Multi-label Classification algorithm technical details Notes _ Natural Language processing

One: Multi-label classification algorithm threshold function (Threshold calibration)

The threshold function here is: For each label's predictive value, when it reaches how much is determined to exist for this label, less than how much is judged as the word label does not exist.
There are mainly two kinds of threshold functions: constant threshold function and threshold function based on training set. 1. Constant threshold function

The common choice is 0 or 0.5. For example, 0.5 is judged to exist when a given label predicts a value greater than it, whereas it does not exist. 2. Threshold function based on training set two: Multi-label Classification evaluation Index

Third: Optimization of tanh activation function

Its original formula is: (MATH.EXP (x)-Math.exp (-X))/(Math.exp (x) + math.exp (-X))
To improve operational efficiency: conversion to 2.0/(1.0 + math.exp ( -2.0 * inputvalue))-1.0
They make an equivalence transformation.

。。。 Not completed

Reference documents

[1] Zhang M L, Zhou Z H. A Review on Multi-label Learning algorithms[j]. IEEE Transactions on Knowledge & Data Engineering, 2014, 26 (8): 1819-1837.