Machine learning systems rely on data for training, and training data can be broadly divided into two categories: "features" and "labels."
"Features" are data that are input into the machine learning model, such as from sensors, customer questionnaires, website cookies, or historical information.
However, can vary. For example, customers may fill list of armenia whatsapp phone numbers in the questionnaire haphazardly or ignore the questions; the sensor may malfunction and return incorrect data; even if the user's web page behavior is clear, the information reported by the website cookie may not be the same. whole.
In addition, the data may also contain noise. When unnecessary information is mixed in, the machine learning model will be misled and make incorrect predictions.
Compared with "features", the accuracy and stability of "labels" are more important. Labels are the final output of the machine learning model. So the machine learning model needs to be taught with the correct results at training time. Label sparsity is also a problem, which is a phenomenon that occurs when the system has mastered a large amount of input data but is not sure about the output results. In such a case, it will be difficult for the model to detect and optimize the correlation between its features and labels, and it may even require additional human intervention to associate labels with input data.
Machine learning relies on the correlation of input and output data in order to have sufficient generalization capabilities to predict future actions and provide relevant recommendations. Therefore, if the input data is too messy, incomplete, or biased, it may be difficult to understand why a certain output/label is generated. In recent years, machine learning has also developed many advanced methods such as semi-guided learning and transfer learning to deal with such problems.
