multilabel classification
In plain wordsTagging each item with any number of labels at once — a photo can show a beach, a dog and a sunset — instead of picking just one category.
A classificationclassification, n.: Sorting things into categories — spam or not spam, which animal is in a photo, which language a text is written in — by learning from examples that are already labeled. task in which an input may carry none, one or several labels from a fixed list of L, so a model answers L yes-or-no questions instead of choosing one class, and each target is a 0/1 vector of length L. Neural networks usually give every label its own sigmoidsigmoid function, n.: An S-shaped curve that squeezes any number into a value between 0 and 1, so the result can be read as a probability or as how far a switch is turned on. output trained with binary cross-entropycross-entropy, n.: A score of how surprised a model's predictions are by what actually happens; lower is better. Language models are trained mainly by driving this number down on real text., the network form of fitting a separate classifier per label (binary relevance). Methods that exploit correlations between labels either pass the predictions for earlier labels to the classifiers for later ones (classifier chains) or recast each label combination seen in training as a single class of a multiclassmulticlass classification, n.: Sorting each item into exactly one of three or more categories — for example, deciding which of the ten digits a handwritten number shows. problem (label powerset). Metrics score individual label slots, as Hamming loss (the share of slots wrong) and per-label or averaged precisionprecision, n.: Of all the items a model flagged as positive, the share that really were positive; a spam filter with high precision rarely sends genuine emails to the spam folder., recallrecall, n.: Of all the items that really were positive, the share the model managed to find; a cancer screen with high recall misses very few people who actually have cancer. and F1 scoreF1 score, n.: A single score that combines how many of a model's alarms are right and how many real cases it catches, and stays low unless both are good. do, or the whole set at once, as exact-match (subset) accuracy does. Uses include document tagging, protein-function prediction and moderation across several harm categories.
An article can be about politics, economics and sport at once, so we trained a multilabel model with one sigmoid per topic.
Origin
Studied under this name in late-1990s text categorization, where documents often have several topics: Andrew McCallum's 'Multi-label text classification with a mixture model trained by EM' (1999) and Robert Schapire and Yoram Singer's boosting work, including the BoosTexter system (2000), were influential early examples. Grigorios Tsoumakas and Ioannis Katakis's 2007 survey consolidated the vocabulary, and Jesse Read and colleagues introduced classifier chains (2009; journal version 2011).
Usage note
Distinct from multiclass classificationmulticlass classification, n.: Sorting each item into exactly one of three or more categories — for example, deciding which of the ten digits a handwritten number shows. (exactly one label) and from multi-output regression (several continuous targets).
Related
- multiclass classificationmulticlass classification, n.: Sorting each item into exactly one of three or more categories — for example, deciding which of the ten digits a handwritten number shows.
- binary classificationbinary classification, n.: Sorting things into exactly two groups, such as spam versus not spam or sick versus healthy. Most yes-or-no decisions made by AI systems are of this kind.
- classificationclassification, n.: Sorting things into categories — spam or not spam, which animal is in a photo, which language a text is written in — by learning from examples that are already labeled.
- one-vs-restone-vs-rest, n.: A way to handle many categories with a method that only gives yes-or-no answers: train one detector per category ('this one or not?') and pick the category whose detector is most confident.
- sigmoid functionsigmoid function, n.: An S-shaped curve that squeezes any number into a value between 0 and 1, so the result can be read as a probability or as how far a switch is turned on.
- F1 scoreF1 score, n.: A single score that combines how many of a model's alarms are right and how many real cases it catches, and stays low unless both are good.
- multiclass classificationmulticlass classification, n.: Sorting each item into exactly one of three or more categories — for example, deciding which of the ten digits a handwritten number shows.
Sources
- Schapire, R. E. & Singer, Y. (2000). BoosTexter: A Boosting-based System for Text Categorization. Machine Learning, 39, 135–168.
- Tsoumakas, G. & Katakis, I. (2007). Multi-Label Classification: An Overview. International Journal of Data Warehousing and Mining, 3(3), 1–13.
Mentioned in
classificationclassification, n.: Sorting things into categories — spam or not spam, which animal is in a photo, which language a text is written in — by learning from examples that are already labeled., image classificationimage classification, n.: Teaching a computer to say what a picture shows by choosing a label, such as 'cat', 'pneumonia' or 'stop sign', from a fixed list of possibilities., binary classificationbinary classification, n.: Sorting things into exactly two groups, such as spam versus not spam or sick versus healthy. Most yes-or-no decisions made by AI systems are of this kind., multiclass classificationmulticlass classification, n.: Sorting each item into exactly one of three or more categories — for example, deciding which of the ten digits a handwritten number shows., one-vs-restone-vs-rest, n.: A way to handle many categories with a method that only gives yes-or-no answers: train one detector per category ('this one or not?') and pick the category whose detector is most confident., sigmoid functionsigmoid function, n.: An S-shaped curve that squeezes any number into a value between 0 and 1, so the result can be read as a probability or as how far a switch is turned on., text classificationtext classification, n.: Sorting pieces of text into categories automatically — what topic an article covers, whether a message is spam, which team should handle a customer's complaint., sound event detectionsound event detection, n.: Teaching computers to recognize everyday sounds in a recording, such as a dog barking, glass breaking or a siren, and to say exactly when each one starts and stops.