Keyword Recommendation System for E-commerce Search Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
E-commerce websites face challenges in matching product titles with search queries, leading to inefficient product discovery and increased workload for search users and engines, as inappropriate keyword selection reduces the visibility of product information in search results.
Innovation Solution
A keyword recommendation system that parses product titles into elements, correlates them with historical search queries, and generates a ranked list of keywords based on supply and demand, Gini, and industry indices to enhance title relevance, thereby increasing the likelihood of product information being displayed in search results.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If seller users manually select keywords for product titles, then the product information can be submitted quickly, but the keywords may not match search queries effectively, reducing product visibility
Solution Approach 1:
The system performs preliminary analysis of historical search queries and product titles to pre-calculate relevance relationships before actual product submission. By establishing a query library with pre-computed relevance scores in advance, sellers can quickly submit products without manual keyword selection while still achieving accurate matching, as the relevance evaluation has already been prepared beforehand.
Solution Approach 2:
The patent introduces an intermediary relevance evaluation mechanism between product titles and search queries. Instead of direct manual matching or simple keyword overlap, the system uses pre-computed relevance relationships from historical data as an intermediary layer to bridge products and queries, enabling both efficient submission and accurate matching through this intermediate relevance assessment layer.
2Measurement precision
If search users repeatedly revise their search queries to find desired products, then they can improve search accuracy, but their workload increases and search efficiency decreases
Solution Approach 1:
The system performs preliminary relevance evaluation by pre-analyzing historical search queries and their relationships with product titles. This advance preparation creates a query library with pre-computed relevance scores, allowing the system to quickly retrieve and rank relevant products without requiring users to repeatedly revise queries, thus maintaining high search accuracy while reducing time loss.
Solution Approach 2:
The system implements feedback mechanisms by analyzing historical search behavior and query revisions to continuously improve the relevance evaluation model. By incorporating feedback from actual search patterns and user behavior, the system learns to prioritize more relevant products in search results, reducing the need for query revisions and decreasing search time while maintaining or improving accuracy.
3Measurement precision
If the search engine processes multiple query revisions from users, then search accuracy can be improved, but the workload for the search engine increases
Solution Approach 1:
The system performs preliminary processing by pre-analyzing historical search queries and building a query library with pre-computed relevance relationships with product titles. This advance preparation significantly reduces the computational workload during actual search operations, as the system can quickly retrieve and rank products based on pre-established relevance scores rather than processing multiple query revisions in real-time.
Solution Approach 2:
The patent creates a copied and simplified representation of the complex search problem by building a query library that captures essential relevance relationships from historical data. This copied structure allows the search engine to work with pre-processed relevance information rather than re-processing raw queries and products, significantly reducing computational complexity while maintaining search accuracy.
Data Source
AI summary
Recommending keywords is disclosed, including: receiving a set of product information including a product title; extracting and parsing the product title into a set of parsed elements; finding a plurality of candidate keywords corresponding to at least a subset of the set of parsed elements based at least in part on stored mappings between parsed data and keywords; determining a plurality of composite correlation scores for corresponding ones of the plurality of candidate keywords; sorting at least a subset of the plurality of candidate keywords into a ranked list based on at least a subset of the plurality of composite correlation scores; and selecting a set of one or more keywords to recommend from the plurality of candidate keywords based at least in part on the ranked list.