END to End Deeply Personalized Searches Now Available
I remember when Matt Cutts used to discuss how Google searches are becoming more personalized. If you were typing in Pizza in 1998 Google you would find information about Pizza but in 2013 Google you would receive localized results based on your location. These results would show the restaurants and ratings along with their location on a map.
What Matt mentioned that Google doesnt have the capability to do at the time however was that if you typed in “Tomorrow” it wouldn’t know if you wanted to learn about the day after the current date, the song by Silverchair, or the song by Chris Young made in 2011.
Well now the personalization is getting to a whole new level thanks to a team at UCLA with their EDAM. It will see what searches you’ve done previously but also integrate viewings on Youtube and other clues to personalize the search so if you listen to Nirvana and Alice in Chains you probably are looking for Silverchairs song Tomorrow vs. Chris Young.
Special thanks to this team who will be the Amit Singhals of the engineering and search team of the future:
From the piece: Personalized item retrieval for online content-sharing platforms without any descriptive information based on the query-aware attention mechanism with external key memory and locality preservation. Experimental results and analysis on the large-scale dataset from areal-world commercial online content-sharing platform also demonstrate the effectiveness and the robustness of EDAM. The insights can be concluded as follows: (1) user history is helpful for personalized item retrieval; (2) learning external key item embeddings for estimating attention weights is beneficial, especially for the users with shorter item history; (3) sequential information in user history is sensitive for item retrieval so that EDAM with locality preservation outperforms baselines of sequence models such as ARNN.
White Paper: https://dl.acm.org/doi/pdf/10.1145/3366423.3380051