A user behavior prediction method and device

Abstract

Embodiments of the present invention provide a user behavior prediction method and device, which relate to the field of big data machine learning technology, and can reduce the deviation caused by high-order feature sparsity in the user behavior prediction process. The specific solution is: the prediction device obtains the weight corresponding to each feature in the first sample set; each sample in the first sample set includes a behavior label and a first feature vector, and the first feature vector includes multiple first-order features and multiple A high-order feature, the high-order feature is composed of multiple first-order features; the prediction device records the frequency of occurrence of each feature; determines the first frequency boundary; corrects the weight corresponding to each feature in the first sample set; where , the correction specifically includes: reducing the weights corresponding to the features whose frequencies are less than the first frequency boundary; predicting the probability correspondence between the first feature vector of the target and the value of the behavior label according to the weights corresponding to each feature after correction. The embodiment of the present invention is used to predict user behavior.

Description

technical field [0001] The embodiments of the present invention relate to the technical field of big data machine learning, and in particular to a user behavior prediction method and device. Background technique [0002] At present, user behavior prediction, as an important technology in the fields of personalized recommendation and accurate advertisement delivery, has been widely concerned and used. Taking click-through rate prediction in user behavior prediction as an example, in actual application scenarios, whether a user will click on a content is not only related to the user's inherent preference characteristics and attribute characteristics of the content, but also to the contextual characteristics that may affect the user's decision-making at that time. related. The existing technology incorporates contextual features into the feature vectors of training samples and generates high-order features through feature combination to reflect the joint effect of different fe...

AI Technical Summary

Problems solved by technology

[0003] However, some high-order features tend to appear too infrequently in the training sample set for the higher-order features generated by the simple feature combination method in the prior art, that is, the high-order features are sparse, so they are easy to appear in the training process. Local overfitting occurs in , so that the weights corresponding to some high-order features obtained through training do not have the generalization ability, resulting in deviations in user behavior prediction based on the weights corresponding to these high-order features

Images