Text Classification

Search for an optimal model between the model algorithms available. Auto Mode selection leverages the Multinomial Naive Bayes and Linear SVC algorithms. For each model, it searches through a small range of corresponding parameters. Auto Mode then outputs the optimal combination of algorithm and hyperparameters. To fine-tune your model, choose one of the algorithm-specific dropdowns.

The multinomial Naïve Bayes algorithm is a probabilistic classification model. The Naïve Bayes classifier builds a model that predicts the probability that a piece of text belongs to a label. To build your model, use training data in the form of rows of text and their associated labels (also known as classes or target). The algorithm assumes that all features are independent of one another. The advantages of the Naïve Bayes classifier are that it is scalable and generally performs well with a small training set.

Alpha is an additive smoothing parameter that you can use to control model complexity. A value of 0 indicates no smoothing. A value greater than 0 might improve your results if a word in the test data doesn't exist in the training data.

The tool searches for the best model based on a range of Alpha values that you define. To create these Alpha values, enter the range you want to search (From–To) and the Number of Steps within that range.

Cross-Validation is a resampling technique that uses different portions (or folds) of your data for model training and validation. Choose how many folds to use during cross-validation.

The need to convert raw text into numerical data is a required step for text classification. This vectorization step allows the model to interpret your data. For the Text Classification tool, we use a Term Frequency-Inverse Document Frequency (TF-IDF) vectorization technique. These are the TF-IDF settings:

Linear SVC belongs to the Support Vector Machine class of models. You can apply this algorithm to data with 2 (binary) or more classes. Once fitted to your data, the model finds the best hyperplane that divides your data into the correct categories. Linear SVC is effective in high dimensional space such as text, however, it might be slow when applied to a large training dataset.

Choose the norm used in the penalization. Note that the L2 norm (also known as the Euclidean norm) is the standard used in Support Vector Classification. The L1 norm results in sparse coefficient vectors.

Choose a loss function. Hinge is the standard choice for this algorithm.

C is a regularization parameter. It must be greater than 0. Large values of C correspond to less regularization and a model that attempts a close fit to the training data. In contrast, small values of C correspond to increased regularization.

The tool searches for the best model based on a range of C values that you define. To create these C values, enter the log range you want to search (From–To) and the Number of Steps within that range.

Cross-Validation is a resampling technique that uses different portions (or folds) of your data for model training and validation. Choose how many folds to use during cross-validation.

The need to convert raw text into numerical data is a required step for text classification. This vectorization step allows the model to interpret your data. For the Text Classification tool, we use a Term Frequency-Inverse Document Frequency (TF-IDF) vectorization technique. These are the TF-IDF settings: