{ "id": 363962, "title": "Recall@20 Explained", "url": "/competitions/otto-recommender-system/discussion/363962", "author_name": "moth", "post_date": "2022-11-03T20:57:23.596000", "votes": 21, "comment_count": 0, "views": 0, "content": "

Recall definitions

\n

Recall is a common evaluation metric used in Information Retrieval. Many different definitions of recall can be found in the internet such as:
\n$$ Recall = \\frac{True Positives}{True Positives + False Negatives} $$

\n\n

Recall @ 20

\n\n

To illustrate with an example let's look at Jaime Arguello's example:

\n

\"\"

\n

It's simple to understand:

\n
    \n
  1. Set a K threshold (in our case 20) of items that will be retrieved.
    2. \n
  2. How many of the retrieved items are relevant. Relevant is defined by the ground truth.
    3. \n
", "messages": [ { "id": 2016221, "postDate": "2022-11-03T20:57:23.597Z", "content": "

Recall definitions

\n

Recall is a common evaluation metric used in Information Retrieval. Many different definitions of recall can be found in the internet such as:
\n$$ Recall = \\frac{True Positives}{True Positives + False Negatives} $$

\n\n

Recall @ 20

\n\n

To illustrate with an example let's look at Jaime Arguello's example:

\n

\"\"

\n

It's simple to understand:

\n
    \n
  1. Set a K threshold (in our case 20) of items that will be retrieved.
    2. \n
  2. How many of the retrieved items are relevant. Relevant is defined by the ground truth.
    3. \n
", "rawMarkdown": "# Recall definitions\n\nRecall is a common evaluation metric used in Information Retrieval. Many different definitions of recall can be found in the internet such as:\n$$ Recall = \\frac{True Positives}{True Positives + False Negatives} $$\n\n- **Recall:** The proportion of relevant items that are retrieved.\n- **Recall:** The probability that a relevant item will be retrieved.\n\n# Recall @ 20\n\n- **R@K:** ratio of relevant items that are in the top-K recommendations.\n\nTo illustrate with an example let's look at [Jaime Arguello's](https://ils.unc.edu/courses/2013_spring/inls509_001/lectures/10-EvaluationMetrics.pdf) example:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3197853%2F51ef579456f375f64add00911a892c32%2FScreen%20Shot%202022-11-03%20at%2017.52.28.png?generation=1667508765486628&alt=media)\n\nIt's simple to understand:\n1. Set a K threshold (in our case 20) of items that will be retrieved.\n2. How many of the retrieved items are relevant. **Relevant is defined by the ground truth.**", "votes": 21 } ], "comments": [], "raw_markdown_by_id": { "2016221": "# Recall definitions\n\nRecall is a common evaluation metric used in Information Retrieval. Many different definitions of recall can be found in the internet such as:\n$$ Recall = \\frac{True Positives}{True Positives + False Negatives} $$\n\n- **Recall:** The proportion of relevant items that are retrieved.\n- **Recall:** The probability that a relevant item will be retrieved.\n\n# Recall @ 20\n\n- **R@K:** ratio of relevant items that are in the top-K recommendations.\n\nTo illustrate with an example let's look at [Jaime Arguello's](https://ils.unc.edu/courses/2013_spring/inls509_001/lectures/10-EvaluationMetrics.pdf) example:\n\n![](https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F3197853%2F51ef579456f375f64add00911a892c32%2FScreen%20Shot%202022-11-03%20at%2017.52.28.png?generation=1667508765486628&alt=media)\n\nIt's simple to understand:\n1. Set a K threshold (in our case 20) of items that will be retrieved.\n2. How many of the retrieved items are relevant. **Relevant is defined by the ground truth.**" } }