Query Recommendation System Using Search Activity Co-occurrence Analysis
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing electronic marketplace search systems face challenges in providing effective search results due to vague or overly specific queries, and lack of explicit feature enumeration in product information, leading to inefficient search refinement and feature identification.
Innovation Solution
A system and method that captures search activity data to infer relationships between queries and identify item features by analyzing search session and selection data, using collaborative filtering and normalization techniques to generate related query indices and provide query recommendations.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If merchants provide detailed descriptive information for items in electronic catalogs, then item information quality improves, but the complexity of indexing and processing this information increases
Solution Approach 1:
The patent replaces manual evaluation and indexing of item features with automated text analysis and natural language processing systems. The system automatically extracts features from descriptive text, images, and other content without requiring manual intervention, thereby maintaining high information quality while reducing processing complexity.
Solution Approach 2:
The system creates structured representations (copies) of unstructured item information by extracting and organizing features into standardized formats. This allows the system to work with simplified data structures that are easier to index and process while preserving the full detail of the original item descriptions.
2Quantity of substance
If the search system provides more search results for vague queries, then the quantity of results increases, but the relevance and helpfulness of results decreases
Solution Approach 1:
The system performs preliminary analysis of item features and search patterns before queries are submitted. By pre-processing item information to extract and organize features, and by analyzing historical search data to understand user intent, the system can quickly retrieve relevant results even for vague queries without requiring multiple refinement steps.
Solution Approach 2:
The patent introduces feature-based intermediaries that bridge the gap between user queries and item results. Instead of directly matching queries to items, the system uses extracted features as intermediate representations that enable more accurate matching, thereby improving result relevance while maintaining quantity.
3Measurement precision
If the search system accepts more specific queries, then search precision improves, but the quantity of relevant search results decreases
Solution Approach 1:
The system dynamically adjusts search parameters and feature weights based on the specificity of the query. For highly specific queries, the system modifies which features are prioritized and how strictly they are matched, allowing it to maintain result quantity by finding items that match the essential features while being more flexible on less critical attributes.
4Measurement precision
If merchants manually evaluate item information to identify features, then feature identification accuracy improves, but the time and resources required increase significantly
Solution Approach 1:
The patent replaces manual feature evaluation with automated text analysis systems that use natural language processing, machine learning, and pattern recognition to extract features from item descriptions, images, and other content. This automation maintains high accuracy while eliminating the time and resource costs of manual evaluation.
Solution Approach 2:
The system performs feature extraction continuously and automatically as items are added or updated in the catalog, rather than requiring discrete manual evaluation steps. This ongoing automated process ensures features are always current without requiring dedicated time and resources for periodic manual reviews.
Data Source
AI summary
Embodiments may include a system configured to receive search session data that indicates, for each of multiple search sessions performed by a respective user, multiple search queries submitted by that user during the search session. The system may also receive search selection data that indicates, for each of multiple search queries resulting in a set of search results, a particular item selected from that set of search results by a respective user. The system may be configured to perform a co-occurrence analysis on the search data in order to generate one or more search indices that specify, for a given search query, one or more search queries determined to be related to the given search query according to the co-occurrence analysis. The system may be configured to process a client request for related queries that are related to a query of interest submitted by the user within that request.