Method and system for sorting search

Abstract

The application provides a method and system for sorting search. The method comprises following steps: obtaining a search word and obtaining texts matched with the search word by means of static sorting index, wherein the obtained texts are used as mass-selection recall texts; obtaining rough sorting feature weight by fitting the pre-counted fine sorting score of the mass-selection recall texts by means of the machine learning method, and calculating the rough sorting score of the mass-selection recall texts by means of the rough sorting feature weight; selecting first M texts which are sorted at front according to the rough sorting score, wherein the selected texts are used as rough sorting texts; calculating the user behavior probability of the rough sorting texts corresponding to the search word and calculating the fine sorting score of the rough sorting texts by means of the machine learning method using the user behavior probability; sorting the rough sorting texts according to the fine sorting score. The application can increase search quality and improve the user search experience.

Description

technical field [0001] The embodiments of the present application relate to the technical field of search engines, in particular to a method and system for searching and sorting. Background technique [0002] The search engine plays a major role in Internet technologies, and is used to search and return documents (Documents, Doc for short) that meet user needs from massive amounts of information according to user query words. Search engines usually create a full index based on static sorting in descending order (static sorting index) at regular intervals. When users search for query words, the search engine will search for Doc through audition recall, rough sorting, and fine sorting. Generally speaking, a search engine calculates a static ranking score for each Doc, which mainly calculates the static ranking score based on the text description quality and historical performance of the Doc, and then creates an index according to the reverse order relationship based on the sta...

AI Technical Summary

Problems solved by technology

[0003] In traditional search engines, the four links of static sorting index, audition recall, rough sorting, and fine sorting are relatively independent, and each link has its own independent goals. The goals are not unified, so that the Doc returned in the current link is of great importance to the overall situation. language is not optimal

Throughout the entire search process, sea selection recall and rough sorting are mainly used to select high-quality tens of thousands of Docs from massive indexes, and fine sorting is used to sort the screened Docs according to user feedback. Statistics show that based on sea The Docs screened out by the recall and rough sorting coincide with the Docs selected from the index based on the fine sorting by about 50%, that is, nearly half of the high-quality Docs are lost in the sea selection recall and rough sorting, which seriously affects the user experience

Images