Facebook is changing to the way it runs its Trending Topics feed, following an internal investigation.

The social network has announced more training for staff and that the feed will no longer rely on a list of news organizations, including the Washington Post, the BBC and Buzzfeed News, to validate subjects.

The feed, which lists popular headlines along with a brief description, has been accused of political bias.

However, Facebook’s report found no evidence of this.

The investigation analyzed 3,000 reviewer decisions following allegations that conservative issues were being suppressed, Facebook general counsel Colin Stretch said.

Facebook was accused by anonymous former employees of tampering with its Trending Topics feature, promoting “progressive” views and websites over content presenting views from the American right.

Current and former staff were also interviewed by the company.

The findings were revealed in a 12-page letter, addressed to Senator John Thune but also published online, in response to Senator Thune’s questions about the workings of Trending.

The Trending Topics feed currently works as a mixture of AI and human input, with potential subjects being suggested via algorithm and then reviewed by staff.

They are a mixture of popular subjects discussed on the social network and sourced from 1,000 media organizations. There was also a list of 10 organizations used to determine importance.

However, “as much as half” of the topics suggested algorithmically are rejected “because they do not make sense at the time or are duplicative”, Colin Stretch said.

So-called “stale topics” – events still popular in discussions after two days but with no new developments – and “junk hashtags” – popular topics not related to actual events – are also sidelined, he added.

Colin Stretch added that topics with sources in foreign languages may also not be included on the grounds that the team may be unable to identify them.

The report did find that historically some topics that were discussed over a long period of time did not show up algorithmically.

For example, hashtags relating to the Black Lives Matter campaign failed to appear in December 2014 and were not manually inserted by the Trending Topics review team either.

However, the topic “Ferguson”, which related to the police shooting of Michael Brown in Missouri, was added to compensate for this, wrote Colin Stretch.