A Personalized Movie Recommendation Method Based on Neural Network and Collaborative Filtering

Abstract

The invention discloses a personalized movie recommendation method based on neural network and collaborative filtering, which adopts Bert neural network to extract features of movie plots, forms a feature matrix about the item to form a connection with Funk-SVD, and then uses matrix decomposition technology to generate A complete U‑I matrix, a fast and efficient way to get all predicted scores. First use the Bert neural network to extract features of the movie plot, and obtain a feature matrix about the movie item; then connect the obtained feature matrix with the collaborative filtering algorithm Funk-SVD algorithm, and then use matrix decomposition technology and gradient descent method to optimize, Obtain a complete U-I matrix with the smallest error, and finally obtain a series of operations such as all prediction scores; the present invention adds auxiliary information, namely the plot of the movie, on the basis of the original explicit feedback and implicit feedback, and obtains the item more accurately. The feature matrix, which reduces the minimum error by 2.40%, improves the accuracy of prediction.

Description

technical field [0001] The present invention belongs to the field of personalized recommendation based on artificial intelligence. Specifically, the Bert neural network is used to extract the features of the movie plot, and a feature matrix about the item is formed to connect with Funk-SVD, and then a complete U-I matrix is ​​generated by matrix decomposition technology. A fast and efficient way to get all predicted ratings. Background technique [0002] Currently, there are three main ways to implement recommendation systems: content-based recommendation (CB), collaborative filtering recommendation (CF) and hybrid recommendation. [0003] CB: Compare items with the user's previous favorite items, and then recommend the best matching items. But the main problems of this method are the cold start problem and similar user reliability problem. [0004] · CF: Collaborative filtering is the most popular algorithm in the recommendation system. It analyzes the interaction data be...

AI Technical Summary

Problems solved by technology

These two processing methods will limit the results of pre-training to generate word vectors. The main reason is that its standard language model is one-way, which limits the structure used in training and leads to a decrease in the accuracy of feature extraction.

Images