Annif: Research backends

[hortongn]

Learn more about backends in Annif and narrow in on the backend(s) we may want to use.

https://github.com/NatLibFi/Annif/wiki

[scherztc]

Backend: TF-IDF
Backend: fastText
Backend: Omikuji
Backend: MLLM
Backend: STWFSA
Backend: YAKE
Backend: SVC
Fusion/Ensemble backends that combine results from other backends
Backend: Ensemble
Backend: PAV
Backend: nn_ensemble
Special backends
Backend: HTTP
Backend: Dummy

[haitzlm]

**Back End Summaries:

• TF-IDF: The TF-IDF (Term Frequency-Inverse Document Frequency) backend uses a statistical method to weight words in a document based on their frequency and relevance to the document. It can be used for document classification and information retrieval tasks.
• fastText: The fastText backend is a deep learning model that is designed to handle text classification tasks. It uses a neural network architecture that can handle subword information, making it particularly effective for handling misspellings and out-of-vocabulary words.
• Omikuji: The Omikuji backend is a probabilistic method that uses the Naive Bayes algorithm to classify documents. It is based on the Bag-of-Words model and is a relatively simple and lightweight method that can work well for small to medium-sized datasets.
• MLLM: The MLLM (Maximum Likelihood Language Model) backend is a probabilistic method that models the probability of a document given its classification category. It is based on a language model that captures the statistical relationships between words in a document.
• STWFSA: The STWFSA (Suffix Tree Weighted Finite State Automaton) backend is a method that uses a combination of suffix trees and finite state automata to represent and process text. It can be used for text classification and information retrieval tasks.
• YAKE: The YAKE (Yet Another Keyword Extractor) backend is a text mining algorithm that identifies keywords and keyphrases in documents. It can be used to generate keywords that can then be used as input to other classification algorithms.
• SVC: The SVC (Support Vector Classifier) backend is a popular machine learning algorithm that can be used for text classification tasks. It works by finding a hyperplane that separates the different classes in the feature space.
• Fusion/Ensemble backends: These backends combine the results of other backends to generate a final classification result. For example, the Ensemble backend can combine the results of multiple backends using a simple voting scheme, while the nn_ensemble backend uses a neural network to learn the optimal combination of backends.
• PAV: The PAV (Pool Adjacent Violators) backend is a method for constructing monotonic classification models. It works by finding a partition of the feature space that maximizes the difference in class probabilities between adjacent regions.

Special backends:

• The HTTP backend allows Annif to interface with external services using a REST API.
• The Dummy backend provides a simple baseline for comparison with other backends, by always returning the same classification result regardless of the input document.