Antidote Recommendation Tool

This repository contains code for an Antidote Recommendation Tool that utilizes vector-based and topic modeling approaches to recommend antidotes for fake claims. In addition the tool leverages the bipartite graph representation and topic modeling techniques to identify relevant antidotes from a corpus of claims.

Contents

• AntidoteRecs/gfc.py: Contains functions for retrieving claims from the Google Fact Check API.
• AntidoteRecs/vectorizers.py: Contains the Vectorizer class for transforming claims into embeddings using pre-trained models.
• AntidoteRecs/antidote.py: Contains functions for retrieving antidotes based on vector-based and topic modeling approaches.
• AntidoteRecs/claim_topic.py: Contains functions for extracting topics from a corpus of claims.
• AntidoteRecs/graphs.py: Contains functions for building bipartite graphs based on user-submission interactions.
• AntidoteRecs/preemptive.py: Contains functions for predicting the next topic of interest for a given user.

Usage

To use the Antidote Recommendation Tool, follow these steps:

Vector-Based Approach

1. Import the necessary functions and classes from the respective files.
2. Retrieve claims from the Google Fact Check API (optional) using the get_claims function from AntidoteRecs.gfc or load the instance of claims using the load_data function from AntidoteRecs.gfc or any other suitable method.
3. Initialize a vectorizer using the Vectorizer class from AntidoteRecs.vectorizers. Choose a pre-trained model appropriate for your task.
4. Utilize the get_antidotes function from AntidoteRecs.antidote to retrieve antidotes based on the vector-based approach. Pass the fake claim, corpus, vectorizer, and the desired number of top antidotes to retrieve.

Topic Modeling Approach

1. Import the necessary functions and classes from the respective files.
2. Load the corpus of claims using the load_data function from AntidoteRecs.gfc or any other suitable method.
3. Initialize a vectorizer using the Vectorizer class from AntidoteRecs.vectorizers. Choose a pre-trained model appropriate for your task.
4. Perform topic modeling on the corpus using the extract_topics function from AntidoteRecs.claim_topic. Specify the number of topics desired and provide the corpus and vectorizer.
5. Utilize the get_antidotes function from AntidoteRecs.antidote to retrieve antidotes based on the topic modeling approach. Pass the fake claim, corpus, vectorizer, number of top antidotes, and the topic of the fake claim obtained from topic modeling.

Preemptive Recommendations

1. Import the necessary functions and classes from the respective files.
2. Build a bipartite graph representation of user-submission interactions using the extract_graph function from AntidoteRecs.graphs.
3. Create a user-to-topic bipartite graph using the user_to_topic_graph function from AntidoteRecs.graphs, providing the user-submission graph, corpus, topic modeling results, and vectorizer.
4. Predict the next topic of interest for a given user using the predict_next_topic function from AntidoteRecs.preemptive.
5. Retrieve the claims and their corresponding reviews that belong to the predicted next topic using the topic information.