Deepopinion Workflow

    • Ahmed Al-Ali |
    • 0 step |
    • 69 minutes
    Log-in to DeepOpinion Studio [](
    Give your model a name, here we named it "Hotels Reviews" On model type choose "Aspect-based Sentiment Analysis" from the drop down menu for creating opinion mining model Click "CREATE"
    You will then find your new custom model added to your page. Click on it to start building your model

    Custom Model Building

    Tip! Creating a custom model happens in 4 simple steps: 1. Data Upload 2. Aspect Definition (listing the topics you want the model to detect) 3. Training Data (annotating examples that the AI model can learn from) 4. Model Training (the model here builds the required skills to solve your task based on example provided in the training data)
    Tip! If you have training data you can directly jump to step 3

    1. Upload Text Data

    You will now upload your text data. Your file format can be "xlsx", "csv", "json", or "jsonl" Click "Upload Documents" Click "BROWSE FILES" and select your file Select the column that has the text by ticking the checkbox Click "CREATE DOCUMENT GROUP"
    To build an intuition for the data you can search for keywords and check if they appear in the text. This can help you in deciding on the topics if you don't have a predefined list
    Use the back arrow to return to the document upload page to upload more files as needed

    2. Define Topics

    Now Navigate to Step 2 where you will list the topics that you want the model to detect in the text.
    Tip! You can either add topics manually or upload the whole list at once
    Click "Add an aspect"
    Type "Hotel [[enter]]"
    Click this icon.
    Type "Food [[enter]]"
    You can choose to upload your topics list as a "csv" file Click "UPLOAD ASPECTS" Click "BROWSE FILES" Click "UPLOAD ASPECTS"
    You can also review, edit, or delete topics at any time Avoid doing that after annotating data. It is tidier to create a new model if you would like to edit topics after training

    3. Label Data

    You are now ready to go to Step 3 "Training Data"
    Tip! You can either annotate examples yourself or directly upload it if you already have training data
    To start click on "Create a new labelling session"
    Tip! Aim to be neutral when labelling data by only selecting topics and sentiments that are explicitly mentioned in the text. This is the only way a model can replicate this task with high accuracy as it only can access the text with no other external knowledge
    Tip! Aim to annotate between 25-50 examples for each topic so that the AI model can learn the context of the topic and give high accuracy. You can track the total number of labeled examples for each topic on the list in Step 2
    Click the "Annotation" field. Click "CREATE LABELING SESSION" For every text segment select the topic and choose the expressed sentiment from the drop down menu
    You can also directly upload training data if you already have it. As a minimum, you require 3 columns: "Text", "Topic", and "Sentiment" Click "UPLOAD" Click "BROWSE FILES" Tick the checkbox on the "text" column Tick the checkbox on the "aspect" column Tick the checkbox on the "sentiment" column Tick the checkbox on the "segment_text" column (optional: for granular analysis) Tick the checkbox on the "span_start" column (optional: for granular analysis) Tick the checkbox on the "span_end" column (optional: for granular analysis) Click "CONFIRM AND UPLOAD TRAINING DATA" You will see that progress bar increase to 100%
    You can also go through the training data that you just uploaded to check the quality and edit the labels as needed
    Now you are ready to move to the final step in creating a model "Model Training"

    4. Train A Model

    Tip! You can choose to train on our pre-trained industry models to benefit from its exposure to your use case domain. Our "Base English" model is also a great start if none match your use case
    On this page you create your model Click "Train a model" Give it a name, I chose "Hotels Reviews v1", the v1 helps in tracking progress when labelling more data iteratively Select "Hotels (id: 57) [en]" from the model type drop down menu Click "START TRAINING" You will notice that the model training progress bar will increase to reach 100%
    Once the model training is completed the Status will change to "Ready" Now you actually can start using the model for text analysis
    You can also have a look at the model metrics and accuracy in details
    With a single click on this upward arrow you deploy the model to production and it is ready to use both through the Studio or as an API

    5. Analyzing Data

    You will also have a shortcut to this model under "My Models" on the models page
    Once you click on it you will be on the demo tab where you can take it for a test drive on single examples. This is helpful to have a quick test of model performance on unseen data
    Now you can navigate to the "Batch" tab from the sidebar to analyze data as a batch. You can upload files as "xlsx", "csv", "json", or "jsonl". Uploading your training data here can be a good start. You can also analyze new data file here for the same task.

    Want to make guides like this in seconds? Yes, it's really that fast.

    This Scribe is in tip-top shape!Leave feedback if there are any issues with this Scribe