Solutions for AI-Hackathon-2023 Final Contest Held by Irancell.

My teammate was Sina Asghari and our team ranked No.1 in the contest score and No.5 overall.
Due to the Contest Code of Conduct, we cannot share datasets publicly, but I will try to explain the schema. ( Sorry in advance ).
I hope you Enjoy :).
The contest contained 2 questions that were judged by an automated system, here I will explain these two.

Problem No.1:

After unzipping the dataset, you will see folders whose names represent a date.
Each of these folders contains folders with the names of news sources that were observed on that date and their news was collected.
Inside each of the folders related to news sources, the HTML files collected from that news source are placed.

data
├── 2023-08-05
│   ├── 7sobh
│   │   ├── 0.html
│   │   └── 1.html
│   └── farsnews
│       ├── 0.html
│       └── 1.html
└── 2023-08-04
    ├── varzesh3
    │   ├── 0.html
    │   └── 1.html
    └── snn
        ├── 0.html
        └── 1.html

Queries :

• How many unique news sources are there in the dataset?
• Which news source has the most HTML files in the entire dataset?
• Consider all HTML files related to varzesh3. Get the number of repetitions of the words کشتی, والیبال, and فوتبال in all p​​ tags, respectively.
• Find the most repeated word in h2 tag among all news collected on 2023-08-01. (You could use the stopwords array placed in the initial file)

Problem No.2:

In this question, you have to build a model that predicts the topic (label) of the news.
Each row in the data set corresponds to a news item and the topic of each news item is placed in the tags column.
This column has different topics, but your model prediction for the test data set should be one of the following topics:

• social (Should be Labeled as 0 )
• Economic (1)
• Iran_provinces (2)
• International (3)
• political (4)
• Scientific_cultural_sports (5)

As a result, the target column (tags) in the initial dataset may require changes. Note that at this stage of the contest, you are allowed to tag news with a maximum of one main category.
F1 Score is used to evaluate your model.
Our model accuracy was about 85%