SLIDE 1
S8768- Contextual Product Search With Vectorized Part Descriptions
Danny Godbout | Data Scientist
Application
Engineering Part Search
S8768- Contextual Product Search With Vectorized Part Descriptions - - PowerPoint PPT Presentation
S8768- Contextual Product Search With Vectorized Part Descriptions - - PowerPoint PPT Presentation
S8768- Contextual Product Search With Vectorized Part Descriptions Danny Godbout | Data Scientist Application Engineering Part Search Which Vehicles Were Built With Aerodynamic Fairings? N33-1068-199AC3 FAIRING, ROOF J83-1001-193AD3
SLIDE 2
SLIDE 3
Which Vehicles Were Built With Aerodynamic Fairings?
SLIDE 4
N33-1068-199AC3 FAIRING, ROOF J83-1001-193AD3 LIGHT , ROOF S78-1201-3117 SEAT , DRIVER 15-1068-199AC3 FAIRING, ROOF 15-6023-5J22K FRNG, ROOF BRACKET N33-1201-3117 FRNG, FUEL TANK AACS16034-033123 SEAT , FASTENER M12
SLIDE 5
N33-1068-199AC3 FAIRING, ROOF J83-1001-193AD3 LIGHT , ROOF S78-1201-3117 SEAT , DRIVER 15-1068-199AC3 FAIRING, ROOF 15-6023-5J22K FRNG, ROOF BRACKET N33-1201-3117 FRNG, FUEL TANK AACS16034-033123 SEAT , FASTENER M12 N33
SLIDE 6
N33-1068-199AC3 FAIRING, ROOF J83-1001-193AD3 LIGHT , ROOF S78-1201-3117 SEAT , DRIVER 15-1068-199AC3 FAIRING, ROOF 15-6023-5J22K FRNG, ROOF BRACKET N33-1201-3117 FRNG, FUEL TANK AACS16034-033123 SEAT , FASTENER M12 FAIRING
SLIDE 7
N33-1068-199AC3 FAIRING, ROOF J83-1001-193AD3 LIGHT , ROOF S78-1201-3117 SEAT , DRIVER 15-1068-199AC3 FAIRING, ROOF 15-6023-5J22K FRNG, ROOF BRACKET N33-1201-3117 FRNG, FUEL TANK AACS16034-033123 SEAT , FASTENER M12 FAIRING, FRNG, N33, …
SLIDE 8
N33-1068-199AC3 FAIRING, ROOF J83-1001-193AD3 LIGHT , ROOF S78-1201-3117 SEAT , DRIVER 15-1068-199AC3 FAIRING, ROOF 15-6023-5J22K FRNG, ROOF BRACKET N33-1201-3117 FRNG, FUEL TANK AACS16034-033123 SEAT , FASTENER M12 FAIRING, FRNG, N33, …
SLIDE 9
Approach
Interpret vehicle options like expert reading BoM
SLIDE 10
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 11
Base P/N Lowercase Remove Non-Alpha Jaro-Winkler Match Re-Sample
nthreethree FAIRNG, TRUCK ROOF 45” CUSTOMERASPECIAL nthreethree fairng, truck roof 45” customeraspecial nthreethree fairng, truck roof 45” customeraspecial nthreethree fairing truck roof customeraspecial nthreethree fairing truck roof customeraspecial
Pre-Processing
N33-1078-03123 | FAIRNG, TRUCK ROOF 45” CUSTOMERASPECIAL
nthreethree fairing truck
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 12
Vectorization- Word2Vec
Image Source: “Vector Representations of Words | TensorFlow.” TensorFlow, www.tensorflow.org/tutorials/word2vec.
𝐾
()
= log(𝑅 𝐸 = 1 | 𝑢ℎ𝑓, 𝑛𝑏𝑢 ) + log (𝑅 𝐸 = 0 𝑨𝑓𝑐𝑠𝑏, 𝑛𝑏𝑢))
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 13
BlazingText – GPU Acceleration
Source: Gupta S., Khare V. “BlazingText: Scaling and Accelerating Word2Vec using Multiple GPUs” MLHPC’17: Machine Learning in HPC Environments, November 12–17, 2017, Denver, CO, USA
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 14
Word Vectors
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 15
Part Vectors
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 16
Kernels
Aerodynamic Attachment
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 17
Description Matrix ● Search Kernels = Term- Kernel Similarity Max Pool(Kernel Similarity) = Description Vector Roof Fairing Bracket
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 18
Part Search
Search Term: FAIRING AND NOT BRACKET Part Vectors 𝑡𝑗𝑛𝑗𝑚𝑏𝑠𝑗𝑢𝑧 = cos 𝜄 = 𝐵 𝐶 𝐵 | 𝐶 |
Pre- Process Vectorize Define Kernels Cosine Distance
SLIDE 19
Results
Match quality, speed
SLIDE 20
Measures
Kappa Sensitivity (TPR) Fall-Out (FPR) Naïve Search 0.28 0.30 0.07 Raw Word2Vec 0.61 0.88 0.21 Kernel Search 1.00 1.00 0.00
SLIDE 21
SLIDE 22
Recap
Word Vectors applied to domain-specific vocabulary Improves search reliability in multi-generational part catalog GPU acceleration: More iterations in a given timeframe