Collaborative recall method based on user clicking and conversion duration feedback

Abstract

The invention discloses a collaborative recall method based on user clicking and conversion duration feedback. The collaborative recall method comprises the following steps of acquiring a historical behavior log of a user; storing the filtered data of the historical behavior log in a first database; calculating the historical average conversion duration of each click video in the historical behavior log and storing the historical average conversion duration in a second database; performing interval division on each click video in the historical behavior log, calculating a preference score of the user for each time interval, and storing the preference score in a third database; recalling the candidate video set pushed by the recommendation system, and calculating a sorting score of each candidate video; and recommending the first N candidate videos to the user according to the ranking score. According to the method, the problem that only the feedback click rate is considered in an existing video recommendation technology, and other duration factors are not considered, so that the overall duration of the system is shortened, is effectively solved, and a final recommendation result ismore accurate.

Description

technical field [0001] The invention relates to the technical field of video recommendation, in particular to a collaborative recall method based on user click and conversion time feedback. Background technique [0002] With the rapid development of the Internet, the information that people can access every day is increasing explosively. In order to push the correct information to the correct users, the recommendation system came into being. The current mainstream recommendation system mainly consists of two modules: recall and ranking. The recall module mainly uses various strategies and algorithms (for example, collaborative filtering) to generate candidate sets from the perspectives of the user's historical behavior and real-time behavior. Generally, the number of candidate sets is relatively large. Due to the requirements of the system response, the subsequent sorting module cannot timely The entire candidate set is processed. Therefore, after the candidate set is gene...

AI Technical Summary

Problems solved by technology

However, in actual scenarios, videos with high feedback click-through rates often have phenomena such as short video duration and title partying, so the front part of the rough sorting results are all short in duration; at the same time, after users click on these short videos, The click-through rate of its feedback has been improved, so that the final recommendation results are always concentrated on these short videos, which is very detrimental to the overall duration optimization of the system

Images