MachineLearning
Upton, Lance T.
A pattern recognition algorithm to classify questions asked at a library service desk, based on a term document matrix (TDM).
Raw Data
Predictors
## location format date
## info desk : 79 walk-up:48738 Min. :2015-05-01
## media : 56 phone : 6989 1st Qu.:2016-01-16
## tech desk : 5301 Median :2016-08-22
## staff desk: 5 Mean :2016-07-12
## west desk :24596 3rd Qu.:2017-01-12
## east desk :25654 Max. :2017-08-01
## off-desk : 36
## datetime question answer
## Min. :2015-05-01 11:18:00 Length:55727 Length:55727
## 1st Qu.:2016-01-16 15:50:00 Class :character Class :character
## Median :2016-08-22 10:51:00 Mode :character Mode :character
## Mean :2016-07-12 19:31:22
## 3rd Qu.:2017-01-12 14:48:30
## Max. :2017-07-31 20:56:00
##
Categories
## tag1 tag2 tag3
## Reference :27830 internal_directions:21505 other/na :11692
## Directional :25362 policy :10895 classroom/space: 7704
## Miscellaneous: 2535 technology : 6849 equipment : 7408
## known_item_lookup : 3500 desk : 3255
## supplies : 3265 hours : 2987
## external_directions: 3017 account : 1994
## (Other) : 6696 (Other) :20687
NOTE: Categories are hierarchical (tag1 > tag2 > tag3).
Method
Step 1: Base Model
Questionis cleaned.- A TDM is generated and features are extracted.
- Data is split into training and testing sets.
- Models are created to predict
tag1. - Models are run and tested.
- A model with satisfactory accuracy is used to split data by
tag1. tag2is predicted (and so on...).
Step 2: Add predictors
- Different combinations of other predictors are added to the model.
- ???
- Profit.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent