News recommendation algorithm based on global graph enhancement

Author | Wang Hao

Reviewer| Chonglou

## News App It is an important way for people to obtain information sources in their daily lives. Around 2010, popular foreign news apps included Zite and Flipboard, while popular domestic news apps were mainly the four major portals. With the popularity of new era news recommendation products represented by Toutiao, news apps have entered a new era. As for technology companies, no matter which one they are, as long as they master the sophisticated news recommendation algorithm technology, they will basically have the initiative and voice at the technical level.

Today, we take a look at a paper nominated for the Best Long Paper Award in RecSys 2023 - Going Beyond Local: Global Graph-Enhanced Personalized News Recommendations (Paper download address: https://www.php.cn/link/195d221c982e47eb58347e5d06ce3180

The overall architecture diagram of the algorithm is shown below:

We first define the news text content as follows (we only Using the word vector

##This formula is the feature representation on the user side. We define

below, which is the local entity feature. In fact, Put all the news titles into an array, and then calculate it using the above formula.

What we introduced above is local feature expression and user-side feature expression. We will use GNN below To express the global news side feature vector:

The final expression of the global news side feature vector is actually to combine these feature vectors Put together:

The final training loss function of the entire news recommendation system is as follows:

Next, let’s take a look at the experimental comparison results:

After comparison (above table) , we found that our newly designed algorithm (GLORY) is better than similar algorithms in many indicators, so it is a rare and excellent news recommendation algorithm. The whole algorithm design idea is very simple, but it uses heavyweight deep learning Technology. The author must have done a lot of technical work in the process of designing the algorithm, making the final effect of the algorithm outstanding.

The following is the use of different Graph Encoder to give news Experimental comparison effect of text-like encoding. It can be seen that using GNN has the best effect:

GLORY is a very excellent one that has appeared in recent years. News recommendation algorithm. Although this algorithm does not escape the old framework of content-based similarity calculation, the author makes full use of new technology and puts it in old matryoshka dolls to generate new value. This article puts old wine in new bottles. Thesis is very worthy of our serious study.

Introduction to the author

Wang Hao, former head of Funplus Artificial Intelligence Laboratory. He has held technology and technology executive positions in ThoughtWorks, Douban, Baidu, Sina and other companies. Working in Internet companies, financial technology, games and other companies for 13 years He has profound insights and rich experience in fields such as digital museum. Published 39 papers in international academic conferences and journals, and won the IEEE SMI 2008 Best Paper Award, ICBDT 2020 / IEEE ICISCAE 2021 / AIBT 2023 / ICSIM 2024 Best Paper Report Award.

The above is the detailed content of News recommendation algorithm based on global graph enhancement. For more information, please follow other related articles on the PHP Chinese website!