20
NLP text recommender system: Journey to auto model training at scale Aditya Sakhuja Engineering Lead, Salesforce Einstein @sakhuja

Journey to auto model training at scale NLP text

  • Upload
    others

  • View
    7

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Journey to auto model training at scale NLP text

NLP text recommender system: Journey to auto model training at scale

Aditya Sakhuja

Engineering Lead, Salesforce Einstein@sakhuja

Page 2: Journey to auto model training at scale NLP text

Agenda● Goal● Scenario● Approach & Metrics● ML System Architecture

○ Recs Serving○ Feature Engineering○ Model Training

● ML System Evolution● Training CI, Deployments & Rollbacks● Cloud Native● Challenges & Takeaways

Page 3: Journey to auto model training at scale NLP text

Customer Service Agent Assist

Goal

Page 4: Journey to auto model training at scale NLP text

Agents rely on traditional search results for finding relevant answers to often long and time sensitive customer questions.

Scenario

Page 5: Journey to auto model training at scale NLP text

Approach

Page 6: Journey to auto model training at scale NLP text

Business Metrics● Agent Time to Resolution● Agent Time spent per case● Case-Article Attach Rate

● # of recommendations served● MAO, MAU● Serving Latency

Page 7: Journey to auto model training at scale NLP text

ML System Architecture

Page 8: Journey to auto model training at scale NLP text

Recommendations ServingLayer 1 : Candidate Generation● NLP : Extract POS, NER, Noun and key terms from user query● IR specific Query Formulation ● Candidates Generated

Layer 2 : Ranking Model● <question, article> pairwise feature generation ● Candidates evaluated by model● Candidates above the threshold are recommended

Page 9: Journey to auto model training at scale NLP text

Recommendations Serving

Page 10: Journey to auto model training at scale NLP text

Data Prep & Feature Engineering

● Multi tenant data ingestion pipeline● Data Cleansing and Sanity checks● Precompute TDF, Corpus Statistics● Feature Vectors computation● 100+ of NLP features across different statistical feature categories● Serving Training Drift

Page 11: Journey to auto model training at scale NLP text

Model Training● Ranking Model● Auto tuned hyperparams● Auto Model comparison● Metrics

○ AUC○ F-Measure○ Precision, Recall○ Hit Rate @K

Page 12: Journey to auto model training at scale NLP text

Model Auto Training Pipeline

Page 13: Journey to auto model training at scale NLP text

ML System Evolutionversion 0● Heuristic based answer recommendations POC. First pilot sign up.● Communities use case: community selected bestAnswer, as positive label.● Generic model trained on open source dataset Stanford SQuAD

version 1● Ranking model : <question, answer> pairwise probability● Notebooks based on-demand training● Static configured data filtering

Page 14: Journey to auto model training at scale NLP text

ML System Evolutionversion 2● Dynamically configured training dataset attributes● Model retraining ● Multilingual Support● Multitenant Auto-trained models● Observability● Trained Model Deployments & Rollbacks

Page 15: Journey to auto model training at scale NLP text

Model Deployment, CI & Rollbacks

Page 16: Journey to auto model training at scale NLP text

Cloud Native Training

Page 17: Journey to auto model training at scale NLP text

Challenges● Data

○ Privacy and sharing compliances – GDPR, HIPAA, Accessibility○ Freshness / Hydration○ Handling encrypted data at rest and in motion○ Too sparse, not meeting thresholds ○ Too dense, training performance SLA not met

● Custom, non standard fields and datatypes● Building ML Infrastructure along the way● Training Serving Skew● Cold start problem

Page 18: Journey to auto model training at scale NLP text

Takeaways● Start small, Ship and Iterate● Prioritize ML infrastructure● Start with simple interpretable models● Scale model learning to the size of your data● Prioritize Observability● Prioritize Data privacy over model quality

Page 19: Journey to auto model training at scale NLP text

Thank you!

Page 20: Journey to auto model training at scale NLP text

Feedback

Your feedback is important to us.

Don’t forget to rateand review the sessions.