Rambler enabled filter "adult" sites
June 1, 2010 - on Children's Day - Rambler launched a filter of sites with erotic content
')
Search filter
“Adult” content is now excluded from the search results of the “Rambler” by default, unless the user intentionally searches for it. At the same time, we leave to users the opportunity to find erotic content if they are interested in it. Filtering occurs when the issuance of "adult" sites is inappropriate.
In the future, we plan to launch “Family Search”, which will protect children from undesirable appearances in the issuance of not only “adult” content, but also some other topics (gambling, drugs, etc.).
"Family Search" is under development, but you can try the product today on the Advanced Search page:
With the “Family search” enabled, “Rambler” not only filters porn, but does not search for sites with erotic content at all, even if the request contains an intention to find such sites.
How do we do it
We define the characteristics of sites containing adult content. Moreover, these signs depend on the type of "adult content". For example, for erotic sites, the signs will be different from those where hard porn. As attributes, specific lexical content is taken into account in various parts of the document, link texts, the presence of pictures, and so on. Weights featured in the classifier are “distributed” automatically using machine learning methods.
Depending on the chosen degree of filtration and on the final assessment of the degree of “maturity” of the resource, various steps are possible. Some resources are generally excluded from the search results, some are reduced in positions. At the same time, however, such resources are not excluded from the search index (unless they contain signs of spam).
Approximately the same thing happens with requests: whether the request expresses an intention to receive “adult” content, is determined automatically by different characteristics of the request: lexical, behavioral (for example, analysis of user clicks on search results).
We have just begun work on filtering inappropriate content and assume that the algorithms may be wrong. We will be grateful if you inform us about possible errors and inaccuracies. We will definitely take into account all comments to improve the operation of filters.
Marina Anisimova
Press Secretary "Rambler"
')
Search filter
“Adult” content is now excluded from the search results of the “Rambler” by default, unless the user intentionally searches for it. At the same time, we leave to users the opportunity to find erotic content if they are interested in it. Filtering occurs when the issuance of "adult" sites is inappropriate.
In the future, we plan to launch “Family Search”, which will protect children from undesirable appearances in the issuance of not only “adult” content, but also some other topics (gambling, drugs, etc.).
"Family Search" is under development, but you can try the product today on the Advanced Search page:
With the “Family search” enabled, “Rambler” not only filters porn, but does not search for sites with erotic content at all, even if the request contains an intention to find such sites.
How do we do it
We define the characteristics of sites containing adult content. Moreover, these signs depend on the type of "adult content". For example, for erotic sites, the signs will be different from those where hard porn. As attributes, specific lexical content is taken into account in various parts of the document, link texts, the presence of pictures, and so on. Weights featured in the classifier are “distributed” automatically using machine learning methods.
Depending on the chosen degree of filtration and on the final assessment of the degree of “maturity” of the resource, various steps are possible. Some resources are generally excluded from the search results, some are reduced in positions. At the same time, however, such resources are not excluded from the search index (unless they contain signs of spam).
Approximately the same thing happens with requests: whether the request expresses an intention to receive “adult” content, is determined automatically by different characteristics of the request: lexical, behavioral (for example, analysis of user clicks on search results).
We have just begun work on filtering inappropriate content and assume that the algorithms may be wrong. We will be grateful if you inform us about possible errors and inaccuracies. We will definitely take into account all comments to improve the operation of filters.
Marina Anisimova
Press Secretary "Rambler"
Source: https://habr.com/ru/post/95129/
All Articles