Search Intent Classification via Machine Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Vertical search engines face challenges in providing relevant search results tailored to user intent while maintaining the convenience of a single search box, as they often return content from multiple classes regardless of the user's intended search class.
Innovation Solution
A supervised machine learning algorithm, such as a neural network or SVM, is used to identify the likely class of content corresponding to the user's search intent, limiting and formatting search results only if the confidence rating exceeds a threshold, based on normalized search results and class-specific information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If a single search box is provided to allow users to search across multiple classes of content, then ease of operation is improved, but search result relevance deteriorates because results from all classes are returned regardless of user intent
Solution Approach 1:
A machine learning classifier is introduced as an intermediary component between the search box and the search results. The classifier receives the user query and predicted search intent as input, and outputs a predicted content class that mediates which results are ultimately displayed. This intermediary layer enables the system to maintain a single search box for ease of operation while filtering results to improve relevance based on predicted user intent.
Solution Approach 2:
The system uses feedback from multiple sources including predicted search intent (derived from query analysis), result counts from different content classes, and relevance scores to dynamically adjust which results are displayed. This feedback mechanism allows the system to learn from the data and improve its ability to return relevant results while maintaining the convenient single-search-box interface.
2Reliability
If search results are limited to a single content class based on predicted user intent, then search result relevance is improved, but device complexity increases due to the need for machine learning algorithms
Solution Approach 1:
The machine learning classifier is designed to handle multiple content classes (programme content, support, products, and future classes) through a single universal model. This multi-functional approach allows the system to improve search result relevance across different content types without proportionally increasing complexity, as the same classifier infrastructure serves multiple purposes.
Solution Approach 2:
The system changes parameters such as confidence thresholds and result count thresholds to control the behavior of the search engine. By adjusting these parameters, the system can balance between returning highly relevant results (higher threshold) and maintaining simpler operation (lower threshold), thereby managing the trade-off between reliability and complexity without fundamental architectural changes.
3Measurement precision
If a high confidence threshold is set for class prediction, then search result precision is improved, but productivity decreases due to fewer results being returned
Solution Approach 1:
The system applies partial action by selectively limiting results only when the confidence threshold is met, rather than always filtering. When confidence is high, results are limited to improve precision. When confidence is low or thresholds aren't met, the system falls back to returning broader results, thereby maintaining productivity. This partial application of filtering resolves the contradiction between precision and productivity.
Data Source
AI summary
A computer-implemented method of determining search intent, comprises: receiving a search query; searching content across a plurality of content classes using the search query, so as to obtain a plurality of search results; deriving summary data from the search results; applying the summary data to a trained machine learning model; and determining from the machine learning model a selected one of the content classes corresponding to the search intent of the search query.