Learning to Rank: From Theory to Production - Malvina Josephidou & Diego Ceccarelli, Bloomberg

[@ashique6246663]

LTR Steps: 8:29 Collecting training data: 10:16 Model evaluation: 18:41 Feature latency: 22:03 Index static features: 24:05 Changing to docvalues: 25:06 Grouping: 29:00 Rolling out: 33:00 Takeaways: 35:00

[@deepakkumarjoshi]

Whats the difference between Ranking and Sorting? If we are using a formulae e.g. 2 * featureA + 4.5 * featureB to generate the score, then we can use this formulae with traditional programming also and sort the results, why do we need ML here?

[@adrianhishow8134]

@@deepakkumarjoshi The model figures out the exact weights for each feature. Based on those features the most relevant documents according to BM25 are re-ranked and then returned to the user

[@deepakkumarjoshi]

Whats the difference between Ranking and Sorting? If we are using a formulae e.g. 2 * featureA + 4.5 * featureB to generate the score, then we can use this formulae with traditional programming also and sort the results, why do we need ML here?

[@shakes115]

To figure out what 2 and 4.5 should br