ShowCase.md

Problem Statement: In today’s world, with increase usage of social media, fake news has become a serious threat to many people. These kinds of false information can influence people’s views and can affect the informed choices of the country’s subjects. Fake news can reduce the impact of real news and hence it is important to detect such news. Affects economy of the nation Impact on democracy 60% of news in the year 2020 was fake 10.9% year-on-year increase of population using social media

Objectives: To make users aware by detecting fake news spread through social media. Model a web application, web extension and mobile app to detect fake news. Create a database through scraping and testing data taken from some trustworthy websites. Validate the input given by users by checking through our database, whether the information is from trustworthy sources or not. Indicate the specifics of what causes an information to be fake or real.

Impact And Potential: Users and Use cases (including defining Economic benefit, Time benefit and Social-benefit): User: Survey and Protection:

• Fake News Detection on Social-media: Social-media is one of the most easily accessible and low-cost platforms for people to consume news. As an increasing amount of our lives is spent interacting online through social media platforms, more and more people tend to seek out news from social media rather than professional news organizations. On the other side of this it enables wide spread of fake news with intentionally false information which can have major political, educational and health related outcomes. Hence using our fake news detection model helps provide educational and political benefits.
• Spread of anti-vaccination misinformation is one of the examples on how fake news can have serious negative effects particularly during these times of an on going pandemic. So, it becomes even more important to use fake news detection model today to help mitigate the negative effects of misinformation and thus helping in social benefits.
• Financial Impact: Let us suppose I like a particular blogger and looking at his posts I feel like buying a product that costs 50 dollars which might be ineffective and harmful for my health but because I believe that guy, I buy the product without any verification just because he said it is good. In a case like this instead of trusting it blind folded I can use the fake news detection model and verify if that product is worth it.
• Economic, Social and Time benefit: If we take an example of a company like sprite and if some rumor is being spread about them as they are mixing some kind of poisonous drug in it, then everyone will think whether to drink it or not and would prohibit it. This will ultimately lead to the market of sprite getting down. But if they use our fake news detection model it can help them decide whether they can drink it or not with the help of some real information. So, it has a social benefit and it also gives sprite company an economic benefit that it prevents their economic loss by assessing that the news about them is fake and it also saves user time because if they really want to check about the rumor on internet, they need so much time to get information by going and reading through different sources and further check if those sources are fake or not. So instead if they use the fake detection site, they can get the information about fakeness in a few seconds. So, our model gives social, economic and time benefits.
• Political Impact: Many still believe on the influence of fake news on twitter during 2016 US presidential elections and its potential impact on Donald Trump’s victory and believe that if there was a way to detect those misinformation maybe the outcome would have been different. So, we can also use our fake news detection model to detect misinformation leading to political benefits.

Approach: Developing a Minimum viable product

Minor Release: Fake detection website with minimal front end using some nlp and indexing techniques (version 0.1) Our fake news detection website take's inputs from user and extracts important keywords out of it and send it to Apache solr( a free and open source software used as a search engine) where it maintains an indexing schema(to make searching faster) and a database containing information of some trustworthy websites so that when an input comes it will be searched through the documents and the similar documents will be returned back as the results. Then , the model will calculate the cosine similarity between the actual input sentence and the documents from solr if the similarity is less than 0.1 it will say it as trustworthy otherwise we will use textual entailment methods to decide whether the most similar sentence is actually entailing the given input sentence or not and then we will give the result accordingly.

Team Details Leads :

• Vidit Ostwal
• Aniket Viramgama

Developers :

• Aashutosh Pandey
• Vidit Ostwal
• Advika S
• Koteswarudu Akula
• Pronoma Banerjee
• Aniket Viramgama
• Ayushi kaul
• Chitvan Agarwal
• Omkar Mahesh Garad
• Osama Zameer
• Sabbisetti Hemanth
• Vedant Vijay Dalimkar
• Vinay Verma
• Aryaman Jeendgar
• Aryan Gandhi
• Prakhar Gupta
• Akshat singhal
• Shambhavi Sumedha

Testers:

• Pronoma Banerjee
• Ayushi kaul
• Aryan Gandhi

Mentors :

• Deep Kumar
• Hari Sai

Clients :

• Deep Kumar

Communication :

• Project Group Meeting Medium : BBB
• Daily Standup Time : 9:30 AM - 9:45 AM
• Daily Dev-Sprints Time: 10 AM - 5 PM
• Mentor Meeting Time : Saturday 6:30 PM - 7:30 PM
• Client Meeting Time : Saturday 6:30 PM - 7:30 PM

Work Division : The job is distributed in a systematic manner across all team members. Generally, all of the members are divided into two teams, one for machine learning and the other for web development. In both teams of there are further sub-extensions of 3-4 that cater to the team's individual demands. Web development team:

• Aashutosh Pandey
• Chitvan Agarwal
• Sabbisetti Hemanth
• Osama Zameer
• Aniket Viramgama
• Aryan Gandhi
• Prakhar Guptha
• Akshat singhal
• Shambhavi Sumedha

Machine Learning Team:

• Vidit Ostwal
• Advika S
• Koteswarudu Akula
• Pronoma Banerjee
• Ayushi kaul
• Omkar Mahesh Garad
• Vedant Vijay Dalimkar
• Vinay Verma
• Aryaman Jeendgar

Contribution Tracking: Tasks are assigned to all the members, and they all contribute on a regular basis to those tasks. Every day, a regular meeting is held for project discussion, and practically all of the members attend. Most of us are new to this field and still need to research and learn to move forward.

Delivery Deadline: The project's projected delivery date is in the next 3-4 weeks, and all individuals are working on their tasks accordingly.

Next Steps: We will try to solve Three main problems in our model as given below Indexing : We will try to improve indexing schema at solr because the present indexing is not working well if we change few words in the input sentence. Cosine similarity and Textual Entailment: At present our model will give output as not fake if the similar document in database and the input sentence has cosine similarity less than 0.1 but it won’t work in all cases because if we input a large sentence that is already present in our database and if we flip the sentence meaning to it’s negative by introducing a ‘NOT’ in between then our model predicting it as not fake but it has to tell it is fake because the word ‘not’ can’t significantly change the value of cosine similarity of a large sentence so as a result cosine similarity will be less than 0.1 and the output will be given as not fake.So, we will try to change this behaviour by introducing textual entailment to each matched sentence so that the model can understand the difference between positive and negative sentence. Correlation between different sentences and understanding the meaning of sentence using nlp techniques is another feature we want to introduce so that our model can understand whether two sentences are giving the same meaning or not.