Thomas Wood NLP/data science consultant Past projects Boehringer - - PowerPoint PPT Presentation

Thomas Wood NLP/data science consultant

Past projects

• Boehringer Ingelheim - pharma
• CV Library:

○ Predict industries/salaries from CV - word2vec + CNN ○ Predict search terms from CV - LSTM

• Forensic stylometry demo

○ Identifying author

• Chatbots

○ Intelligent home etc ○ Question answering about products

• Document clustering, classification, trend detection, sentiment analysis
• Cambridge Masters: anaphora resolution it’s raining

Boehringer Ingelheim

• Before running a clinical trial a

pharma company writes a 200 page PDF called a protocol.

• I developed an ML model which

extracts important data from the protocol: type of treatment, toxicity, number of subjects, etc.

Boehringer Ingelheim (2)

• Company has factories all over the world. Most medicines go through

multiple facilities and countries before going to market.

• When manufacturing defect occurs it is written in free text in local

language by factory worker, e.g. temperature deviation of 5 degrees due to crack in vial probably occurring in transit

• I ran unsupervised topic detection to identify commonest problems in

various categories of products from the unstructured text data.

CV-Library

• Upload CV
• Goes through word2vec
• Recommends industry
• Use TensorFlow NMT to

recommend search term

○ Repurposed Viet translator

• Trained on 12 million CVs
• Deployed on GCP
• 7% increase in signups - £££

When you upload a CV, it gets converted to TXT and passed through deep NN ... Then some fields which candidate previously filled out, get autofilled! Result: more engagement, fewer dropouts This was 2.5 years ago, before ELMO/BERT

Chatbots Artificial Solutions

• Worked building chatbots for mobile and

web

• Shell, AT&T, IKEA, Samsung, HTC, Rightmove
• Integrated smart home with voice

commands

○ turn on the coffee machine every Tuesday when I

• pen the downstairs front door

Forensic stylometry

• https://www.fastdatascience.com/author-prediction-demo
• Oxford University workshop on NLP every summer

Document analysis, trend detection

• Developed NLP pipeline for English and

German at Pattern Science AG, near Frankfurt

• Used for document classification
• Trend detection
• Emerging topics

Masters Cambridge

• Unsupervised learning for identifying

pleonastic pronouns

○ It seemed that things would never get any better ○ It surprised me to hear him say that

• Download available