CS224W fall 1.2 Applications of Graph ML

For giving a heterogeneous diagram that describes the relationship between users' interests in goods ： One product for one   Whether people are attractive If you just embed by setting features of different dimensions The effect will be better than   At the same time, the goods 、 People, goods and human connections are embedded together as information Bad come

because ： At the same time, the goods 、 The connection between people, goods and people is used as information , At this time, the input is not only the product information 、 Human information ; What is more important is the relationship between someone and many commodities he was interested in in in the past , This structural information stay When judging whether a person is interested in a new product It is useful to . It is much better than considering only the characteristics of the commodity itself （ As shown in the screenshot above （9：14）.

therefore Images （ Commodity information 、 Human information ） add The graph structure Cause more than the picture itself Better advice

PS： The Chinese translation in the video will translate “Graph” Translate into “ Chart ”, In fact, it should be “ chart ”