### Welcome to the Summer Challenge Whether you are entering the field of AI and want to get familiarised with language models or simply test your skills and have fun, this challenge is for you. You're about to encounter a unique task even those who've dabbled in ml will find intriguing In this challenge, you will create a model to detect whether the same author has written two texts. The data we provide you is collected from articles from our [blog](https://tantusdata.com/insights/) and two publicly available datasets of [news](https://archive.ics.uci.edu/dataset/217/reuter+50+50) and [blog posts](https://www.kaggle.com/datasets/rtatman/blog-authorship-corpus). The total data set is small, enabling you to iterate quickly by only using resources available on Kaggle or your computer. But also to test how you can manage with limited data, which is a situation that you will encounter in many cases. The key links and resources for you to keep track of: - [data](https://www.kaggle.com/competitions/authorship-comparison/data) - here, you will find the relevant files and a description of available data. - [code](https://www.kaggle.com/competitions/authorship-comparison/code) - here, you can create and share notebooks with code. If you are new to Kaggle, [here](https://www.kaggle.com/docs/notebooks) is notebooks documentation. Notebooks created through this page will have competition data automatically attached. - [discussion](https://www.kaggle.com/competitions/authorship-comparison/discussion) - here is a place for discussion. We have created [a thread](https://www.kaggle.com/competitions/authorship-comparison/discussion/425956) to ask questions. - [leaderboard](https://www.kaggle.com/competitions/authorship-comparison/leaderboard) - here, you can see the ongoing results. Note that the public leaderboard is calculated using ~30% of test data, and the remaining 70% will be visible after the competition ends - [evaluation](https://www.kaggle.com/competitions/authorship-comparison/overview/evaluation) - here is the information about the evaluation metric and submission file format. - [submissions](https://www.kaggle.com/competitions/authorship-comparison/submissions) - here you can see your submissions (all attempts) and choose up to three for the final scoring that will take place after the deadline - [rules](https://www.kaggle.com/competitions/authorship-comparison/rules) - here you can see all the competition rules. - [team](https://www.kaggle.com/competitions/authorship-comparison/team) - here you can add a link to your solution—more on that is explained in your task section. ### Your task Participants are challenged to create a model to determine if two texts come from the same author. We provide a training dataset (linked above), but you can also use an external dataset. But please note that the goal of this competition is not to train a huge model on all data available on the internet but to come up with a smart solution which can take advantage of even a small dataset. That is why you should also write a description and make your notebook public after the submission deadline so we can understand your approach and check whether it complies with the rules (to check them out, use the link above). [Here](https://www.kaggle.com/discussions/product-feedback/373153), you can find a description about creating a write-up attached to your best submission, which will be visible after the competition ends. You don't need to access external software, and the code can be run using Kaggle Notebooks. Now, a bit about the dataset: It was created from three sources [Blog Authorship Corpus](https://www.kaggle.com/datasets/rtatman/blog-authorship-corpus), [Reuter 50 50](https://archive.ics.uci.edu/dataset/217/reuter+50+50), and articles from our [blog](bloghttps://tantusdata.com/insights/). The training data consists of content text and author. For the test set, participants are provided with texts and pairs of text IDs for which they have to predict whether it comes from the same author. To access the data and for detailed files description head to the [data page](https://www.kaggle.com/competitions/authorship-comparison/data). > ### 💡 Tips and advice > To make things easier for those looking for more guidance, we have created [an example of a solution](https://www.kaggle.com/code/tantusdata/example-solution/notebook). You don't need to follow it, and it is only an idea to get you started. Those looking to get more familiar with ML before starting this task are welcome to use the following resources/we recommend: > - [this blog post](https://jalammar.github.io/illustrated-transformer/) is an excellent description of what the architecture looks like > - [NLP course on Huggingface](https://huggingface.co/learn/nlp-course/chapter1/1?fw=pt), which goes through all key aspects of training and using language models. You can jump ahead and start if you're not interested in extra tips or learning resources. But there is no need to hurry, and we do not evaluate by the solution date, but only the score and description. We will start the evaluation after the deadline. What we are looking for, aka how to score high upon submission: obviously, you want your score to be good, but remember that it should be coupled with a description of a scaled and optimal solution that best fulfils the task without needlessly overusing resources. It is part of the challenge to figure out what is too complicated and what is just right. The task should be solved independently; that is why we disabled teaming. To see all the rules, head to the tab linked above. ## Keep in mind There is an extra side contest that will run during this competition, so make sure to sign up to our [LinkedIn](https://www.linkedin.com/company/tantusdata/mycompany/) to be notified when it starts and how to win. If you have any questions, use the enquiries thread in the discussions. You are also encouraged to respond to your fellow challenge participants. If you want to share tips, you are welcome to. The deadline to submit is the 18th of August, 2023, 23:59 CEST. No late entries will be considered, so plan ahead! Well, that's the basics covered, make sure to have a look through the listed tabs and good luck.