Awesome

FAWOC the FAst WOrd Classifier

FAWOC is a TUI program for manually labelling a list of words. It has been developed to support the efficient clustering of documents based on topic modeling algorithms such as Dirichlet Latent Allocation.

The programs reads a CSV file containing the terms and allows the fast association of labels to the terms.

Each term is presented to the user, who can associate to the term one of the labels with the press of a key.

Some statistics are provided in the user interface to have a clue about the number of classified terms and the remaining ones.

The terms are sorted according to their frequency in the set of documents, which is a value that must be made available to FAWOC.

Example of usage

fawoc terms.csv

The input file terms.csv needs to have at least one column with the header (first column) called term.

Labels

The supported labels are the following:

• relevant: the term is relevant for a topic, but it may not significantly contributes to its characterization
• keyword: the term is a keyword and thus significantly contributes to the characterization of a topic
• noise: the term is a common word that does not contribute to the characterization of a topic
• not-relevant: the term originates from a document that should not belong to the set of relevant documents (typically due to a false positive in the query)
• stopword: the term should have been considered a stopword
• postponed: the term will be classified later (usually to continue the classification in presence of terms that may require more attention to be classified)
• autonoise: same as noise, but it was classified using the autonoise function (see the section below)

Commands and keybindings

The following keybindings are associated to the availabe labels:

• r relevant
• k keyword
• n noise
• x not-relevant
• s stopword
• p postponed
• a autonoise

The autonoise function automatically classifies as autonoise all the remaining terms having the same number of "grams" as the current term. E.g., if you are classifying the bi-grams related to a given word, the autonoise function will mark as autonoise all the remaining bi-grams, and will present the first tri-gram for classification. This is useful when it appears clear that the terms start to become rare or have some common pattern that allow to identify them as noise in a batch.

Other keys allow to save, undo or quit:

• u undo the last change
• w save immediately
• q quit

FAWOC automatically saves the changes on closing. Moreover, it autosaves the changes every 10 classified words.

Logging

FAWOC writes profiling information into the file profiler.log with the relevant operations that are carried out.

Files

FAWOC reads the terms from a tsv file with the following structure:

• id: identification number of the term. Must be unique. For backward compatibility this column may be missing. In this case, FAWOC assigns an id to each term that will be saved in a newly created column id on the first save;
• term: the term itself. For backward compatibility with old files, this column can be called keyword. This name is deprecated, and FAWOC will change it to term on the first save;
• label: a string describing the label assigned to the term.

FAWOC will load other information from two service files. These files are named after the input file, by removing its extension, adding the suffix fawoc_data and then adding the proper extension. The service files are:

• *_fawoc_data.tsv: it contains static information about each term. It is saved only on FAWOC closing. Currently, it is used to load the number of occurrences of each term;
• *_fawoc_data.json: it contains information used by FAWOC to correctly handle the undo command.

The --no-info-file command line option can be used to tell FAWOC to not load (and save) the *_fawoc_data.tsv. With this option, FAWOC will not display the count value.

*_fawoc_data.tsv

The format of this tsv file is:

• id: identification number of the term;
• term: the term itself. This field is not directly ised by FAWOC, and it is here only the make this file more readable;
• count: the number of occurrences of the term.

For backward compatibility with old files, if the *_fawoc_data.tsv file is missing, FAWOC searches for the count column in the input file. If this column is found, then FAWOC will use this value, otherwise the value -1 will be used. A new file *_fawoc_data.tsv is created on the first save with the loaded values of count.

*_fawoc_data.json

The format of this JSON file is a dictionary. The keys of this dictionary are the id of the terms. The values are dictionaries with the following format:

• order: number indicating the order in which each term is classified;
• related: related term selected at the moment the term is classified.

For backward compatibility with old files, if the *_fawoc_data.json is missing, FAWOC searches the order and the related fields in the input file. If they are not found, then FAWOC will not be able to handle the undo of the classifications made before. Each new classification will have its own entry in a newly created *_fawoc_data.json.