File size: 711 Bytes
b109984
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
Commercial

# Source:

D. Y. Fu, M. F. Chen, F. Sala, S. M. Hooper, K. Fatahalian, and C. Ré. Fast and three-rious: Speeding up weak supervision with triplet methods. In ICML, pages 3280–3291, 2020.


# Labels:
0: negative (the graph is not commercials)

1: positive (the graph is  commercials)



4 Labeling functions

LFs: In this dataset, there is a strong signal for the presence or absence of commercials in pixel histograms and the text; in particular, commercials are book-ended on either side by sequences of black frames, and commercial segments tend to have mixed-case or missing transcripts (whereas news segments are in all caps). We use these signals to build the weak supervision sources.