Entity Sentiment Analysis Framework

Entity sentiment analysis aims to determine sentiment of different entities identified in a text. Our framework will be using the Google Cloud Natural Language API, which is a web API that uses NLP to process text and provide insight on key factors such as sentiment, salience, etc. To use the framework, you should create implementations of the DataPlugin and VisualizationPlugin interfaces (in the plugin directory) to properly format input texts and create visualizations of analysis results. Remember to edit the files in the resources folder to register your plugins.

Data Processing

This framework incorporates machine learning by aggregating the processed text files and calling the Google Cloud Natural Language API entity-sentiment analysis API on each. The visualizations present the ML analysis results.

Data Plugin

THe main method in the data plugin is convertToString. The job of this function is to convert the contents of a single given file into a String. The parameter of this method is of the type Path from the Java NIO package. The Path object can be easily converted to a File object, as necessary.

There are some sample Visualization Plugins in the plugin/data folder. PDFPlugin utilizes the Java PDFBox Library to parse a PDF file, and TxtPlugin utilizes the Java NIO Library to parse a TXT file.

Register a new data plugin by adding the path to the plugin to /backend/src/main/resources/META-INF/services/edu.cmu.cs214.hw6.framework.core.DataPlugin.

Visualization Plugin

The main method in the visualization plugin is visualizeData. The only parameter of visualizeData is a list of AnalyzeEntitySentimentResponse objects which are objects defined by the Google Cloud Natural Language API. The documentation for Entity, EntityMention, and Sentiment objects can be found online. Example code for retrieving AnalyzeEntitySentimentResponse information can be found here. In the visualizeData method, image files should be created in the plugin/charts folder in the backend, and the path to the visualized image should be returned as a String.

Note that the list of AnalyzeEntitySentimentResponse objects are ordered based on the order of texts in the directory.

There are some sample Visualization Plugins in the plugin/visualization folder. These plugins make use of 3rd party library JFreeChart to generate visualization images.

Register a new visualization plugin by adding the path to the plugin to /backend/src/main/resources/META-INF/services/edu.cmu.cs214.hw6.framework.core.VisualizationPlugin.

Set Up GCloud

Before running the program, this is a critical step necessary in order to make Google Cloud Natural Language API Calls.

TLDR Steps

1. Create a project in Google Cloud and enable the Cloud Natural Language API, which requires a billing account.

2. Set up credentials for ADC as described here.

Detailed Steps

1. Install gcloud: follow https://stackoverflow.com/a/52438247 until step 7
2. Get Google Cloud Platform API credentials in JSON: https://www.youtube.com/watch?v=rWcLDax-VmM
3. In your Google Cloud console (https://console.cloud.google.com/), navigate to your project and search for “Cloud Natural Language API”.
4. Click “Enable” and set up a free trial (must provide your billing information).
5. Set your environmental variable: use the path to the JSON file you downloaded from step 2 and run the following command in your terminal.

$ export GOOGLE_APPLICATION_CREDENTIALS=“<your_path”

It will look something like

$ export GOOGLE_APPLICATION_CREDENTIALS=“/Users/sunnyliang/Downloads/hw6-esa-framework-project-some-alphanumeric.json

Verify this worked by running the following

$ echo $GOOGLE_APPLICATION_CREDENTIALS

6. Now you are ready to make API calls

Running the Program

1. Set Up Backend Server

Either run the Java backend by using your IDE or by typing

mvn exec:exec

in the back-end folder. This will start the Java server at http://localhost:8080.

2. Set Up Frontend Server

In the front-end folder, run

npm install
npm start

This will start the front-end server at http://localhost:3000. You can update the front-end code as the server is running in the development mode (i.e., npm start). It will automatically recompile and reload.

Usage Examples

Since the required GCloud setup is lengthy, here are examples of using our framework.

CMU Reviews

Given a local directory of txt files (see at /docs/txts/), each containing a student-written review about CMU, the framework will generate a bar graph of the aggregate salience of the entities in the reviews.

Using the framework

The result

Political Articles

Given a local directory of 2 PDF files (see at /docs/pdfs/), one PDF containing a CNN article about Trump and one PDF containing a Fox News about Trump, the framework will generate a pie chart of the averaged sentiment magnitudes of the entities of the texts.

Using the framework

The result