a result generated by inputting the plurality of search query candidates and the context to a natural language processing model trained using information relating to purchases on the EC site by the user.