Behavior prediction method and device based on zoom dot product attention, equipment and medium

Abstract

The invention relates to the field of artificial intelligence, and provides a behavior prediction method and device based on zoom dot product attention, equipment and a medium. Firstly, a long time sequence is split based on a time interval, a user layer and a behavior sequence layer are separated by a time interval layer and are not directly connected, the sequence length in the interval is greatly reduced, and parallel computing is facilitated; the cross calculation amount is reduced, the operation efficiency is improved, encoding is carried out based on a scaling dot product attention mechanism, data features are optimized, to-be-predicted data is subjected to standardization processing to conform to the input data format of the model, the model is conveniently utilized for prediction, and then automatic prediction of behaviors such as purchasing is achieved based on an artificial intelligence means; and the accuracy and the prediction efficiency are higher. The invention also relates to a blockchain technology, and the behavior prediction model can be stored in the blockchain.

Description

technical field [0001] The present invention relates to the technical field of artificial intelligence, in particular to a behavior prediction method, device, equipment and medium based on scaling dot product attention. Background technique [0002] In order to predict the behavior of users such as purchases, it is usually necessary to obtain user behavior, and analyze and predict based on user behavior. [0003] However, the behavior sequence models currently used are mainly attention (based on the attention mechanism), RNN (Recurrent Neural Network, recurrent neural network) and other models. These models are originally in the field of natural language processing, and cannot be fully used in behavior sequence modeling scenarios fit, the main problems are as follows: [0004] (1) The behavior sequence is long, which makes the training and engineering calculation of the model more difficult. The RNN model has insufficient ability to process long sequences and will lose a l...

AI Technical Summary

Problems solved by technology

[0004] (1) The behavior sequence is long, which makes the training and engineering calculation of the model more difficult

The RNN model has insufficient ability to process long sequences and will lose a lot of information; the attention model will cross-calculate the attention weight of the sequence when processing long sequences, and the time complexity and space complexity of the cross-calculation are O(n 2 ), the complexity is too high, making it difficult to support production applications

[0005] (2) The time span is large, adjacent behaviors may be separated by a long time, and the coupling of discontinuous behaviors in business is serious

This coupling of discontinuous business behaviors will make the model unable to effectively learn important information

Images