Introduction
Welcome to the Natural Language Processing service. This document describes how to use the NLP Auto Learning Platform and find help.
Billing for the NLP Auto Learning Platform started on September 23, 2019. Each new Alibaba Cloud account receives a one-month free trial. For detailed pricing, see NLP Auto Learning Platform pricing. To increase your queries per second (QPS) or if you have other questions, you can join the "Alibaba Cloud NLP Auto Learning Platform User Q&A Group 2" (DingTalk group ID: 44619071).
Service workflow
The basic workflow of the NLP Auto Learning Platform is shown in the following figure. The entire process is interactive and does not require an engineering or algorithm background. With more than 500 annotated data entries, you can achieve a model accuracy that exceeds 85%.
1. Activate the service
You must activate the Natural Language Processing service before you can purchase and use products on the NLP Auto Learning Platform.
First, make sure you have an Alibaba Cloud account. If you do not have an account, click here.
If you have an Alibaba Cloud account, click Activate NLP Auto Learning Platform Service to activate the service.
After you activate the NLP Auto Learning Platform, you must purchase a resource plan for service calls. You can purchase the NLP Auto Learning Platform Resource Plan for Calls.
To train models, you also need to purchase a monthly model subscription. You can purchase the NLP Auto Learning Platform Model Service.
2. Create a project
You can create and manage your projects. You can create 13 types of projects: text entity extraction, text classification, key phrase extraction, sentiment analysis, text relation extraction, short text matching, product review parsing, resume extraction, sentence pair classification, bidding information extraction, contract element extraction, contract extraction, and dialogue text classification.
You can add more project administrators to a project. Administrators have all permissions for the project, such as managing the project, data, and models. You can add administrators by entering their Alibaba Cloud account IDs. You can also use RAM users to manage projects. For more information, see the FAQ about RAM user authorization.
3. Manage data
After you open a project, you can manage your data in the Data Center. You can create data in two ways:
1. Create an annotation task. 2. Upload a dataset.
3.1 Create an annotation task
Step 1: Upload documents and add annotators
The creator and project administrators are annotators by default. You can also assign annotation tasks to your RAM users. Assigned users can log on to the platform with their RAM user credentials to start data annotation.
RAM user logon instructions:
1. Go to the RAM user logon page: https://signin.aliyun.com/login.htm
2. After logging on, click My Projects > Create Annotation Task.
Note: Only data files encoded in UTF-8 are supported.
Step 2: Set up items for annotation
The item types for annotation vary based on the project type. For example:
For a text entity extraction project, the items are entity names:
For a text classification project, the items are class names:
Step 3: Annotate data
After you create the annotation task, go to the Data Center and click Annotate to open the annotation center. You can then start to annotate the documents. Each document is annotated only once.
Step 4: Perform a quality check (Optional)
In the Data Center, find the project in the list and click the quality check button in the Operations column to open the quality check page.
You can filter and search for annotated documents to review them and ensure high annotation quality.
3.2 Upload a dataset
In addition to creating an annotation task, you can also upload your own pre-annotated training data. You must format the data according to the example file before you upload it.
In the Data Center, click the button to upload annotated data. This action opens the upload dataset window.
4. Manage models
In the Model Hub, you can train a model with one click, view model evaluation metrics, and perform interactive online tests. After testing, you can call the API.
4.1 Train a model
In the Model Hub, click Create Model to open the model creation page. Enter the basic model information, select the annotated data, and train the model with one click. Model training takes 30 to 60 minutes.
4.2 View a model
You can view the model's evaluation metrics, which include precision, recall rate, and F1 score. You can also add new model versions for version control.
Note: If you have fewer than 100 data entries for training, the model performance may be poor. The evaluation metrics may fluctuate significantly and are not a reliable reference. To achieve good and stable model performance, use more than 500 data entries for training.
4.3 Test a model
After a model is published, you can test it directly on the platform and correct any inaccurate prediction results.
For your convenience, you can publish two versions of the same model at the same time. To publish a third version, you must manually unpublish one of the existing published versions.