Data Description

Data Use Update: The Competition Sponsor has updated the permitted use of this competition's dataset. You may access and use the Competition Data for any purpose, whether commercial or non-commercial, including for participating in the Competition and on Kaggle.com forums, and for academic research and education.

The dataset for this challenge contains a sample of users’ page views and clicks, as observed on multiple publisher sites in the United States between 14-June-2016 and 28-June-2016. Each viewed page or clicked recommendation is further accompanied by some semantic attributes of those documents. For full details, see data specifications below.

The dataset contains numerous sets of content recommendations served to a specific user in a specific context. Each context (i.e. a set of recommendations) is given a display_id. In each such set, the user has clicked on at least one recommendation. The identities of the clicked recommendations in the test set are not revealed. Your task is to rank the recommendations in each group by decreasing predicted likelihood of being clicked.

As a warning, this is a very large relational dataset. While most of the tables are small enough to fit in memory, the page views log (page_views.csv) is over 2 billion rows and 100GB uncompressed. We have also uploaded a sample version of this file with the first 10,000,000 rows. The MD5 checksum of page_views.csv.zip is 3742c116bab4030e0a7ea1c0be623bd9.

Data Fields

Each user in the dataset is represented by a unique id (uuid). A person can view a document (document_id), which is simply a web page with content (e.g.  a news article). On each document, a set of ads (ad_id) are displayed. Each ad belongs to a campaign (campaign_id) run by an advertiser (advertiser_id). You are also provided metadata about the document, such as which entities are mentioned, a taxonomy of categories, the topics mentioned, and the publisher.

File Descriptions

page_views.csv is a the log of users visiting documents. To save disk space, the timestamps in the entire dataset are relative to the first time in the dataset. If you wish to recover the actual epoch time of the visit, add 1465876799998 to the timestamp.

• uuid

• document_id

• timestamp (ms since 1970-01-01 - 1465876799998)

• platform (desktop = 1, mobile = 2, tablet =3)

• geo_location (country>state>DMA)

• traffic_source (internal = 1, search = 2, social = 3)

clicks_train.csv is the training set, showing which of a set of ads was clicked.

• display_id

• ad_id

• clicked (1 if clicked, 0 otherwise)

clicks_test.csv is the same as clicks_train.csv, except it does not have the clicked ad. This is the file you should use to predict. Each display_id has only one clicked ad. Note that test set contains display_ids from the entire dataset timeframe. Additionally, the public/private sampling for the competition is uniformly random, not based on time. These sampling choices were intentional, in spite of the possibility that participants can look ahead in time.

sample_submission.csv shows the correct submission format.

events.csv provides information on the display_id context. It covers both the train and test set.

• display_id

• uuid

• document_id

• timestamp

• platform

• geo_location

promoted_content.csv provides details on the ads.

• ad_id

• document_id

• campaign_id

• advertiser_id

documents_meta.csv provides details on the documents.

• document_id

• source_id (the part of the site on which the document is displayed, e.g. edition.cnn.com)

• publisher_id

• publish_time

documents_topics.csv, documents_entities.csv, and documents_categories.csv all provide information about the content in a document, as well as Outbrain's confidence in each respective relationship. For example, an entity_id can represent a person, organization, or location. The rows in documents_entities.csv give the confidence that the given entity was referred to in the document.