1# Text classification 2 3Use a TensorFlow Lite model to category a paragraph into predefined groups. 4 5Note: (1) To integrate an existing model, try 6[TensorFlow Lite Task Library](https://www.tensorflow.org/lite/inference_with_metadata/task_library/nl_classifier). 7(2) To customize a model, try 8[TensorFlow Lite Model Maker](https://www.tensorflow.org/lite/models/modify/model_maker/text_classification). 9 10## Get started 11 12<img src="images/screenshot.gif" class="attempt-right" style="max-width: 300px"> 13 14If you are new to TensorFlow Lite and are working with Android, we recommend 15exploring the guide of 16[TensorFLow Lite Task Library](../../inference_with_metadata/task_library/nl_classifier) 17to integrate text classification models within just a few lines of code. You can 18also integrate the model using the 19[TensorFlow Lite Interpreter Java API](../../guide/inference#load_and_run_a_model_in_java). 20 21The Android example below demonstrates the implementation for both methods as 22[lib_task_api](https://github.com/tensorflow/examples/tree/master/lite/examples/text_classification/android/lib_task_api) 23and 24[lib_interpreter](https://github.com/tensorflow/examples/tree/master/lite/examples/text_classification/android/lib_interpreter), 25respectively. 26 27<a class="button button-primary" href="https://github.com/tensorflow/examples/tree/master/lite/examples/text_classification/android">Android 28example</a> 29 30If you are using a platform other than Android, or you are already familiar with 31the TensorFlow Lite APIs, you can download our starter text classification 32model. 33 34<a class="button button-primary" href="https://storage.googleapis.com/download.tensorflow.org/models/tflite/text_classification/text_classification_v2.tflite">Download 35starter model</a> 36 37## How it works 38 39Text classification categorizes a paragraph into predefined groups based on its 40content. 41 42This pretrained model predicts if a paragraph's sentiment is positive or 43negative. It was trained on 44[Large Movie Review Dataset v1.0](http://ai.stanford.edu/~amaas/data/sentiment/) 45from Mass et al, which consists of IMDB movie reviews labeled as either positive 46or negative. 47 48Here are the steps to classify a paragraph with the model: 49 501. Tokenize the paragraph and convert it to a list of word ids using a 51 predefined vocabulary. 521. Feed the list to the TensorFlow Lite model. 531. Get the probability of the paragraph being positive or negative from the 54 model outputs. 55 56### Note 57 58* Only English is supported. 59* This model was trained on movie reviews dataset so you may experience 60 reduced accuracy when classifying text of other domains. 61 62## Performance benchmarks 63 64Performance benchmark numbers are generated with the tool 65[described here](https://www.tensorflow.org/lite/performance/benchmarks). 66 67<table> 68 <thead> 69 <tr> 70 <th>Model Name</th> 71 <th>Model size </th> 72 <th>Device </th> 73 <th>CPU</th> 74 </tr> 75 </thead> 76 <tr> 77 <td rowspan = 3> 78 <a href="https://storage.googleapis.com/download.tensorflow.org/models/tflite/text_classification/text_classification_v2.tflite">Text Classification</a> 79 </td> 80 <td rowspan = 3> 81 0.6 Mb 82 </td> 83 <td>Pixel 3 (Android 10) </td> 84 <td>0.05ms*</td> 85 </tr> 86 <tr> 87 <td>Pixel 4 (Android 10) </td> 88 <td>0.05ms*</td> 89 </tr> 90 <tr> 91 <td>iPhone XS (iOS 12.4.1) </td> 92 <td>0.025ms** </td> 93 </tr> 94</table> 95 96\* 4 threads used. 97 98\*\* 2 threads used on iPhone for the best performance result. 99 100## Example output 101 102| Text | Negative (0) | Positive (1) | 103| ------------------------------------------ | ------------ | ------------ | 104| This is the best movie I’ve seen in recent | 25.3% | 74.7% | 105: years. Strongly recommend it! : : : 106| What a waste of my time. | 72.5% | 27.5% | 107 108## Use your training dataset 109 110Follow this 111[tutorial](https://www.tensorflow.org/lite/models/modify/model_maker/text_classification) 112to apply the same technique used here to train a text classification model using 113your own datasets. With the right dataset, you can create a model for use cases 114such as document categorization or toxic comments detection. 115 116## Read more about text classification 117 118* [Word embeddings and tutorial to train this model](https://www.tensorflow.org/tutorials/text/word_embeddings) 119
