News recommendation method and topic characterization method based on RNN and attention mechanism

Abstract

The invention relates to an RNN and attention mechanism-based news recommendation method and a topic representation method, which combine a traditional topic model with a neural network word vector and can effectively improve the accuracy of semantic extraction and representation of news content texts. The RNN is used for describing the seriality characteristics of news browsing of the user, so that the timeliness of the personalized news recommendation content can be greatly improved; influence weights of different news on recommendation prediction are distinguished by utilizing an attentionmechanism, user interest migration can be captured, and the accuracy and novelty of personalized news recommendation contents are improved; and finally, in combination with an attention mechanism of aDBSCAN density clustering algorithm; performing heuristic discovery on the new topics and the old topics through density clustering, and dynamically calculating influence weights of news by utilizinga topic clustering result, so that the novelty of the recommended topics is improved.

Description

technical field [0001] The invention belongs to the technical field of data mining and neural network, and in particular relates to a news recommendation method and topic representation method based on RNN and attention mechanism. Background technique [0002] With the development of information technology and the Internet, people have gradually entered the era of information overload from the era of information scarcity. As the main way for the public to obtain all kinds of information, various news websites are flooded with massive amounts of information every day. How to screen and filter news content, how to grasp the personalized needs of different users, how to follow up the migration of users’ reading interests, and how to provide users with novel, accurate and timely personalized recommendations have become the challenges faced by news websites. main challenge. Under the background of this demand, personalized news recommendation system came into being. [0003] A...

AI Technical Summary

Problems solved by technology

Disadvantages: It is impossible to distinguish the needs of different users' personalized reading interests. The recommended content needs to be filtered by users themselves, and it is impossible to describe the transfer of users' interests.

Disadvantages: There are still obvious differences between users. When the number of users in the system is small, it is difficult to find users who are particularly similar to this user

Disadvantages: Does not consider the differences in interests between users, it is easy to repeatedly recommend too many similar news, and cannot capture the migration of users' interests

[0015] Disadvantages: (1) Feature extraction of news content is generally difficult

(2) Unable to dig out the potential interests of users

(3) Unable to generate recommendations for new users

[0023] (1) The current content-based news recommendation methods are mostly based on content-based keyword or topic extraction. Due to the limited extraction performance of traditional topic models (it is easy to extract some common topics that do not have much meaning), the ability to express news content semantics is relatively low. Weak, greatly reducing the quality of recommended news content

[0024] (2) The current recommendation algorithm seldom considers the timing characteristics of users reading news, and ignores the impact of the order in which users read news on the timeliness and novelty of recommendations

[0025] (3) At present, most recommendation systems will recommend a large number of news topics that users often watch, but ignore the impact of new topics on recommendations. It is difficult to find the migration of user interests. As a result, the recommendation results are always around a few old topics, and users cannot be found. topics of potential interest

Traditional content recommendation algorithms seldom consider the sequential characteristics of their content, so some similar topics will be repeatedly recommended, "forgetting" the user's previous concerns, and unable to capture long-distance interests

Images