How to caculate distance with the vectors?

[mmpang]

I got the user vector and item vector with nemf(collaborative similarity embedding)，which distance should I use to do ann search，inner product (dot product) or cosine distance or Euclidean distance ?

[chihming]

both dot product and cosine similarity are okay

[mmpang]

@chihming Thanks for your reply. Another question, in a recommander system, can i do u2i or i2i retrieval with the vectors directly ?

[chihming]

yes, such vector-based retrieval is able to achieve a reasonable performance