Search result sorting method and device, storage medium, electronic device

Abstract

The present invention provides a search result sorting method and device, a storage medium, and an electronic device, wherein the method includes: generating offline feature data and online feature data according to sample data, and using the offline feature data to include an initial feature extractor and a second feature extractor. The first combination model of a classifier is trained in parallel offline features to obtain the trained feature extractor, and the trained feature extractor is kept unchanged, and the online features are used to perform online real-time training on the second combination model composed of it and the second classifier to obtain The second model: receiving a job search request and obtaining a corresponding search result list, and using the second model to reorder multiple pieces of candidate job information records in the search result list. The present invention solves the technical problem of unreasonable sorting of the search list in the related art, shortens the training time of the model while reducing the memory resources occupied by the training model, and improves the model training efficiency.

Description

technical field [0001] The present invention relates to the field of computer technology, in particular to a search result sorting method and device, a storage medium, and an electronic device. Background technique [0002] In related technologies, sorting is a very critical part of the search business. Its main function is to sort the candidate information recalled from the massive information according to the degree of relevance to the search field, so that the information that is more in line with the user's intention Arrange as high as possible in order to achieve a better product experience. [0003] In related technologies, the method used to achieve sorting is to use machine learning to learn the relationship between a series of information such as query intent, user characteristics, retrieval information characteristics, and target (target content), depending on the target Depending on the setting and the optimization of the Loss Function (loss function), the sortin...

AI Technical Summary

Problems solved by technology

[0005] In terms of model, the generality of a single model is not strong. The tree models used for sorting include XGBoost and LGBM models. The two models can mine and combine high-order statistical features. Unfriendly, poor performance, more dependent on continuous statistical features, this is largely because the model has learned the numerical information of the id feature, but the numerical information of the id feature is meaningless and there is a lot of interference sex

If the id feature is encoded in one-hot form, it will cause high-dimensional sparse features, which cannot be handled well for tree models.

[0006] For the above-mentioned problems existing in related technologies, no effective solution has been found yet

Images