Neural Network Sorting Model for Search Result Relevance
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current search engines face challenges in accurately reflecting user search demands, leading to increased computing load and reduced information-obtaining efficiency due to inadequate sorting of search results, especially when queries include content word pairs and relationship words.
Innovation Solution
A method and apparatus for building a neural network-based sorting model using search logs with relationship triples, where the model is trained on query-clicked title pairs to sort search results based on the description of content word pair relationships, improving the relevance of search results and user experience.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If search results are sorted using traditional methods, then the sorting process is simple, but the search results fail to accurately reflect user search demands, requiring users to spend more time and request more computing resources
Solution Approach 1:
The patent applies preliminary action by pre-building a sorting model using relationship triples extracted from search logs before actual search operations. The model is trained in advance with relationship data (content word pairs and their relationships) so that during search, results can be quickly sorted based on pre-learned patterns, improving both accuracy and efficiency
Solution Approach 2:
The patent introduces an intermediary sorting model that acts as a mediator between the search engine and user queries. This model uses relationship triples as intermediate representations to bridge the gap between raw search data and meaningful result排序, enabling accurate reflection of user intent without requiring complex real-time analysis
2Loss of time
If users manually search through unsorted or poorly sorted results, then no additional computing resources are needed for sorting, but users expend more time performing operations such as pulling down the page or turning pages
Solution Approach 1:
The patent implements feedback by using search log data (including user click behavior on search results) to continuously train and improve the sorting model. The relationship triples are extracted from actual user search patterns, creating a feedback loop where user behavior informs future sorting improvements, reducing time loss over time
Solution Approach 2:
By pre-processing search logs to extract relationship triples and training the sorting model in advance, the system prepares sorting capabilities before users need them. This preliminary action ensures that when users perform searches, the sorting is already optimized based on historical data, minimizing their time expenditure
Data Source
AI summary
The present disclosure provides a method of building a sorting model, and an application method and apparatus based on the model. The method of building a sorting model comprises: obtaining, from a search log, a query including a relationship triple and a clicked title of a search result corresponding to the query, wherein the relationship triple includes a content word pair and a relationship word of the content word pair; obtaining training data using the obtained query, the clicked title corresponding to the query, and times of click of the clicked title; using the training data to train a neural network-based sorting model, the sorting model being used to sort sentences according to the sentences' description of a relationship of the content word pair. The sorting model may be used to implement the following applications: sorting the search results of the query comprising the relationship triple, determining the sentence describing the relationship of the content word pair; upon displaying the relevant content word with respect to the query including the content word, the search engine can further display the sentence describing the relationship between the relevant content word and the searched content word.