Sentiment Analysis with OpenVINO™

This tutorial is also available as a Jupyter notebook that can be cloned directly from GitHub. See the installation guide for instructions to run this tutorial locally on Windows, Linux or macOS. To run without installing anything, click the “launch binder” button.

Binder Github

Sentiment analysis is the use of natural language processing, text analysis, computational linguistics, and biometrics to systematically identify, extract, quantify, and study affective states and subjective information. This notebook demonstrates how to convert and run a sequence classification model using OpenVINO.

Table of contents:


import warnings
from pathlib import Path
import time
from transformers import AutoModelForSequenceClassification, AutoTokenizer
import numpy as np
from import mo
from openvino.runtime import PartialShape, Type, serialize, Core

Initializing the Model

We will use the transformer-based DistilBERT base uncased finetuned SST-2 model from Hugging Face.

checkpoint = "distilbert-base-uncased-finetuned-sst-2-english"
model = AutoModelForSequenceClassification.from_pretrained(

Initializing the Tokenizer

Text Preprocessing cleans the text-based input data so it can be fed into the model. Tokenization splits paragraphs and sentences into smaller units that can be more easily assigned meaning. It involves cleaning the data and assigning tokens or IDs to the words, so they are represented in a vector space where similar words have similar vectors. This helps the model understand the context of a sentence. Here, we will use AutoTokenizer - a pre-trained tokenizer from Hugging Face:

tokenizer = AutoTokenizer.from_pretrained(

Convert Model to OpenVINO Intermediate Representation format.

Model conversion API facilitates the transition between training and deployment environments, performs static model analysis, and adjusts deep learning models for optimal execution on end-point target devices.

ir_xml_name = checkpoint + ".xml"
MODEL_DIR = "model/"
ir_xml_path = Path(MODEL_DIR) / ir_xml_name
ov_model = mo.convert_model(model, input=[mo.InputCutInfo(shape=PartialShape([1, -1]), type=Type.i64), mo.InputCutInfo(shape=PartialShape([1, -1]), type=Type.i64)])
serialize(ov_model, ir_xml_path)
/opt/home/k8sworker/ci-ai/cibuilds/ov-notebook/OVNotebookOps-475/.workspace/scm/ov-notebook/.venv/lib/python3.8/site-packages/transformers/models/distilbert/ TracerWarning: torch.tensor results are registered as constants in the trace. You can safely ignore this warning if you use this function to create tensors out of constant variables that would be the same every time you call this function. In any other case, this might cause the trace to be incorrect.
  mask, torch.tensor(torch.finfo(scores.dtype).min)

OpenVINO™ Runtime uses the Infer Request mechanism which enables running models on different devices in asynchronous or synchronous manners. The model graph is sent as an argument to the OpenVINO API and an inference request is created. The default inference mode is AUTO but it can be changed according to requirements and hardware available. You can explore the different inference modes and their usage in documentation.

core = Core()

Select inference device

Select device from dropdown list for running inference using OpenVINO:

import ipywidgets as widgets

device = widgets.Dropdown(
    options=core.available_devices + ["AUTO"],

Dropdown(description='Device:', index=1, options=('CPU', 'AUTO'), value='AUTO')
compiled_model = core.compile_model(ov_model, device.value)
infer_request = compiled_model.create_infer_request()
def softmax(x):
    Defining a softmax function to extract
    the prediction from the output of the IR format
    Parameters: Logits array
    Returns: Probabilities

    e_x = np.exp(x - np.max(x))
    return e_x / e_x.sum()


def infer(input_text):
    Creating a generic inference function
    to read the input and infer the result
    into 2 classes: Positive or Negative.
    Parameters: Text to be processed
    Returns: Label: Positive or Negative.

    input_text = tokenizer(
    inputs = dict(input_text)
    label = {0: "NEGATIVE", 1: "POSITIVE"}
    result = infer_request.infer(inputs=inputs)
    for i in result.values():
        probability = np.argmax(softmax(i))
    return label[probability]

For a single input sentence

input_text = "I had a wonderful day"
start_time = time.perf_counter()
result = infer(input_text)
end_time = time.perf_counter()
total_time = end_time - start_time
print("Label: ", result)
print("Total Time: ", "%.2f" % total_time, " seconds")
Total Time:  0.04  seconds

Read from a text file

start_time = time.perf_counter()
with open("../data/text/food_reviews.txt", "r") as f:
    input_text = f.readlines()
    for lines in input_text:
        print("User Input: ", lines)
        result = infer(lines)
        print("Label: ", result, "\n")
end_time = time.perf_counter()
total_time = end_time - start_time
print("Total Time: ", "%.2f" % total_time, " seconds")
User Input:  The food was horrible.


User Input:  We went because the restaurant had good reviews.

Total Time:  0.02  seconds