Implementation of Short Video Click-Through Rate Estimation Model Based on Cross-Media Collaborative Filtering Neural Network
Table 1
Audio feature extraction results.
Function number
Function name
Description
1
Chromaticity deviation
Standard deviation of 12 chromaticity coefficients
2
Chromaticity vector
The 12 elements of spectral energy represent the 12 isothermal pitch classes (semitone spacing) of Western music
3
Mel’s inverse spectral coefficient
Mel frequency cepstrum coefficients forming the cepstrum representation, where the frequency bands are not linear but have to be distributed according to the Mel scale
4
Spectral roll-off point
Below this frequency, 90% of the spectrum’s amplitude distribution is concentrated
5
Spectral flux
The squared difference between the normalized amplitudes of the spectra of two consecutive frames
6
Spectral entropy
The entropy of the normalized spectral energy of a set of subframes
7
Spectral extension
The second central moment of the spectrum
8
Spectral center of mass
The center of gravity of the spectrum
9
Energy entropy
The entropy of the normalized energy of a subframe, which can be interpreted as a measure of the mutation
10
Energy
The sum of squares of the signal values normalized by the corresponding frame length
11
Trans-zero rate
The rate of sign change of the signal during a given frame duration