[HN Gopher] Facebook changes algorithm to boost original reporting ___________________________________________________________________ Facebook changes algorithm to boost original reporting Author : jbegley Score : 55 points Date : 2020-06-30 13:23 UTC (9 hours ago) (HTM) web link (www.axios.com) (TXT) w3m dump (www.axios.com) | tumetab1 wrote: | It seems these days opaque machine learning models means | transparency. | | I would prefer they would say that they're down grading their | preference for user engagement, and time on platform, for what | the trained model says it's original reporting. | r721 wrote: | >In conjunction with those changes, Facebook will also begin to | down-rank news in its algorithm that doesn't have bylines, or | present information about the company's editorial staff on the | publishers' website. | | Will they down-rank The Economist, I wonder? :) | mc32 wrote: | I imagine they can whitelist known trustworthy entities that | follow some modicum of rules. | mgraczyk wrote: | This is the actual press release. Shame on axios for not linking | to it. | | https://about.fb.com/news/2020/06/prioritizing-original-news... | djohnston wrote: | is this irony? | caiobegotti wrote: | This is a critical bit that won't address the problem worldwide: | | _For now, the tech giant is focusing on stories in English. It | hopes to expand to other languages in the future._ | chillacy wrote: | It's pretty typical for companies to start with English and | expand from there. Siri had a similar arc, now it supports | 20ish and continues to grow. | rmrk wrote: | "it will use artificial intelligence to analyze groups of | articles on a particular story topic and identify the ones most | often cited as the original source." | | and | | "Facebook will also begin to down-rank news in its algorithm that | doesn't have bylines, or present information about the company's | editorial staff on the publishers' website." | | Okay those are the new requirements for content mills and fake | propaganda outlets. How long before they adapt? | | Actually vetting reporters, reportage and news outlets is really | hard for a team of smart humans editors to do. Even the premiere | organizations like the NYT and Washington Post with their armies | of editors has failed at this from time to time. Algorithms are | not ready for this task yet. | chillacy wrote: | > Google said last year that it adjusted its algorithms and the | guidelines used by the people that rate its search results to | elevate original reporting. | | Have we seen these adaptations yet? It's been half a year since | Google announced their intentions too. | mgraczyk wrote: | The ranking algorithm does not do the vetting. The algorithm | just chooses how to weight the vetting outcome to rank the | content. | ksk wrote: | Fair point, but its likely very hard if not impossible to | create AI algorithms for automatic/guided content | curation/classification without deploying them in a real-world | use-case. | rmrk wrote: | For me this is the crux of the issue with The Platforms | giving rise to "fake news". | | We as a society have decided that rampant mis-information and | propaganda is only worth solving if we can automate it. If we | actually have to pay real people real money to fix it on an | ongoing basis, that's just too expensive. | | Sure there are problems having Humans doing this work too, | but they are still way ahead of AI in this problem space. | | How long do we wait for automated solutions while these | problems impose real costs to society? | luckylion wrote: | I have doubts that you can do it without heavy automation. | Sure, eventually some human can decide whether something is | "factual-ish" or not. But producing content is much easier | and can be automated, so the attackers can flood the | system. | | If you want humans involved, you end up with a gate keeper, | which essentially means "unless you're an accredited media | organization, your content is considered fake", because you | can't vet individual pieces. | rmrk wrote: | I agree with you. It'd take a huge company with tons of | resources at its disposal to do something like this if | it's possible at all. But if anyone could hire and train | the army necessary to do it, it'd be google or facebook. | (Apple already does it. Apple News is edited by real | apple-employed humans but its far smaller in scope). | | I think real solutions are gonna require us to break out | of our tech-focused approaches and find ways to get | Google, Facebook, Twitter to really start to care about | fixing this stuff. Unfortunately I think that means it'll | have to start costing them. | rhizome wrote: | > _" it will use artificial intelligence to analyze groups of | articles on a particular story topic and identify the ones most | often cited as the original source."_ | | Great idea, they should give it a snappy name, maybe something | that rhymes with "stage tank." Of course this does nothing WRT | organizations that tend not to cite earlier reportage when it | originates outside of the company. | | > _How long before they adapt?_ | | Why, that would require creating a staff of fake names, so in a | lot of cases it'll probably be completed sometime around close | of business today. Maybe the end of the week. | basch wrote: | Deduping blogspam and re-reporting of AP/Reuters, and using | that redundancy to uprank the original source, is something | Facebook should have been doing a decade ago. | | It should be more akin to https://techmeme.com (or hn for | that matter) where they editorially try and choose the first | or best source. If a better source becomes available they | swap. Facebook could benefit from this dynamicness, where a | story can bump and replace an existing post. | zitterbewegung wrote: | Well this is the next logical step for AdTech. Since all of the | newspapers and other media outlets have been disrupted by the | rest of the AdTech industry the next step would be to allow for | people to create content on the platform and then you keep your | moat. ___________________________________________________________________ (page generated 2020-06-30 23:01 UTC)