Multi-task prediction model training, event type prediction method and device

Abstract

The embodiment of this specification provides a method and device for training a multi-task prediction model and predicting an event type. In the training method, an initial sample set is obtained. For any first sample, in the first i−1 decision trees, obtain N scores of each leaf node including the first sample. Based on the respective N scores of each leaf node and the N label values ​​of the first sample, several gradients are determined, and the first fusion is performed to obtain the fusion gradient of the first sample. For the current node of the i-th decision tree, obtain the fusion gradient of each sample in the sample set split into the current node. Based on the fusion gradient of each sample in the current sample set, the split feature and feature threshold of the current node are respectively determined from the features of each sample and the feature values ​​of each sample in the current sample set corresponding to each sample feature. Based on the split feature and feature threshold of the current node, the current sample set is segmented, and child nodes corresponding to the current node are generated until the leaf node is reached.

Description

technical field [0001] One or more embodiments of this specification relate to the field of computer technology, and in particular to a method and device for training a multi-task prediction model and predicting event types. Background technique [0002] In traditional techniques, Gradient boosted decision trees (GBDT) can usually only be used to train single-task prediction models. The single-task prediction model here means that it can only be predicted for a single task. For example, a single-task prediction model can only predict whether the current text is an advertisement, or whether the current event is a fraud event, etc. [0003] However, in real-world scenarios, multi-task prediction is usually required. For example, in order to achieve effective risk management and control of events, it is usually necessary to predict the probability corresponding to each type of fraud for an event. Based on this, it is necessary to provide a training method for a multi-task pr...

AI Technical Summary

Problems solved by technology

For example, a single-task prediction model can only predict whether the current text is an advertisement, or whether the current event is a fraud event, etc.

Images