Imagine a computer as a cook who cannot read recipes written in words and understands only numbers. Vectorisation is the process of “translating” the whole recipe into his language. Every ingredient and every step in the recipe is replaced by a specific set of numbers. The word “salt”, for example, gets the coordinates [1, 0, 0] and “pepper” [0, 1, 0]. This lets the computer not only make sense of the individual ingredients but also compare how similar two recipes are, purely on the basis of these numbers. Vectorisation thus converts information from our world into the numerical form an artificial intelligence can handle.
Vectorisation, in the context of machine learning and data processing, is the process of converting unstructured or categorical data into the format of numerical vectors. The aim is to transform data such as text, images or sounds into a form that mathematical models can understand and process. This conversion allows algorithms to quantify and analyse relationships, patterns and semantic similarities between individual data points. Techniques such as Bag-of-Words (BoW), TF-IDF, or more advanced methods like Word2Vec and GloVe are examples of vectorisation used in natural language processing. A vectorised representation is the basic prerequisite for training most artificial intelligence models.