# ADAMS SoSe23 — Data

来源：https://www.kaggle.com/competitions/adams-sose23/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description

##**Data partitioning**
The data has been randomly partitioned into a training set and a test set. You should use the labeled training set for developing your models and validating their quality. Next, you should use the developed models to predict the property prices for the listings in the unlabeled test set.

To prevent overfitting, the test set is divided into a public and a private subset. The leaderboard score of your submission is computed on a private part of the test set. Once the competition is finished, you will see a private score of your submissions. The private score determines the winner of the competition. While the competition is running, you will be able to see the public score of your submissions. In this way, you can track your progress over time, when testing different models.

##**File descriptions**
The data consists of three .csv files containing different information:

- **train.csv**: training set with listing features, description and prices
- **test.csv**: test set with listing features and description
- **reviews.csv**: [optional for master students, mandatory for PhD students] reviews for listings in both training and test set

##**Data fields**
Files train.csv and test.csv have the same columns except for the price variable:

- listing_id: unique property identifier
- price: [target variable] property price per night in US dollars
- summary: [text] listing summary
- description: [text] listing description
- space: [text] space description
- neighborhood_overview: [text] neighborhood description

As indicated above, the text data is provided in several columns devoted to different topics. You will need to think whether to merge them into a single text or consider separately in order to improve your solution.

The remaining columns contain numeric/categorical features with the listing attributes such as location, host information, reviewer scores and more. You would need to perform some data preprocessing and think how to incorporate the tabular data into your modeling pipeline. The column containing the urls to the property images should be dropped.

##**Additional data**
The data on property reviews in reviews.csv is optional for master students. It can be used to further refine the quality of the solution, but master students can also choose to ignore it. By contrast, PhD students are required to include the review data in their submissions.

##**Reviews**
Data dictionary for reviews.csv:

- listing_id: unique property identifier
- review_id: unique review identifier
- reviewer_id: unique reviewer identifier
- comments: [text] review

Reviews can be matched with train.csv by listing_id. You would need to think how to deal with the following challenges:

- reviews may come in different languages
- listings have a different number of reviews
- aggregating signal from multiple reviews