This commit is contained in:
sahil 2023-07-17 21:41:31 -05:00 committed by GitHub
commit dfc3b063a0
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23

View File

@ -1,60 +1,38 @@
# Earlybird Light Ranker
*Note: the light ranker is an old part of the stack which we are currently in the process of replacing.
The current model was last trained several years ago, and uses some very strange features.
We are working on training a new model, and eventually rebuilding this part of the stack entirely.*
*Note: The light ranker is an older part of the stack being replaced. The current model was trained years ago and uses odd features. A new model is being developed, and eventually, the entire stack will be rebuilt.*
The Earlybird light ranker is a logistic regression model which predicts the likelihood that the user will engage with a
tweet.
It is intended to be a simplified version of the heavy ranker which can run on a greater amount of tweets.
The Earlybird light ranker is a logistic regression model predicting user engagement likelihood with tweets. It's a simplified version of the heavy ranker, capable of handling more tweets.
There are currently 2 main light ranker models in use: one for ranking in network tweets (`recap_earlybird`), and
another for
out of network (UTEG) tweets (`rectweet_earlybird`). Both models are trained using the `train.py` script which is
included in this directory. They differ mainly in the set of features
used by the model.
The in network model uses
the `src/python/twitter/deepbird/projects/timelines/configs/recap/feature_config.py` file to define the
feature configuration, while the
out of network model uses `src/python/twitter/deepbird/projects/timelines/configs/rectweet_earlybird/feature_config.py`.
The `train.py` script is essentially a series of hooks provided to for Twitter's `twml` framework to execute,
which is included under `twml/`.
The Earlybird light ranker is a logistic regression model predicting user engagement likelihood with tweets. It's a simplified version of the heavy ranker, capable of handling more tweets. There are two main light ranker models: one for in-network tweets (`recap_earlybird`) and another for out-of-network (UTEG) tweets (`rectweet_earlybird`). Both models are trained using the `train.py` script, and they mainly differ in the features used. The in-network model uses `src/python/twitter/deepbird/projects/timelines/configs/recap/feature_config.py`, while the out-of-network model uses `src/python/twitter/deepbird/projects/timelines/configs/rectweet_earlybird/feature_config.py`.
The `train.py` script serves as a series of hooks for Twitter's `twml` framework, included under `twml/`.
### Features
The light ranker features pipeline is as follows:
![earlybird_features.png](earlybird_features.png)
Some of these components are explained below:
- Index Ingester: an indexing pipeline that handles the tweets as they are generated. This is the main input of
Earlybird, it produces Tweet Data (the basic information about the tweet, the text, the urls, media entities, facets,
etc) and Static Features (the features you can compute directly from a tweet right now, like whether it has URL, has
Cards, has quotes, etc); All information computed here are stored in index and flushed as each realtime index segments
become full. They are loaded back later from disk when Earlybird restarts. Note that the features may be computed in a
non-trivial way (like deciding the value of hasUrl), they could be computed and combined from some more "raw"
information in the tweet and from other services.
Signal Ingester: the ingester for Realtime Features, per-tweet features that can change after the tweet has been
indexed, mostly social engagements like retweetCount, favCount, replyCount, etc, along with some (future) spam signals
that's computed with later activities. These were collected and computed in a Heron topology by processing multiple
event streams and can be extended to support more features.
- User Table Features is another set of features per user. They are from User Table Updater, a different input that
processes a stream written by our user service. It's used to store sparse realtime user
information. These per-user features are propagated to the tweet being scored by
looking up the author of the tweet.
- Search Context Features are basically the information of current searcher, like their UI language, their own
produced/consumed language, and the current time (implied). They are combined with Tweet Data to compute some of the
features used in scoring.
Components explained below:
- Index Ingester: An indexing pipeline responsible for handling tweets as they are generated. This component serves as the primary input for Earlybird. It creates Tweet Data, which includes basic information about the tweet (text, URLs, media entities, facets, and more) and Static Features, which are features that can be computed directly from a tweet (such as whether it has a URL, cards, or quotes). All information computed by the Index Ingester is stored in an index and flushed as each realtime index segment becomes full. When Earlybird restarts, this information is loaded back from the disk. It's important to note that some features might be computed in non-trivial ways, such as determining the value of "hasUrl". These features could be computed and combined from raw information within the tweet and data from other services.
- Signal Ingester: Responsible for Realtime Features—per-tweet features that can change after indexing. These include
social engagements like retweet count, favorite count, and reply count, as well as future spam signals. They are
collected and computed in a Heron topology by processing multiple event streams and can be expanded to support more features.
- User Table Features: A separate set of per-user features sourced from the User Table Updater, which processes a stream
written by the user service. It stores sparse realtime user information, and these per-user features are propagated to
the tweet being scored by looking up the tweet's author.
- Search Context Features: Information about the current searcher, such as their UI language, their own produced/consumed language,
and the current time. These features are combined with Tweet Data to compute some of the features used in scoring.
The scoring function in Earlybird uses both static and realtime features. Examples of static features used are:
- Whether the tweet is a retweet
- Whether the tweet contains a link
- Whether this tweet has any trend words at ingestion time
- Whether the tweet has trend words at ingestion time
- Whether the tweet is a reply
- A score for the static quality of the text, computed in TweetTextScorer.java in the Ingester. Based on the factors
such as offensiveness, content entropy, "shout" score, length, and readability.
- A score for the static text quality, computed based on factors like offensiveness, content entropy, "shout" score, length, and readability.
- tweepcred, see top-level README.md
Examples of realtime features used are: