Fast iterative ordering method and fast iterative ordering system for search engines

Abstract

The invention discloses a fast iterative ordering method for search engines. The fast iterative ordering method comprises an offline step and an online step, wherein the offline step comprises the following specific steps: training to obtain a plurality of candidate online ordering models, carrying out storing and proportional distribution on all the ordering models, and storing the ordering models and proportions into a search server cache database at fixed time for reading by the online step; the online step comprises the following specific steps: receiving user requests and distributing the ordering models according to user information, retrieving related files from an index, reading the ordering models in the search server cache database, calculating the priority and returning to users, and counting search behaviors of the users at present time. According to the fast iterative ordering method disclosed by the invention, by a set ordering model description mode, one ordering model can be described by using a character string, and then is sequentially stored into a relational database and a key-value database through a graphical interface and a timing task, so that the persistence of data storage can be guaranteed, and online service can quickly acquire the data.

Description

technical field [0001] The invention relates to the technical field of search sorting, in particular to a fast iterative search engine sorting method and system. Background technique [0002] With the rapid development of big data technology, the use of features in search engine systems is getting more and more in-depth. Text relevance, page rank value and URL link length are all good sorting features. The more features selected, the more likely it is to objectively reflect the user's behavior preferences. Google's search engine ranking system uses as many as more than 200 features, and these features are not simply added linearly, but are represented by a complex neural network, which not only makes full use of each feature of the document, but also The relationship between features can be exploited. However, at this time, it has become unrealistic to expect to manually fit the weights of each feature or even a complex neural network model, and ranking learning technology...

AI Technical Summary

Problems solved by technology

However, the offline learning algorithm needs to interrupt the service every time the algorithm model is updated, and the process is relatively cumbersome, which greatly prolongs the iteration cycle of the project.

Images